Not because I think that’s a bad idea, but because I’m sufficiently familiar with the messaging surrounding the mid-level practitioner that I don’t trust it when it’s being sold that way.

For years, the mid-level practitioner has been pitched differently to different audiences. It is sometimes an alternative to veterinary school for those who, for whatever reason, have concluded that a four-year terminal degree isn’t for them. And sometimes it’s raising the ceiling for veterinary technicians.

This work has been compiled by yours truly but heavily formatted with LLM-use. Look, I spay cats for a living. Mastery of website, Word, and Google Docs formatting eludes me still. Substack doesn’t natively support tables, so I had Claude turn the table into a series of three PNG files. I can tie a slipknot in a matchbox with my eyes closed in less time than it takes me to read this sentence but tables and spacing? Not my thing.1

This is a document built from primary sources like Dr. Wayne Jensen’s presentation to the Colorado legislature, the CACVT statement on the mid-level practitioner, and various other publicly available records.

Four lanes here. One for the venue. One for what the technicians were told, one for what the prospective students, legislators, and profession were told, and the replies, meaning the survey results, position statements, and votes.

“So what?” some may ask.

The credential that makes someone a licensed, registered, or certified veterinary technician is, in most of the country, an associate degree. It’s two years, 60-70 credits, heavy on clinical skills, light on the general-science sequence that a bachelor’s requires.

The AVMA accredits 220 veterinary technology programs, and only 28 of them offer a four-year degree. One hundred and ninety-two others offer the associate track and nothing further.

Colorado State University’s MS-VCC requires a bachelor’s degree to apply. It also required, in the version Dr. Jensen presented to the legislature in 2023, 35-37 credits of prerequisites — including algebra, statistics, genetics, chemistry, gross anatomy, physiology, physics, and microbiology. Some, but not all, of the prerequisite pathway is found in a veterinary technician or associate’s degree curricula. And often in an applied form.

So when the VPA is described as a path for technicians, it must be stated that said path runs through a bachelor’s degree that most technicians did not get and a year (or so) of coursework most of the programs did not teach.

Hence: the hedge.

A real raise in the ceiling for veterinary technicians is a premise worth supporting, and to my recollection it was a topic gaining meaningful traction before AI news, noise, and hype drowned out pretty much everything else.

The CACVT wrote down what a real raising of the ceiling would look like for veterinary technicians in January 2023. And then the same people who told technicians that the VPA was their next step also told prospective students it was an alternative to veterinary school, and told the funders that the building was for the DVM program, and then built a master’s degree program that excludes most credentialed veterinary technicians.

Then they wrote the technician pathway into the statute as “may.” Then opened applications with an “or” in the sentence.

The statements weren’t exactly false, they were just true of different people.

So when somebody asks whether I’d support the VPA if it were meant for technicians, akin to the VTS programs mentioned by the CACVT, my answer is yes. But I’d quickly and emphatically point out that the VPA, as it exists today on CSU’s websites, is not meant for the majority of veterinary technicians. And I’d point to the Colorado Association for Certified Veterinary Technicians, who, in their own words, are still waiting on “one of many critical promises made during the Proposition 129 campaign.

And I’d make a point that Aesop’s satyr made long before I did: that the man who talks for sides is not to be trusted by either.

And a recent series of decisions from Zoetis Animal Health absolutely floored me. Not one bad actor, but a series of unbelievably bad decisions. How could so many smart people make so many bad calls? One right after another?

On July 30th, 2026, Columbia Veterinary Hospital, a four-doctor practice in The Dalles, Oregon, filed a complaint against Zoetis in Wasco County Circuit Court (Case No. 26CV40283). The complaint makes two claims: fraud in the inducement and violation of Oregon’s Unlawful Trade Practices Act. The prayer (the ask) is $50,000; a sum so modest as to suggest that the money isn’t the point.1

You probably saw approximately the same headlines I did. “Oregon veterinary hospital sues tech company, claims AI misdiagnosis led to dog’s death” and “The Algorithm in the Exam Room: Who’s To Blame When AI Misdiagnosis Turns Deadly?”

I’m all about AI and vet med, but none of the articles actually described the complaint in detail that I felt was sufficient. David Kincaid’s was pretty good, but some of the articles admit that they didn’t read the complaint or even the marketing materials cited!

But other articles, including one by Dr. Eric Fish, incorrectly state that no studies exist and that the complaint and research is not public record. I mean, c’mon! If you’re going to spill the tea, then spill the tea!2 It is public record and five studies exist.

Keep in mind, as you read my coverage of the complaint, that all of it rises to the level of allegation. This is one side’s story, told by lawyers, and Zoetis’ folks haven’t answered yet.

Evidence and Witnesses

In August 2024, CVH signed up for Zoetis’ VETSCAN IMAGYST.3 VETSCAN IMAGYST is an in-hospital AI diagnostic platform that Zoetis markets as “the world’s most capable veterinary AI analyzer” and promises “expert-level results in minutes” backed by “rigorous validation studies showing that VETSCAN IMAGYST’s AI-derived diagnostic results are comparable to results from expert… clinical pathologists.” The complaint quotes all of these claims at Zoetis, and the quotes are real. You can find them on the Zoetis website.

In June 2025, Zoetis launched a new Imagyst4 capability called AI Masses. You point a scanner at a cytology slide and the algorithm screens it. Its advertised findings include plasma cell tumors, histiocytomas, and mast cell tumors, among other things.5

On October 22, 2025, an eleven-year-old Belgian Tervuren presented to CVH with a rapidly growing mass on its neck. The hospital6 did a fine-needle aspirate, stained the slides, and scanned it into the Imagyst under the “Skin/subcutaneous Mass Clinical Profiles” category.

The report came back the next day. According to the complaint, the report’s microscopic findings included two lines:

• “Aggregate Mast Cell — Present”

• “Spindle Cells — Present”

And then the report’s conclusion:

• “Suggestive of inflammatory lesion” with “Indeterminate results — The findings do not match any established diagnostic categories.”

My reading of the complaint is that the machine found the neoplastic cells and, for whatever reason, talked itself out of calling it cancer. For those of you who didn’t have to take histology as many times as I did, the presence of mast cells and/or spindle cells in a rapidly growing mass is, in general, not good news. For purposes of interpretation, it means cancer. The doctors at CVH, I have to assume, would’ve had the opportunity to see both the “Findings” and the “Conclusion.” These things are generally presented distinctly but both included in a report.

On October 30, CVH removed the mass as an inflammatory lesion, which is to say that it was removed with narrower margins than would’ve been indicated for a mast cell, spindle cell, or other sorts of tumors.

During surgery, it appeared “too vascular” for the “inflammatory lesion” conclusion. A fast-growing mass, as you might reason, needs to be more vascular to feed its growth than would be expected of a lesion that is simply inflammatory.7 CVH sent the mass to Oregon State’s diagnostic lab to be read by, presumably, an actual human veterinary pathologist and, on November 6, the OSU report came back as a high-grade mast cell tumor with dirty margins on every plane. This is an aggressive cancerous mass, and “dirty margins” is shorthand for “didn’t get all of it.”

In mast cell tumors, the canonical recommendation is two centimeters laterally and one fascial plane deep. But that’s for lower graded tumors, which this was not. Regardless, dirty margins on the original surgery meant the recommendation was for a second, more aggressive, surgery.

On November 17th, 2025, CVH performed that surgery and the patient died of complications following that procedure.

Sad story with a bad outcome. And it could’ve stopped there.

(It did not stop there.)

The morning of the second surgery, CVH allegedly tried to reprint the original Imagyst report and couldn’t access it. Zoetis tech supported routed them to a different portal. The version that came out of that system had replaced two lines from the original report. What had once read “Aggregate Mast Cell — Present” and “Spindle Cells — Present” had both been replaced with “Unknown.” Notably, this came just a few days after CVH had sent the OSU report to Zoetis.

Then, per the complaint, there was a phone call where Zoetis apologized, admitted fault, and said other Imagyst users had hit the same problem.

Then CVH got locked out of the platform until it agreed to a new liability waiver, one reading “Last updated 10/22/2025.” The same day the Tervuren originally presented.

When CVH asked to see the “rigorous validation studies,” even offering an NDA, Zoetis allegedly refused, citing trade secrets.

The last detail that’s especially relevant for a veterinary audience? There are five peer-reviewed validation studies of Imagyst published in the literature and all five are about fecal parasites.8 At the time of this writing, I have not found a peer-reviewed study validating Imagyst’s AI Masses.

Duties of a Law-Abiding Clinician

Most of my readership is, predictably, of the veterinary inclination. And so I’m fairly sure that those folks will have clocked it already, but I’ll lay it out for everybody else: this case is barely about the patient, and it’s barely about AI.

The complaint spends two paragraphs on the dog’s life and fifteen pages on its paperwork. And I can’t help but linger on that fact. “Zoetis” is derived from the Greek word zoetic, which “of or pertaining to life.” And not “life” with the story attached to it, that would be bios, instead the company named itself after the bare fact of being alive. All the more poignant because the allegation here isn’t that Zoetis mishandled the living thing. The allegation is, in part, that once the living thing was gone, the company went back and edited the record of it. Zoetis didn’t revise the fact of life, the zoē, but the story of the life, the bios.

I’m a practicing clinician. I read histopathology reports and performed surgeries this morning. If a cytology report reads “aggregate mast cells — present” and/or “spindle cells — present” on a fast-growing mass in an eleven-year-old anything, that is not getting treated as an “inflammatory lesion” or “indeterminate” mass no matter how much the Conclusion hedges and regardless of whether the Conclusion’s author is a famed pathologist or a newly-minted algorithm.

There was, according to the complaint, a week between the first report and the surgery. A week in which a paid expert review of the same digital slide was available from Zoetis within hours (allegedly), or the slide could’ve gone to any reference lab on earth. Heck, I could’ve walked from The Dalles to Corvallis in a week.9

Then there’s the hospital, and the owner, that opted to revise the surgery rather than refer the case elsewhere.

And the patient’s passing? It followed complications from a second, more aggressive, surgery, on a geriatric patient, with a high-grade mast cell tumor. That carries, on a good day, a guarded prognosis.

The headlines that draw the throughline of causation from a fuzzy AI output to that outcome are ignoring a lot of human input along the way.

And the complaint is pretty clear about that. The complaint doesn’t make it about the patient or the humans or even the tumor.

The complaint makes it about what Zoetis said in its advertising materials and what it did when they got caught being wrong. And that’s a more interesting story.

Merely Thoughtless

For about the dozenth time in one of my essays: I’m not a lawyer. I don’t know enough to tell you, to even opine on, whether or not CVH has a winning case.

But I can count. And I can sort of read a stock chart. And I can tell you that Zoetis has shed something like $36 billion in market value over the past year for reasons that having nothing to do with lawsuit. The case did not move that number in a meaningful way.

But it did land at a time when it wouldn’t be great if the story went everywhere. And the story went everywhere. The local Oregon press, national syndication, the pet-media circuit, and a permanent entry in the AI Incident Database as Incident 1631.

I know little about stocks that aren’t meant for cattle, but I’d be willing to wager that “international news story with a dead dog about your flagship AI product missing cancer” costs more than $50,000. The headlines will cost Zoetis clients because every practice owner evaluating the product will Google “Imagyst” and find those stories. The headlines don’t even have to be turned purple to be damaging.

And now we’ve (finally) reached the baffling bit to me.

I get the surgery. I get that surgeries go wrong. I get that patients have complications. I get that FNAs are far from perfect.

What I struggle to get is how this happens.

Cover-ups in movies have a villain, they lead an evil corporation hellbent on world domination, and their nefarious schemes can only be thwarted by the release of this crucial piece of information.

This isn’t a movie and Zoetis isn’t a villainous company. CEO Kristin Peck has a podcast, not a secret lair.10 The company makes medications that are crucial to the practice of medicine for me and tens of thousands of other veterinarians treating hundreds of millions of patients.

And it’s with no small sense of irony that I note the title of Ms. Peck’s podcast as “On Purpose,” because I don’t think this cover-up was. I’m not even sure it was a cover-up.

I think this cover-up, such as it is, was distributed. I think it was a cascade of failures that piled on top of the first thing that went wrong. I think an engineer gets an annoyed call from a sales rep about a pissed off customer. The engineer does their job, opens a new portal, and closes a ticket. The updated portal publishes a different version of the report. A legal team responds the way legal teams respond, by pushing an updated waiver to the whole system, one stamped with a date that happens to match the day the Tervuren walked into CVH. Whether that's coincidence, some other client's bad day, or a lawyer picking a date that predates the problem can’t be known from the complaint. Then a well-meaning support rep gets on the phone with a frustrated doctor and apologizes like a human being, saying something meant to be empathetic and comforting, and in doing so creates the most damaging paragraph in the 17 pages of the complaint.

Every act is locally defensible. Stack them and it’s a conspiracy.

Lots of small but bafflingly bad decisions that amount to a pretty damning review of Zoetis’ internal processes. I still cannot get my head around how nobody in the cascade of failures ever said, “hey, stop trying to ‘fix’ this and take responsibility.”

In failing to ever own the problem, Zoetis finds itself paying Arendt on it.

“It is not offering an alternative. It is offering a ceiling.”

“The VPA is not a threat to indebted new graduates; it is part of the answer.”

Well, shoot. If it’s not one thing, it’s another.

If you’ve spent as much time reading Claudese as I have,1 you’ll have caught the more than a dozen repetitions of the construction, “it’s not X, it’s Y” in the August 12, 2026 statement of response from the Coalition for the Veterinary Professional Associates to Mission Pet Health’s August 6, 2026 statement on the VPA. The above quotations are from the CVPA document.

For those joining the story in medias res, Mission Pet Health is the >900-hospital network formed when Southern Veterinary Partners and Mission Veterinary Partners merged. Earlier this month, MPH published a position statement opposing the veterinary professional associate, the mid-level practitioner role Colorado voters created with Proposition 129 and the legislature codified into HB25-1285.

Mission’s argument, briefly: the role lacks the training the work demands, it piles risk and oversight burden onto supervising veterinarians, and the workforce math on which it was predicated now relies on force rather than math to work.

Six days and five pages later, the CVPA, an advocacy group led by Dr. Robert Murtaugh and Mr. Steven J. May, published a rebuttal, inviting MPH to “move from opposition to participation.”

I don’t have a degree in English or Literature, but I read, what seems to me, a lot. So take what follows as observation rather than accusation

Mission Pet Health’s statement reads like it was written by committee. There are inconsistencies in subject-verb agreement, there are grammatical and syntactical mistakes, and the sentence describing the scope of the VPA breaks parallelism mid-list. It’s got an institutional boilerplate structure to it, rather than an essay. All things typical of a document composed by decision-makers and a communications team working together.

The CVPA statement, meanwhile, seems to have all the hallmarks of an LLM-drafted document. While Dr. Murtaugh and Mr. May don’t pull a Byers and fabricate citations or build arguments from hallucination, it has the hallmarks of an LLM-generated, human-decorated document. Its machine-smooth uniformity is interrupted by clunky sentences tacked on at the end of paragraphs, where I suspect an actual human reviewer chimed in.

Yeah, I’m a snob about these things. Blame my high school English teachers. And parents.

Still, I can’t help noticing the difference between the documents and wondering what it says about the people and organizations behind them.

Subscribe now

I don’t think the CVPA is acting inappropriately by publishing LLM-drafted work. Dr. Murtaugh is the president, Mr. May is the executive director, and if the leadership chooses to publish a document, it is certainly within their purview to do so. They’re not doing anything dishonest or fraudulent in doing so.

The documents make for an interesting pair: one riddled with the stylistic tics of LLM-prose, the other with the fingerprints of a corporate communications team project. Observation, not accusation.

Which brings me to a disclosure, which I think is important practice. I’m on friendly terms with Dr. Jay Price, Mission’s CEO, and Dr. Dan Markwalder, Mission’s CVO, and a number of other folks in the Mission ecosystem. I had an, ahem, enthusiastic exchange of ideas with Dr. Murtaugh on the FlareVet podcast earlier this year. I have never met Mr. May.

Given those disclosures, you may weigh my next statement accordingly: Mission Pet Health’s position and statement is not, I believe, one of pure altruism.

That statement isn’t a values statement dreamed up by the marketing team, it isn’t a sly way of positioning themselves for purposes of doctor recruitment, and it isn’t Jay and/or Dan going all misty-eyed about all creatures great and small.

Mission Pet Health is a >900-hospital network. The Chief Executive Officer, the Chief Veterinary Officer, and all the rest of the C-suite run a company of thousands of employees as well as serious financial backing from Shore Capital Partners, Silver Lake, and others. We’re talking billions of dollars. The people who run a company like that have a fiduciary duty to the people who backed them financially. Not duty about which the Pirates of Penzance sing, but duty with a legal definition and responsibilities.

That duty doesn’t evaporate because a couple of veterinarians got their hackles up about a mid-level practitioner. They’re not off the hook because they have feelings.

A Substack essay is a cheap thing to publish. A position statement like Mission’s? One that’s public, signed, quotable forever from one of the largest veterinary employers in the United States? That can be expensive.

The lazy read of the consolidator opposing a new, ostensibly cheaper provider tier is one of turf protection and regulatory capture. The incumbents pulling up the ladder.

But go look at the CVPA’s own star exhibit: the Smith-Lloyd-Harris model. The paper is heavily referenced and cited in the CVPA response, and features prominently on the front page of the CVPA website. It projects that adding a single full-time VPA to a typical practice increases net profit by $177,850 per year. Taking the Coalition’s math at face value, for now, and multiply it against Mission Pet Health’s footprint? You get in the ballpark of $160M in hypothetical profit every year. Drs. Price, Markwalder, and their organization just publicly declined.2

I don’t have insight into Mission’s actual P&L’s, of course, but that’s the math as predicted by the Smith-Lloyd-Harris model on (roughly) the logic of cheaper labor, bigger margins. But Mission looked at the same numbers the CVPA did and said, “no, thanks.”

So does Mission not believe the math or do they believe that there are things the math doesn’t capture? But Dr. Price and Dr. Markwalder have both done something that Dr. Murtaugh and Mr. May haven’t done: they’ve run veterinary hospitals from inside the building this century. And I, with all the bias you might expect, believe that the value of a veterinary hospital is the quality of medicine practiced inside it. I believe that a workforce of doctors and technicians is an asset you compound over decades rather than a cost to be arbitraged over fiscal quarters. Those of us who can still hit veins and pass tubes know it, those of us who still wear stethoscopes around our necks know it, those of us with people who count on us to take care of their animals know it.

I think the Coalition would like you to believe that Mission is playing defense. I do not believe that to be the case. I think Mission is playing for 2046.

Subscribe now

And since the CVPA all but insisted, I did exactly what you’d expect me to do with that Smith-Lloyd-Harris paper: I read it.

The model is presented in the CVPA response as “peer-reviewed research” by authors from Pepperdine University, Animal Health Economics, and Michigan State University. It has the air of independent thought about it, doesn’t it?

But the CVPA letter doesn’t note that acknowledgement section of the paper, where it notes that the research was funded by Lincoln Memorial University’s College of Veterinary Medicine. LMU’s College of Veterinary Medicine, you may recall, is facing probation from the AVMA’s Council of Education for a lack of exposure to research opportunities for its students. LMU has stated in court filings that “research offerings are almost solely the purview of existing traditional land grant, Ivy League, and/or state-funded universities, all of which are overwhelmingly driven and funded by (a) federal grants obtained through the National Institutes of Health (‘NIH’) and the U.S. Food and Drug Administration (‘FDA’), (b) state funding, and (c) large endowments.” Nevertheless, the resourceful folks at Lincoln Memorial found room in the budget for this one.

And yes, that would be the same Lincoln Memorial University where the Master of Veterinary Clinical Care is being studied. And sold. The school selling the credential paid for a study that, gasp, found the credential to be incredibly profitable.

That’s not a crime, but it is something I would disclose if my goal were an accurate record in, as the CVPA puts it, “the spirit of professional dialogue.”

The model’s inputs come from interviews, “more than forty structured interviews with industry leaders and practice owners,” per the letter, which makes it sound like a survey of the profession. Per the paper, the interviewees were primarily drawn from LMU’s own veterinary college and advisory board and industry leaders, “many of whom a priori support the concept of mid-level practitioners.” That’s pulled from page 46 of the Journal of Applied Business and Economics, Volume 27(5), 2025, where the authors of the paper are doing exactly what they’re supposed to do in academic research.

Finally, the paper relies on a world of excess demand. One that is hard to believe with visits falling nationwide. The paper assumes that clients keep coming, that practices keep growing, the capacity gets absorbed, that VPAs charge the same fees as DVMs, and prices don’t drop. All those assumptions, held up next to the CVPA statement and all the claims about “access to care,” start to look awfully dicey to me. The CVPA wants to “lower the price of access” based on an academic model that, unequivocally, doesn’t lower the price of access.

Either the profit argument, which needs prices to hold, or access argument, which needs prices to fall. Those claims are both made in the CVPA document. Pick one.

Call it bias if you like, but I’m willing to bet that a company as big and as well-run and as successful as Mission Pet Health did their homework on this one. I’m willing to bet that they ran the math, more elaborately and accurately than I did, and concluded, more or less, “this ain’t it.”

Which brings me back to the sentence structure favored by LLMs and the CVPA: “this is not an unregulated shortcut; it is a defined, examined, board-registered professional tier.” For the moment I’ll take their word on the credential, and apply it to the CVPA’s own letter. Because whether or not the credential is a shortcut, the letter is. It’s largely unexamined by the signatories’ own pens, built on a study funded by a seller, sampled from the faithful, and robust only in a world its own footnotes call fragile.

The profession, led by the AVMA, AAHA, NAVTA, and now, importantly, Mission Pet Health, put teams of people together and challenged the CVPA with questions about training and about trust. The Coalition answered, with a machine-drafted and self-contradicting piece built on highly specific research, and in doing so demonstrated exactly how much of its own effort those questions deserved.

The CVPA wants the debate to be about X number of hours, X number of credits, X ratio, and X dollars per practice per year. X is comfortable, X can be negotiated and modeled until the objections fade away. But the letter, read next to its own sources, answers a different question entirely: not what the VPA is, but what all the machinery is for.

Because, in the end and after all, it’s not X, it’s why.

The scribe runs, the SOAP note appears, you fix two things, you sign it. Great. You have successfully used an AI tool in veterinary practice. You’re a tech wizard. You might even get that printer working later.1

Subscribe now

But “AI” is not one tool. It’s a category of tools in a larger toolbox, and right now many of us are using the hammer for everything. Screws, bolts, drywall anchors, and the occasional jar of peanut butter? Grab hammer, swing hard. It works, sort of, until it doesn’t, and then we blame the hammer.

Hammer use has become sufficiently widespread in veterinary medicine as to merit refining the conversation. Here are six tools that I think you should know, I’ll tell you what each one does, show you what it looks like in practice, and when to put it down and pick up a different one.

1. Prompts

The thing you’re already doing when you interact with a large language model.

A prompt is one question, one answer. You type something, the model responds, you do something with what it gave you. That’s it. No memory of last time. No connection to your records. Just a conversation.

It’s good at: Anything where you need the model to think once, especially about something specific. Rewriting a client email so a layperson can understand it. Brainstorming a differential for a weird presentation. Summarizing a journal article you don’t have time to read. Explaining a complex diagnosis using a football analogy because you know your client — but maybe not football — and you know what’s going to land.

It’s bad at: Anything you need done exactly the same way twice. A prompt doesn’t remember what you told it yesterday (it could in the right circumstances, but don’t take it for granted). It can’t reach into your records. It can’t send emails. It generates text, presents it to you, and its job is done.

The skill that makes this useful: Want a better answer? Ask better questions. It sounds snarky and glib, but it’s most of what works.

“Write a discharge summary” gets you something generic and vaguely robotic. You know the voice — the em dashes,2 the word “commendable,” the phrase “it’s not just a X, it’s a Y.” Your clients, who, unless we’re neighbors, are somewhat unlikely to be Amish, know that voice too.

But type something like: “Write a discharge summary for a 7-year-old neutered male Golden Retriever presenting for acute vomiting, diagnosed with foreign body obstruction on rads, treated with exploratory laparotomy and enterotomy, recovered uneventfully, discharging on tramadol and metronidazole with recheck in 10 days. I’ve attached the radiology report, surgery report, and pre-anesthetic lab results. Sixth-grade reading level, warm and reassuring, and for the love of all that’s decent don’t say ‘furry family member.’”

That gets you something you’d actually send home. Well, it gets me something that I would actually send home with a specific patient. You’d want to tailor your prompt to suit your own use case.

Something to try: Take the last discharge summary you wrote. Paste it into a paid version of Claude or ChatGPT and type: “Rewrite this at a 5th-grade reading level. Keep the medical accuracy, but minimize the jargon or explain it where it can’t be avoided. Make it sound like a kind but professional veterinarian, not a robot trying to human. Remove any instructions on which the client can’t act.”

Then compare the final output to what you sent home. Then rewrite the prompt (iteration is crucial); add your preferred tone, specify the structure, tell the model what to include and leave out, and notice how the output changes. Try it in a new chat in incognito mode to see how the outputs differ. Now you’re a prompt engineer using the same model but with better instructions you’re seeing a dramatically better result.

2. Scripts

The calculator. No thinking, just math.

A script is code — a few lines of Python, a spreadsheet formula, whatever — that does the same thing the same way every single time. No judgment. No creativity. If the input is 30 kg and the dose is 0.5 mg/kg, the output is 15 mg. Always.

What it’s good at: Anything where the answer should never vary. Drug dose calculations. Reformatting lab imports. Generating a weekly inventory report. Sending automated appointment reminders. Reconciling end-of-day payments. The logic is “always do exactly this,” and you want it done exactly that way forever.

What it’s bad at: Judgment calls. A script can calculate that the metoclopramide dose for a 30 kg dog is 15 mg. It cannot decide whether metoclopramide is the right drug given a patient’s cardiac history and concurrent meds. That’s what the doctor (sometimes with the help of an AI model) is supposed to be doing.

Why you should care even though you’re not a coder: Two reasons. First and most importantly: AI can write scripts for you now. You can describe a calculation to Claude in plain English and have working code in 30 seconds. You don’t need to know Python, need to know what you want calculated. You don’t always need to know how to do it.

Second and this one matters too: the script runs without the AI. No hallucination risk, no variability. The dose is the dose is the dose. It’s calculated. It’s deterministic. For anything where you want the exact same result every time, a script is safer, faster, and way cheaper than asking a language model.

Example you can try right now: Open Claude and type: “Write me a Python script that takes a patient’s weight in kg and a drug dose in mg/kg and calculates the total dose in mg, the volume to draw up given a concentration in mg/mL, and flags if the dose exceeds a maximum I set.” You’ll have a working drug calculator in under a minute and it’ll never hallucinate.

Subscribe now

3. Agents

A model with a to-do list.

An agent is an AI that works in a loop. Unlike a prompt where you ask one question and get one answer, an agent takes a goal and figures out the steps: look at this, decide what to do, do it, check the result, move to the next thing. It can call tools along the way. It keeps going until the job is done.

If a prompt is asking the model a question, an agent is giving the model a mission.

What it’s good for: Multi-step work that requires judgment at each step. Say it’s 6 AM, you have four surgeries on the schedule. You (or a predetermined script) tell(s) the agent: “Go through today’s surgical patients. For each one, confirm pre-anesthetic bloodwork is on file and less than 30 days old (if it’s not, determine the date of the most recent bloodwork), check that a signed consent form is in the record for this procedure, pull up the patient’s current medications, and flag anything that should have been held pre-op.” The agent opens the first chart. Checks the bloodwork date, finds that everyone’s is current. Finds the consent forms, all signed. Reads the med lists, notes the patients on carprofen and flags that it wasn’t held for two of the patients. That’s not one question. That’s a loop of clinical decisions, and you get a summary before rounds instead of discovering the missing consent form when the client is already in the lobby.

Other examples may include scanning lab results as they arrive and alerting you to critical values. Going through today’s records, finding any patients discharged on NSAIDs without a documented renal value in the last 90 days, and flagging them. Processing a referral by pulling the patient history, summarizing the case, and drafting the letter.

What it’s bad for: Simple questions (that’s a prompt) and fixed calculations (that’s a script). Don’t deploy an agent to calculate a drug dose. The overhead and setup isn’t worth the time. To set up an agent to do something like I’ve just described, you would need some time for set up and testing. And more autonomy means more ways things can go sideways.

It’s not a quick fix for all that ails the hospital. But it could be darned useful if you take the time to do it right.

The critical rule: As ever, an agent is a tool. It is not a replacement for the sensible checks and balances that exist in clinical medicine. The safe setup is the agent does the work, presents the result, and waits for you to say “yes.” You’re the pilot. The agent is the eager intern co-pilot who runs the checklists and sets everything up, then waits for your clearance.

Example you can try today: If you use Claude and connect it to your email and calendar, try: “Look at my inbox, find any emails from clients in the last 48 hours that mention their pet vomiting or not eating, and summarize each one with the pet’s name and what’s going on.” That’s an agent loop: it searches, reads each email, makes a judgment about relevance, extracts the information, and compiles a summary. Small-scale, but that’s the mechanic. The surgical pre-check above is the same pattern applied to clinical systems.

4. MCP (Model Context Protocol)

The thing that gives the model access.

An MCP connection is what lets the model reach into a real system — that system could be your email, your calendar, your lab platform, a research database — and actually interact with it. Without MCP, the model can only work with what you paste into the conversation. With MCP, it can go get things itself.

What it’s good for: Anytime the model needs information that lives somewhere else. Things like patient records, lab results, the schedule, an email thread, a journal article, and so on. If the data isn’t in the conversation already, setting up access via an MCP is the way to get it done.

What it looks like in practice: I use MCP for too many things to mention. Claude connects to my Gmail, Calendar, Drive, Slack, Notion, and PubMed. I can say “Find the email from the urgent care about the splenectomy referral, I think the dog’s name was a dessert, pull the original bloodwork from my Drive, and search PubMed for the latest metronomic cyclophosphamide protocols.” And it does all of that in one conversation. Four systems, one instruction, no tab-switching! It saves an incredible amount of time and Mochi the spleenless pit bull is the only one happier than me about it.

Why this is the big deal: MCP is what turns AI from a drafting assistant into something that actually can be molded into your workflow. Without it, the model drafts a beautiful discharge summary and you copy-paste it into a text field. With it, the model drafts the summary and writes it to the record. Without it, the model suggests reorder quantities and you type them into inventory. With it, the model checks stock, calculates reorder points, and generates the purchase order for you to approve.

That gap between “AI helped me write this” and “AI did this inside the system I actually work in” is closed by the Model Context Protocol.

Example you can try right now: If you haven’t connected your LLM of choice to your email or calendar, do it. It takes about two minutes in Settings. Then try: “What’s on my schedule tomorrow and are there any emails I haven’t replied to from this week?”3 That’s MCP in action — the model reaching into two live systems, pulling real data, and giving you a useful answer. Once you see it work, you’ll immediately start thinking about what else you’d want it connected to.

5. RAG (Retrieval-Augmented Generation)

The model’s reference shelf.

RAG is a setup where you give the model a library of reference documents — your formulary, your clinic’s SOPs, a collection of journal articles, your controlled substance protocols — and build a search system so the model can look things up when it needs them. It doesn’t memorize your protocols. It looks them up, every time, right when it needs them.

How it’s different from MCP: Easy to mix these up, and it matters, so here’s the analogy:

MCP is the phone on the wall. You pick it up and you’re connected to something live. Imagine calling the lab, the front desk, the pharmacy’s inventory system. The information on the other end changes by the minute. Fluffy Henderson’s potassium result wasn’t there an hour ago; now it is.

RAG is the protocol binder on the shelf by your desk. Your IRIS staging criteria, the anesthesia protocols, federal and state controlled substance SOPs. That binder doesn’t change when a patient walks in. It’s your vetted, accumulated knowledge — the stuff you’ve decided is correct — organized so you can find it fast.

You don’t call the lab to find out how to stage chronic kidney disease. You don’t open the protocol binder to find out Fluffy Henderson’s potassium. Same information ecosystem, completely different jobs.

MCP gets you the patient’s data. RAG gets you the protocol for interpreting it.

What it’s good for: Anytime the model needs your specific knowledge, not just general medical knowledge. Language models know a lot of medicine. They also hallucinate drug doses, confuse species-specific protocols, and confidently cite guidelines that don’t exist. RAG solves this by grounding the model in your vetted material. The model doesn’t guess what your NSAID protocol says, instead it retrieves the actual document and reasons from it.

Imagine having your vetted, consolidated, collected anesthesia protocols, searchable when the model is helping plan a procedure. Or your controlled substance SOPs, surfaced when a tech asks a compliance question. A curated collection of JAVMA articles on the condition you’re treating, so the model cites real evidence instead of generating plausible-sounding claims.

When it’s overkill: If your formulary fits on two pages, just paste it into the prompt. Or make a spreadsheet. RAG earns its effort when the knowledge base is too big to fit in a single conversation, as in where there are hundreds of pages of protocols, thousands of articles, years of case records.

Example to understand the concept: You know the “upload a PDF and ask questions about it” feature in Claude or ChatGPT? That’s a simplified version of the same idea. The model reads the document you gave it and answers from that instead of from memory. RAG is the industrial-strength, organized version with hundreds of documents, always available, automatically searched. But the principle is the same: ground the model in your material instead of trusting its training data.

6. Skills

The playbook you don’t have to explain twice.

I mostly use Claude for my LLM use.4 A skill is a reusable set of instructions that makes the model reliably excellent at a specific task. If a prompt is a one-off question, a skill is the recipe card you laminated and stuck to the wall. It tells the model what to include, what format to use, what tone to hit, and what mistakes to avoid — every time, without you re-explaining it.

What it’s good for: Any task you’ve figured out and want done the same way consistently. You’ve dialed in exactly how you want discharge summaries — the structure, the reading level, the follow-up instructions, the voice that sounds like you. Instead of re-explaining all of that every time, you encode it as a skill. Now anyone on your team can invoke “discharge summary” and get output that matches your standards without knowing how to prompt.

Other examples: a SOAP note skill tuned to your emergency workflow. A client callback skill that summarizes the case at a lay reading level. A referral letter skill with your preferred structure and clinical voice. A controlled substance reconciliation skill.

When to skip it: When the problem is genuinely new. Skills encode what you already know works. For something you’ve never dealt with before, you want the model’s full reasoning, not a constrained playbook. Use a prompt, solve the problem, and then — if you’ll see it again — build the skill.

Example you can try right now: Take a task you do regularly — say, writing post-op callback notes. Write out every instruction you’d give a new associate about how you want it done: what to include, what to skip, what tone, what reading level, how to handle complications vs. routine recoveries. Paste those instructions at the top of a prompt and then paste a medical record below them. What you get back will be dramatically better than an unprompted request — and those instructions are now the first draft of a skill.

How They Fit Together

None of these tools replaces the others. They stack.

Picture this: You’re closing out your day. You invoke a skill — “end-of-day reconciliation” — that you built last month. It loads your standards for what to check and how to report it. That triggers an agent that loops through today’s patients. The agent reaches through MCPs into your records, your lab system, and your inventory. For the renal case in Room 2, the agent notices a creatinine of 4.8 and searches your RAG library to pull your clinic’s CKD staging protocol — grounding its assessment in your document, not its training data. For the inventory check, it calls a script that calculates reorder quantities from your trailing 30-day usage — a deterministic calculation, no AI judgment needed. Throughout the whole process, the agent uses prompts within the skill’s framework to assess each record, draft follow-ups, and compile a summary.

Fifteen minutes. Zero tab-switching. You reviewed and approved every output. The model did the work. You made the decisions.

Every component of that workflow exists today. And you don’t need all six on day one. Start where you are — better prompts, a script for your most tedious calculation, connecting Claude to your email — and build up as the problems get more interesting.

You don’t have to use all of the tools or even some of the tools, but you should know how they work, which one you’re reaching for, and why.

I was recently on the FlareVet podcast discussing telemedicine with the wildly impressive Dr. Lori Teller.1

FlareVet's founder and co-host, Dr. Jonathan Gonzalez, has lately been taking Dutch Pet to task on LinkedIn over the company's negative advertising, and it came up on air. Curious, I pulled up Dutch Pet's funding mid-episode and ran some back-of-the-napkin-Rainman math live.

Still, the numbers looked awfully ambitious. My rough count put break-even somewhere around 150,000 active members and 750 consults a day. I'd put the odds of it working about where I'd put the odds of me breaking seven minutes in a 2,000-meter row again: technically possible, but a lot of things have to go very specifically right.

Please consider what follows a hobbyist’s deep dive, written for the fun of the format. My last piece, on the Duffield Foundation, was organized, somewhat cheekily, as a medical record — a workup of the “gift horse” of charitable giving. This time I’ve dressed the analysis up as a venture capital investment memo, a genre in which I have exactly zero professional education or standing. I’m not a VC, an analyst, or anyone’s financial advisor; I’m a veterinarian with a spreadsheet, a couple of AQI-cancelled appointments, and an affinity for public records. Nothing here is investment advice, and everything here is checked and cited from publicly available sources.

That said, I (predictably) took the exercise seriously once I was in it. The research rabbit hole swallowed a full day, and the write-up came out north of 9,000 words — which frankly deserves its own opt-in gate before it lands in your inbox. What follows is the approximately 4,000-word version. That felt like the more humane option.

This one is meant to be a little fun. I hope you enjoy.

Subscribe now

I. The Deal

This is an investment memo for an investment nobody offered.

It is written in the standard format because the format is the point: venture diligence is the one genre of American business writing in which flattering a founder costs the author money, and so the one in which every claim gets checked. Dutch has received a great deal of coverage and no diligence — a Forbes profile, a lead investor’s published memo, a one-hour sponsored interview, five years of press releases — none of which pulled a Form 990, read a statute, or divided a price by a labor rate. This memo does those things.

The company: Dutch (Dutch Pet, Inc.) is a direct-to-consumer veterinary telehealth company launched in 2021 by Joe Spector, a co-founder of Hims & Hers. The product: $132 per year for unlimited video consultations covering up to five pets, medications billed separately through a PetMeds partnership, insurance bundled through Pets Best, and referrals into brick-and-mortar clinics — including Petco’s Vetco locations — for everything a video call cannot do.

Worryingly, the company reports 40,000-plus cumulative customers and over one million “veterinary interactions” across all fifty states, in seven of which, as of July 2026, its complete business model is actually lawful.

The capitalization: Approximately $43 million: a seed led by Forerunner Ventures, a $20 million Series A, an $18 million Series B, and $5 million from Jimmy Fallon, whose diligence process is not in evidence. The founder’s stated ambition, on the record: “We want to go public — and rule the world.”

The arithmetic: For the cap table to work, the company must exit at roughly $400–600 million. Consumer health subscriptions trade at 3–5x revenue, so the exit requires $100–150 million in annual recurring revenue — at $132, something over 600,000 active paying members. Every section that follows measures the distance between that number and this company.

The transaction under review. There isn’t one. But Dutch is raising, continuously, from two classes of investor who will never see a diligence process: pet owners, $132 at a time, on the representation that their veterinarian is overcharging them; and state legislators, twelve statehouses at a time, on the representation that a trade association is a cartel. Whoever supplies the capital gets the memo. This one is addressed to them.

Nothing here is non-public; everything can be checked by any reader with an internet connection — the standard we suggest applying to the company as well. An early draft contained an eighth adverse founder finding and a third market filter; both died on review and are retracted in place, reasoning shown, because the memo’s central finding concerns errors that all point one direction, and the only license to make it is a demonstrated willingness to correct one’s own.

The recommendation, stated up front as the format requires, is pass. The reader is encouraged to check the math. That is rather the theme.

II. Founder Diligence

Founders sell; that is the job. But the ratio of claim to fact is a data point, and one of the most predictive we have. Our primary source is the founder’s late-2025 sponsored interview — by the host’s own disclosure, an advertiser placement — checked against primary documents. Eight material claims reviewed.

Claim 1 — Veterinary prices are doubling annually. “Veterinary care has grown 2x the rate of inflation... So something that was $50 is literally $100 a year later.” The claim refutes itself in two sentences: twice inflation is roughly 6%; $50 becoming $100 is 100% — seventeen times his own cited figure, in the same answer. “2x inflation” is the defensible number; “$50 becomes $100” is the one the audience remembers.

Claim 2 — Routine dental cleaning costs $5,000. Presented without geography, invoice, or case detail. A routine cleaning with anesthesia typically runs in the mid hundreds. A $5,000 dental exists, often at a boarded veterinary dentist, for advanced disease: exactly the pathology that results when the $600 version is skipped for fourteen years.

Claim 3 — The physician-to-patient ratio in human healthcare is 1:20 (versus 1:3,000 for vets). The FSMB census counts 1,082,187 licensed U.S. physicians — one per 314 Americans; the founder’s ratio requires 17 million. Apples to apples, DVMs against physicians: 1:1,300 versus 1:314 — a real four-fold gap, presented as 150-fold. Overstated ~25x; the ratio of truth to embellishment recurs.

Claim 4 — The AVMA and state groups have “hundreds of millions of dollars in annual budgets.” We pulled the 990: $68.9 million in revenue against $69.6 million in expenses — the country’s largest veterinary trade association ran a deficit of $657,006. The full filing cuts both ways: $162.2 million in assets, a $1.37 million marketing contract with a firm called Keep Our Pets Safe, a PAC, ~$3.0 million in annual lobbying and political spend — not a bystander. But an annual budget is the thing the 990 measures, and it is $69 million, spent at a loss.

Claim 5 — Regulators tell vets the FBI will arrest them; “millions of dogs will die.” Attributed to unnamed persons at unnamed conferences and hearings “behind closed doors.” No document, speaker, date, or venue cited; we could not locate the phrase anywhere in AVMA publications or testimony. Suspected to be constructed to be uncheckable — a design feature rather than an accident.

Claim 6 — No other profession’s trade association controls school accreditation. “Law schools don’t have a trade association decide who gets to have a law school.” Falsified by the founder’s own example: the American Bar Association is the federally recognized accreditor of American law schools. The AMA co-accredits medical schools; the ADA, dental schools. This is the standard architecture of American professional education and does not represent a “cabal.”

Claim 7 — The price. “It’s $100 for a year of care.” The company’s own website: $132. In an hour dedicated to the proposition that other people’s prices are dishonest, the one number the founder controls absolutely was wrong by 24% — in the direction that flattered the pitch.

Claim 8 — The founding story. WITHDRAWN. An earlier draft flagged that the corgi origin story carried a $2,000 bill in the interview and $4,000 in Forbes. A Chamber of Commerce profile reconciles them: a $4,000 bill, $2,000 covered by insurance — one invoice, two accurate descriptions. We withdraw the finding, and note why: the innocent explanation was one search away, and the draft didn’t run the search, because an eighth adverse finding completed a pattern the memo wanted complete. Pattern-hunger is not the founder’s pathology alone; the seven findings above survived the same check, and this one did not.

Pattern: seven claims, seven favorable errors, zero unfavorable ones. Random imprecision distributes in both directions; this distribution has a mean. In our experience, it does not improve after investment.

III. Market

As pitched: 90 million pet-owning households; a $100 billion pet sector; 129 million Americans in “veterinary care deserts.” But $100 billion is all pet spending — food, toys, grooming; citing it as the market for a video-call subscription is like citing global food spending as the TAM for a recipe app. And the vet-deserts statistic originates in Dutch’s own marketing research. We do not weight self-sourced TAM studies.

As diligenced: Filter one: legal geography. The binding constraint is whether the veterinarian-client-patient relationship can be established remotely — without that, the video call cannot lawfully originate the prescription carrying the margin. As of July 2026, seven states permit the full model: Arizona, California, Florida, Ohio, New Jersey, Virginia, Vermont — roughly 101 million people, 30% of the country. Filter two: willingness to prepay for virtual-only care. The best proxy — pet insurance, which covers the $4,000 emergency owners actually fear — has reached 4–6% of U.S. pets after two decades of marketing. Apply a generous single-digit rate: perhaps 1 to 2.5 million households.

A third filter is retired here, in place. An earlier draft read the founder’s flagship statistic — half of Dutch’s customers haven’t seen a vet in three years — as evidence of a base selected for non-spenders. Dutch’s own CMO refuted it, citing the PetSmart Charities finding that 52% of U.S. pets receive no veterinary care at all: a base that matches the base rate has not been selected for anything. The statistic is uninformative in both directions; the founder should stop citing it, and so do we. What survives is the company’s own segmentation — a cohort its CMO calls “doubtful pet parents... the skeptics of veterinary medicine,” acquired through anti-incumbent messaging and awaiting conversion.

As observed. The entire U.S. veterinary telehealth market — Dutch, Vetster, Airvet, Pawp, everyone — was approximately $92 million in 2024. The revenue required for the exit management describes exceeds the current size of the entire category. Not Dutch’s share. The category. Even granting analysts’ ~20.6% growth curve, $100 million in ARR by 2030 requires 35–50% of the national market, captured by its fourth-best-funded entrant. And the demand side agrees: AVMA benchmarking shows active clients per veterinarian declining since 2021 — the channel management insists is turning desperate customers away is seeing fewer patients each year. There is a real access problem; the 52% figure describes it exactly. But an access problem is not a market. A market is people who will pay, and the people who will pay are overwhelmingly already paying, in person, where the medicine is.

IV. Unit Economics

Revenue: $132 per year — $11 a month, $2.20 per pet at the household maximum — for unlimited consults. All additional revenue depends on pharmacy attach.

Labor: Dutch’s own job posting for a Virtual Care Veterinarian offers “pay per consult — average $50–$80/hour,” 1099, no benefits, nights and weekends preferred. At three to four consults per hour including documentation, roughly $15–25 per consult; complex cases run to $40. For calibration: TelaVets — a transactional competitor charging $65 flat per visit — advertises $100–150/hour for the identical remote work. Dutch’s rate is the bottom of the observed market, not the market. And note TelaVets’ price sheet doing arithmetic on Dutch’s: a year of unlimited visits costs two market-rate consults. Note also the flagship AI announcement — vets “see twice as many pets per day” — which, for a piece-rate workforce, is a plan to compress the piece rate.

The worked example — management’s own showcase customer. Dutch’s marquee use case is chronic canine anxiety, and the founder describes the care model himself: “you kind of need to have those regular check-ins,” citing an owner “on Dutch for 2 years.” Monthly check-ins is twelve consults a year: $180–300 in direct veterinary labor against $132 in revenue — $300–480 at the tail — before payment processing, free shipping, support, the EMR, or a dollar of acquisition spend. The showcase customer loses money at every point in the range, unless the pharmacy attach rescues the account — and the company has published no attach-rate data, ever.

Adverse selection. The households for whom $132-unlimited is a bargain are the high utilizers; the healthy-pet households who would subsidize them have the least reason to buy. And “your vet is overcharging you” selects the price-sensitive and heavily-utilizing at once: a buffet whose advertising targets the hungriest customers.

Retention flag. The company reports it “doubled member retention” after moving to annual-only billing. When a monthly plan is eliminated, retention measured inside the twelve-month window rises by construction because a prepaid annual commitment “retains” nearly 100% at month twelve regardless of whether anyone is happier. The reported doubling is consistent with a definitional artifact; one cohort table, measured at renewal, would resolve it. None has been published.

Break-even. Charitable case: two consults per member per year, labor $30–80, pharmacy contribution $20–35, processing and support ~$20, fixed costs $8–12 million (all our estimates; the company discloses nothing). Contribution: roughly $50–115 per member-year. Break-even: 70,000 to 240,000 active members, against a lifetime cumulative total of 40,000. The venture case requires 600,000-plus — over a million consults a year, staffed by professionals offered up to double per visit elsewhere in the same job feed.

These economics do not improve with scale. They are the same buffet, with more tables.

Subscribe now

V. Go-To-Market

Dutch runs three scripts at once, and they do not describe the same company.

To consumers, through the founder on sponsored national media, the profession is a “cartel,” a “conspiracy,” a “monopoly,” an industry where a third of revenue “is basically selling you more stuff,” whose exam-room recommendations are “a list of all the services they want to sell to you.” Substitution through distrust.

To the profession, through the CMO on a veterinary podcast: “Our goal here is not to compete but to complement.” Vets are paid per pet helped, not commissioned on products; it saddens her that people are losing trust in veterinarians. Partnership through reassurance.

To buyers, through the paid creative: neither. The app-store listing leads with price arithmetic — unlimited visits “for less than the cost of one vet visit”; nobody price-compares against a thing they are supplementing. And the anxiety creative is a stepped funnel whose terminal step is not a consult but a shipment: risk-free visit, prescribed treatment plan, “treatments delivered right to your door for free.”

The finding is divergence, not dishonesty — the CMO’s no-sales-incentive claim, taken narrowly, is plausibly true, but it describes the labor layer; the creative describes the business, and the funnel it purchased terminates at the pharmacy. Only one of the three scripts was bought at CPM rates, which makes it the revealed strategy. A company is what it pays to say.

Three structural contradictions follow. First, the selection effect: a skeptic cohort, acquired by advertising that manufactures the skepticism, must be converted into recurring pharmacy customers by exactly the recommendation — a vet proposing ongoing treatment — the ads taught them to hear as an upsell. The company spent its marketing budget inoculating its customers against its revenue model. Second, the referral layer: Dutch refers everything acute and hands-on to brick-and-mortar clinics, making the profession its fulfillment layer — then poisons it, so members arrive pre-loaded with the teaching that the treatment plan is a sales script. Third, the labor pool: every DVM Dutch will ever hire works inside the system its ads call a conspiracy. We are aware of no company that has scaled a professional-services workforce while running attack ads against the profession.

The CAC arithmetic closes the section: national podcast sponsorship, CTV, and app-store placement are acquiring customers worth roughly $50–115 in annual contribution. The category’s survivors sell through employers and marketplaces. In motions where someone else absorbs the CAC; Dutch chose the only one where it pays full retail. And its quiet distribution deals (PetMeds, Pets Best, Petco’s clinics) all route through the legacy infrastructure its advertising exists to discredit. The new model distributes through the old one.

VI. Regulatory

In a standard process, this chapter asks whether the rules might change against the company. Here the rules must change for it, in dozens of jurisdictions, on a venture timeline. Regulatory change is not a risk to this business. It is the business.

The map. The full model is lawful in seven states, ~30% of the population. The founder’s “about 20 states don’t allow telemedicine” implies thirty that do; the verified count is seven. Even his description of his own obstacle was favorable, by a factor of four.

The velocity. Four full-model conversions in four years — Arizona, California (2023), Florida (2024), Ohio (2025) — roughly one state per year, decelerating. Since the founder’s late-2025 interview: zero. The 2025 sessions were a rout: Texas’s bill died at sine die, Colorado’s was defeated, reform failed in four states while three reinforced in-person requirements. At one state per year from a 30% base, an open map is a project measured in decades, against fund clocks expecting liquidity within the one. The honest summary: companies with product-market fit acquire customers; companies without it acquire legislation. When a company’s flagship consumer website is a lobbying portal — savepuppies.com — the diligence question answers itself.

Texas, closed. The narrated path into the largest blocked market ran through Hines v. Pardue, whose Fifth Circuit ruling protected one retired vet’s pure online advice. It was not diagnosis, it was not prescribing. Certiorari denied, April 2026; the in-person statute stands. The founder-diligence residue survives the docket: on tape, asked who was litigating against the theory his largest market depended on, the founder answered “Greg. Forget his name.” The Attorney General prosecuting the petition is Ken Paxton and Greg Abbott is the Governor. We have never through a pitch where the founder was hazy on both the opposing counsel and the prize.

The budget asymmetry: the AVMA’s observable advocacy complex totals roughly $4.5–4.7 million a year — real, professionally run, and ~40x below the founder’s “hundreds of millions” — while his “we don’t have the crazy budget these trade groups have” comes from a $43 million company running twelve state campaigns and sponsored national media. The party that publishes its books is not the one that raised $43 million.

The moat inversion. Assume total victory: telehealth VCPR legal in fifty states. The regulatory barrier is the only thing keeping serious capital out of the category. A nationally legal market is immediately addressable by Chewy — $11 billion in revenue, twenty million customers, an existing pharmacy, telehealth already priced at zero for Autoship — by Mars, by Petco, by every insurer. Dutch’s $43 million and its proprietary EMR do not survive contact with Chewy’s checkout page. The strategic box: if the lobbying fails, the market never gets big enough; if it succeeds, the market immediately belongs to someone else. The company is spending its investors’ capital to build a public road and calling it a moat. We generally decline deals where the company must beat the government to survive. We have never seen one where beating the government is also how the company loses.

VII. Comparables

The closest comp is nearly forensic.

Fuzzy Pet Health: subscription-priced 24/7 vet telehealth, ship-to-home prescriptions, embedded insurance — substitute Dutch’s fee, its PetMeds deal, its Pets Best bundle, and the two pitch decks are separated mostly by fonts. Fuzzy raised $80.5 million, nearly twice Dutch’s total, and on June 16, 2023, ceased operations overnight: apps down, paychecks late, 401(k) deductions unaccounted for, the CEO’s profiles vanished. A former employee’s post-mortem required one sentence: “the business model was not economically viable, and the company could not scale.” That is the base rate for Dutch’s exact model, executed with double the money.

The survivors share one feature: every one stopped selling prepaid telehealth subscriptions to pet owners. Airvet pivoted to employer benefits and reports 4x growth since renouncing Dutch’s strategy. Vetster runs a marketplace carrying no labor cost. Otto — raised $43 million, Dutch’s exact capitalization — pivoted into software sold to clinics; now making its living selling tools to the “cartel.” And Chewy supplies the pricing signal: it gives telehealth away free with Autoship. When an $11 billion incumbent prices your entire product as a giveaway, the market has ruled on destination-versus-feature, and the ruling was not close.

Management’s preferred comp, Hims, is a category error. The Hims customer is the patient: he self-diagnoses (the mirror performs the exam), his medications are lifetime generics at software margins, and no VCPR statute stood in the way. Dutch’s patient cannot speak, is examined through a phone camera held by an untrained owner, and generates its highest-value medical events — the blockage, the fracture, the mass — precisely where telehealth cannot go. The founder built a company whose every structural advantage is structurally absent here.

The complete list of companies that have attempted Dutch’s model at venture scale: Fuzzy, and Dutch. Survival rate to date: zero of one, one trial in progress. We are asked to believe this time is different. Diligence found one difference: less money.

VIII. The Bull Case

A memo that cannot state the case for the deal has finished prosecution, not diligence. So, argued as its best advocate would argue it:

Thesis one: the strategic acquisition. The memo’s own findings — telehealth survives as a feature, not a destination; victory hands the market to Chewy, Mars, Petco, the insurers — describe an asset worth more inside a platform than standing alone. What the acquirer buys: a national brand built by $43 million of paid media; five years of fifty-state clinical data; a purpose-built EMR and AI stack; a credentialed 200-vet network; active relationships in twelve statehouses. Every name on the bear’s list of executioners is a name on the bull’s list of bidders. Comparable tuck-ins — Hims’s $190 million Apostrophe deal, Ro’s $225 million-plus Modern Fertility deal — support a $100–200 million range, at which the Series B returns capital and the seed makes 4–5x. Perhaps not enough to “rule the world,” but not a zero. The trigger: one Texas-class state converting, which restarts every acquirer’s build-vs-buy clock while Dutch is the only pure-play of scale on the shelf.

Thesis two: the pivot. Airvet’s employer-channel move is the category’s documented survival play, and it deletes the memo’s worst findings — someone else absorbs the CAC, the conspiracy creative retires, the skeptic-selection problem dissolves in an HR portal — and Dutch’s assets map onto the channel almost without modification.

The founder card. Hims was ridiculed at this exact stage — a Viagra website with venture backing, attacked by the incumbent profession — and went public above a billion dollars. Nearly every objection here has a 2018-vintage twin. Perhaps the pattern detected is simply what this founder’s playbook looks like in year five.

The rebuttals, at full strength. Against the acquisition: the acquirers’ one revealed data point. Fuzzy was available in June 2023 for the cost of a phone call, and no strategic bought it — Mars hired its CVO instead. The observed clearing price for a distressed vet-telehealth platform is a salary. Worse, Mars owns VCA and Banfield — the literal “cartel” of Dutch’s creative; the brand is a contingent liability to its likeliest buyers. (The talent from the last such experiment, Fuzzy’s CVO included, now works for the incumbents.) Against the timing: the option’s expiry is the runway, conversions are decelerating, and the anchor comps priced in a 2021 market that no longer exists. Against the pivot: it fires the growth engine $43 million built, writes off the DTC brand as the B2B liability it becomes, and enters a channel where Airvet has a multi-year head start. The honest label is restart, and restarts prefer new cap tables.

IX. Recommendation: Pass

An earlier draft arrived at the pass easily; revision took the ease away, and the pass survived anyway.

The bull case, answered. The acquisition thesis fails on pricing, and the pricing answer is dispositive: at a $150 million exit the B returns capital on preferences; at $200 million, perhaps 1.5x. That is the bull’s ceiling — contingent on a statehouse, a buyer’s mood, and a window. Venture pricing buys the right tail, and the bull case’s right tail is the bear case’s consolation prize. We decline to pay growth prices for a recovery thesis. The regulatory head start, the bull’s best asset, is worth the most the day before the trigger fires and approximately nothing after, because the victory that creates the acquirer’s urgency converts Dutch’s scarcest asset into public law. The pivot is real, which is why it appears among our reversal conditions rather than our reasons to invest: we do not fund the availability of a pivot, only the execution of one. And the founder card: this process keys on structure and on tape. The structure — self-diagnosis, generic margins, no VCPR — was verifiable at Hims in 2018 and is verifiably absent here; the tape — seven for seven — is founder-specific evidence no 2018 memo could have contained. A prior success raises our prior on execution; it cannot repeal the arithmetic of a $132 fee against $180–480 of showcase-customer labor. Execution in service of negative contribution scales the loss.

Concessions. The problems the founder describes are real: costs above inflation, private-equity consolidation, a genuine access gap — the 52% figure is the most honest number in this memo. This memo defends arithmetic, not the association. But whatever ails this industry, the cure will not arrive from a company whose numbers do not check, whose diligenced customers cannot fund it, and whose legal victory would be its competitive death.

What would change our mind: published clinical outcomes data — any at all; audited active-member counts replacing cumulative press releases; the cohort table, measured at renewal, that would resolve the retention flag; disclosed pharmacy attach rates; positive contribution margin at the utilization management itself advertises; an executed — not announced — distribution pivot; a Texas-class conversion; or a fundraise priced to a realistic outcome. Every item is within management’s power, several would cost less than a month of marketing spend, and none has been produced in five years. When the evidence that would exonerate a company is cheap, available, and perpetually absent, diligence is entitled to draw the inference. We have. The standard we applied to ourselves — retract, correct, show the work — is the standard these conditions apply to management. It is not a high bar. It is a cohort table.

A note for our limited partners. Dutch is not raising from us. It is raising from pet owners, $132 at a time, on the theory that their veterinarian is robbing them; and from state legislators, on the theory that a trade association running an operating deficit is a hundreds-of-millions-dollar cartel. Both should pass, for the same reason we do: not because the problems are fake, but because the numbers attached to them fail verification seven times in seven, always in the same direction — and a man whose numbers fail at this rate is asking for trust he has not earned and, on the evidence of Section II, does not expect to need.

To the profession, one closing note: the company is not the threat; the road is. Dutch will most likely follow its predecessor into the graveyard — or, the better branch, a tuck-in — but the legislative map it is opening will outlive it, and the entities waiting on it have the customer bases, pharmacies, and balance sheets to use it. The task is not to defeat the worst messenger telemedicine has ever had; it is to make sure that when the road opens, veterinarians own the practice of medicine on it. Worry about that. Do not worry about Dutch.

A note on the management meeting. We invited management to defend the thesis in an open forum some time ago. Management declined. Having completed diligence — including the strongest case that could be made on management’s behalf, which we wrote ourselves because no one else had — we no longer consider that a fair request. Of him.

My wife and I have become enthusiastic fans of the Philadelphia Union, and this week will be the team’s first game back after the break for the World Cup.

The Union won the Supporters’ Shield last season but have been sub-par disappointing underachieving bad this one. Currently the Bimbos1 are in last place, with four draws, ten losses, and just one win.

The offseason saw the team lose or sell a number of key players and, as a fan, it’s way less fun to lose than to win.

But even at the bottom of the standings, the players make it easy to root for the team. My wife and I took the boys to a “Meet the Team” event to get autographs last week. And the players, there for hours, in an uncomfortably warm field house, with hundreds of people to meet and greet, were gracious and friendly and kind all the way through. Makes it good to be a fan.

Subscribe now

And so, with all the worrying delusional pathological stubborn optimism of Philly sports fans, we’re very excited about this week’s match.

Firstly, because it will likely be the return of midfielder Quinn Sullivan, a homegrown talent and rising star, who has rehabbed from a season-ending ACL tear late last season. We’ve had fun watching Quinn go from a talented-but-raw prospect to a key starter for the team.

Sullivan was injured in a 6-0 demolition of rival DC United on a rainy September night on Potomac Avenue. The thing about that story that gets me is how Quinn reacted afterwards. Do you know what the then 21-year-old posted to social media? He thanked the other team’s medical staff for the diligence and quality of their care. That shows so much professionalism and class both from DC United and from Sullivan. Stories like that don’t get noticed often enough in sports.

And we will also see the return of left back Kai Wagner, who was sold in January to Birmingham City FC. It’s easy to be inspired by Sullivan’s dedication and hard work that led to his return to the field: injury, grind, rehab, comeback, triumph. Sacrifice toward the summit. But Wagner’s story is different.

Kai Wagner is a 29-year-old German and, after a season that saw his team hoist the Shield and become the Union’s all-time assist leader, he was sold to Birmingham City FC. It was a heartbreaking but understandable move for both the Union and Wagner. The Union sold a player at the peak of his value, and a player in his prime got to play on a bigger stage, in a better league, just one notch below the highest levels of European football.2

After just seven months with the club, Kai Wagner is no longer with Birmingham City.

And not because he was underperforming. He immediately looked capable of playing at the Championship level. In his league debut against eventual league champions Coventry City, he recorded an assist — seriously, the man has a left foot you just can’t believe — in a 3-2 victory. In his second appearance, he scored against Cambridge United.

While he is ostensibly a defender, lining up on the back line, he frequently plays in an attacking role. He created 31 chances (passes or dribbles that lead directly to a teammate's shot) in 16 starts, more than any other player on the team. And, perhaps more tellingly, his team had a +5 goal difference while he was on the field.

He’s gone because he asked to leave Birmingham City. A player in his prime, playing at what may be the peak of his professional achievement, wanted out.

Not because the team was doing poorly, not because of any off the field drama, not because he didn’t fit in with the squad or the coaching staff or even because he felt he was underpaid.3 In fact, the video he posted to social media is unapologetically emotional.

He wanted to go back to Philly because his family was unhappy in Birmingham.

Subscribe now

I’ve written a lot about being a son and being a dad in the past year, and I find myself moved by the decision of a man who would voluntarily step down from the rarest of career opportunities to do what he felt was best for his kids. And that famously (and frustratingly) frugal Philadelphia Union organization paid a higher price than they’d initially received in order to get him home.4

Sure, there’s a fan part of me that delights in the fact that my team will be better for the return of two important players. Or how there’s no small amount of joy in thinking how depressing it must be for the rival New York Red Bulls to come to Philly, anticipating a team on the ropes, and instead they’ll find a revitalized, reloaded, and ready club. I could go on about how Quinn and Kai were key facilitators in the team’s offense, which has, without them, been something of a disappointment hard to watch abysmal bad this season.

But I think the better story, the one that means more to me, is the one about what all that talent and effort is really for.

Kai Wagner was building a European career in professional football, and every on-pitch measurement indicated that he’d succeed. He’d earned the move after seven seasons and, arriving mid-season, in a tougher league, started every match. Man of the Match in his debut, and nicknamed Mr. Consistent5 on the club’s social media.

We tend to talk about, and often subconsciously treat, career accomplishment as if it were a destination all its own. But I’m inclined to think of accomplishment as currency. And currency hoarded is just paper.

It’s only valuable because of what you do with it.

While Quinn Sullivan spent his getting back to the pitch, Kai Wagner spent the currency he’d accumulated on his family’s happiness. I don’t think Kai’s a man who abandoned his ambitions, I think he’s a man who correctly remembered the denominations of those ambitions.

And then there’s the Union, a club that sold him and other stars at the peak of their market value, paid a premium to bring back Kai Wagner. And while an argument can be made that Kai’s return will generate ticket and jersey sales, it will still be a loss on any balance sheet. But I’ve been in the stands to witness the wins, the losses, and the draws over the past few seasons, and I suspect that the closest we’ll get to measuring the impact of Kai Wagner’s return to Subaru Park will be in decibels on Wednesday night.

I came to my office this morning to write about a telehealth company or maybe a governor’s race and a state veterinary board, and my mind kept wandering home to my team. Kai Wagner went to England to build the career of his dreams, and his heart, with his family, kept wandering home to Philly. I used to think of that kind of pull as a distraction, a flaw in the focus, a failure of discipline, something that can be solved with adequate effort instead of considering the reason for the pull in the first place.

Not so much anymore.

So this week we’ll go watch a last-place team with one win, and we’ll cheer as loud as we did the night they hoisted the Shield. Quinn Sullivan will be back on the field he fought a year to reach. Kai Wagner will be with the team he gave up the EFL Championship to play for.

The standings say the season is lost.

But the standings only tell you where the club stands. On Wednesday night, we’ll be there to see what the club stands for.

There’s a well-regarded engineering college in upstate New York that now bears the name of a man who made a promise to a miniature schnauzer.

The engineering college preceded the man, who preceded the miniature schnauzer.

In the early 1980s, David Duffield was in the uncertain years between software companies. It was a stretch before PeopleSoft, when nothing was assured and everything was leveraged. Through it, Maddie, the aforementioned miniature schnauzer, was feisty, devoted and always at the founder’s side. David Duffield would later call her the lighthouse of that stormy period. One evening, playing with her on the living room rug, he and his wife Cheryl made her a promise. If the money ever came, they told her, they would give it back to her kind, so that other people and other animals could have what the three of them had.

The founder and his partner led PeopleSoft, a human resources management system, to success. PeopleSoft would be sold to Oracle in 2005 for more than $10 billion.1 And Mr. Duffield’s next act, Workday, became another, with an initial IPO valuation of $9.5 billion. Mr. Duffield is the founder and co-CEO of Ridgeline, Inc, currently valued at more than $300 million.2

Mr. Duffield kept his promise to Maddie.

It is difficult to overstate, even for me, what that money has done. Maddie's Fund effectively led the no-kill movement into the American mainstream. And at a time when the idea that shelters could stop euthanizing healthy, adoptable animals was considered naïve, the Duffields declared it a national goal and spent relentlessly toward it.

They built the academic field of shelter medicine nearly from scratch, funding the first comprehensive program at UC Davis and seeding others at Cornell, Florida, and Wisconsin. More than $301 million in grants later, euthanasia rates that once seemed immovable have plummeted in community after community, and Maddie's Fund can claim a share of that credit that few institutions in animal welfare can match.

Nor is the family’s generosity confined to animals. The Dave & Cheryl Duffield Foundation, established in 2016 and headquartered in Incline Village, Nevada, holds roughly $873 million in assets and funds, among other things, service-dog programs for veterans with PTSD through Liberty Dogs. And there is the family’s long devotion to Cornell, where Dave pitched for the baseball team and played bass in a rock band in the early 1960s.3 First Duffield Hall, the university’s nanotechnology hub, built and maintained by the family since 2001, and then, announced this year, the largest single gift in Cornell’s history and one that renamed the entire engineering school the David A. Duffield College of Engineering.

This is not a family that hides. Their name is not just on buildings but on colleges. Their origin story is on their websites, told with sincerity, schnauzer and all. By every visible measure the Duffields are exactly what they appear to be: people of extraordinary wealth who love animals, honor their commitments, and give at a scale at which most philanthropists only gesture.

Which is what makes what their money has been doing in Colorado so interesting.

Subscribe now

Presenting Complaint

Here is the thing you must understand before anything else in this story makes sense: the problem the Duffields set out to solve is real.

In December 2018, the Access to Veterinary Care Coalition, a partnership of veterinary providers, animal welfare organizations, and social service professionals working with the University of Tennessee, released the first serious national study of who actually gets veterinary care in America, and who doesn’t.

The study was commissioned with a grant from Maddie’s Fund, which is to the Duffields’ credit; before funding answers, they funded the question. The findings were sobering, with roughly 28% of pet-owning households experiencing a barrier to veterinary care in the previous two years. The dominant barrier, across every category of care, was money: 80% of those who couldn’t obtain preventive care cited financial constraints, 74% for sick care, 56% for emergency care. Dr. Michael Blackwell, the coalition’s chair and a former dean of Tennessee’s veterinary college, called it what it was: a complex societal problem, not a failure of any individual clinic or client.

The University of Tennessee’s follow-up study, a bilingual national survey of more than 5,000 pet families, replicating the 2018 work, found that barriers to care have increased, hitting low-income and younger households hardest.

Meanwhile the AVMA’s own data shows average household spending on veterinary care rising 8.9% in a single year, from $514 to $560. Every practicing veterinarian in America can put faces to these statistics. I’m a clinician, and more than sometimes, and I’ve stood across the exam table from people who love their animals completely and cannot afford what those animals need, and there is no version of that conversation that anyone walks away from feeling good.

So let me be unambiguous, because everything that follows depends on it: the diagnosis is correct. Access to veterinary care is a genuine problem, it is getting worse, and the people who first put rigorous numbers to it deserve gratitude. The Duffields looked at this landscape and saw suffering that money could address. That is what good philanthropists do.

The question this story asks is not whether they saw a real problem, I think, but rather it’s what their money decided the solution was and who got to decide.

Systems Review

If you want to understand what the Duffield Foundation built, it helps to remember what they built before. Dave Duffield did not become incredibly wealthy by selling software; he did it selling systems. He sold platforms that sat underneath thousands of companies and quietly ran their payrolls, their benefits, or their org charts. PeopleSoft and Workday are infrastructure. You don’t see them, you operate inside them.

And the philanthropy works the same way. The Duffields do not, by and large, hand out checks to shelters and hope. They build infrastructure, layer by layer.

At the bottom, research: the 2018 study that defined the problem was Maddie’s-funded, and so is the University of Tennessee’s Program for Pet Health Equity, which has received nearly $12 million from Maddie’s Fund since 2018 to research, develop, and stand up something called AlignCare — a subsidized-care system, modeled loosely on human health-care safety nets, that connects low-income families to participating veterinary practices with the help of social workers.

In 2023, AlignCare graduated from the university and became an independent nonprofit, led by executive director Anna Stout. This is textbook venture philanthropy: incubate inside an institution, then spin out.

Next layer up, practice tools. This February, the Dave & Cheryl Duffield Foundation announced a $1.9 million, three-year grant to Open Door Veterinary Collective, a nonprofit founded by Aimee St. Arnaud with the stated goal of building a national network of financially friendly veterinary care within fifty miles of every American, to launch a program called Yes to Care.

The design is genuinely clever and I can’t help but appreciate it: participating practices get communication training from Open Door, guaranteed payment-plan software from a company called Varidi, an in-house charitable “angel fund” administered through the myBalto Foundation, and $5,000 of Duffield money to seed it. Open Door also runs PetHelpFinder, a national directory connecting pet owners to affordable care, whose program leadership includes veteran animal-welfare executive Heather Cammisa.

And at the community layer: partnerships. Maddie’s Fund appears as a Foundation Partner of Humane Colorado, the Denver institution formerly known as the Dumb Friends League, one of the largest animal welfare organizations in the country. Alongside all of it runs a fellow-traveler network of national funders, most prominently PetSmart Charities, whose co-sponsored Gallup surveys supply the very statistics the ecosystem cites, 52% of pet owners skipping needed care; 94% of veterinarians constrained by client finances.

Step back and admire it, because it deserves admiration as a piece of systems engineering: research to define the problem, deliver organizations to address it, software to operationalize it, community institutions to localize it, and survey data to measure it, publish research data to justify it. Every layer funded (and much of it built) by the same family’s money. It is exactly what you would expect from a man who spent forty years building enormously successful enterprise systems.

Every system, as I suspect Mr. Duffield could tell you, eventually needs to go into production. It needs a deployment environment.

The one this system chose was Colorado.

Localizing the Lesion

Get out a map and place pins in the geography of everything described so far.

The Duffields live in Incline Village, Nevada, on the shore of Lake Tahoe; their family foundation is headquartered there too, and its most personal grants (i.e. the high school, the local humane society) stay close to home. Maddie’s Fund operates from Pleasanton, California. AlignCare was incubated in Knoxville, Tennessee. Open Door Veterinary Collective’s grant announcement carries a dateline of Grand Rapids, Ohio. Cornell University is in Ithaca, New York. Nothing in this network is from Colorado.

With one exception: Humane Colorado, the Denver institution formerly known as the Dumb Friends League, where Maddie’s Fund appears among the Foundation Partners.

And yet Colorado is where the experiment runs.

On November 5, 2024, Colorado voters approved Proposition 129, creating for the first time anywhere in the United States a midlevel veterinary practitioner called the Veterinary Professional Associate. A master’s-degree veterinary professional who is authorized, under a veterinarian’s supervision, to diagnose, develop treatment plans, and perform surgery.

The measure passed with about 52% of the vote. It was placed on the ballot by a committee called All Pets Deserve Vet Care, sponsored by Humane Colorado, which put roughly a million dollars behind it, alongside national supporters including the ASPCA and the Humane Society of the United States. The opposition, led by the AVMA under the banner Keep Our Pets Safe, actually outspent the yes side, raising about $1.5 million, and lost anyway. Colorado State University, which, in a not-so-remarkable-coincidence, had been developing its Master of Science in Veterinary Clinical Care program before the votes were counted, remains the only institution in the country training the new profession; its first class graduates in 2027.

Let me be precise about what the record does and does not show, because precision is the whole point of this exercise. No document I have found shows a Duffield dollar in the Proposition 129 campaign. The campaign’s funders are a matter of public record, and the Duffield Foundation is not among them.

What the record shows is something quieter and, I’d argue, more interesting: that the ballot initiative creating a new veterinary profession, the community institution that sponsored it, the university program training it, the advocacy organizations promoting the access-to-care framework it rests on, and the research base justifying it all draw water from the same philanthropic watershed. A watershed sits a thousand miles from Denver.

The people of Colorado, it should be said, did vote for this. Fifty-two percent is a majority. But fifty-two percent of a general electorate answering a ten-second question about pet care is a different thing from what came next. A ballot measure only creates a profession on paper. Someone has to write the rules: the scope, the supervision, the training standards, the credentialing. In Colorado, that someone is the State Board of Veterinary Medicine.

Findings

The Colorado State Board of Veterinary Medicine has nine members, appointed by the governor. Under the statute that structures it, most seats belong to licensed veterinarians and veterinary technicians. Two do not. Here is the eligibility requirement for those two seats, quoted in full from § 12-315-106(2)(c) of the Colorado Revised Statutes:

“Two members to the board from the public at large who have no financial or professional association with the veterinary profession.”

That is the entire provision. It exists for an obvious reason: on a board otherwise composed of the regulated profession, two seats are, on a plain language reading, reserved for the public. The pet owner, the taxpayer, the outsider with no stake. The public members are the check. That is the design.

Colorado’s two public-at-large seats are held by Anna Stout and Heather Cammisa.

You have met both of these women before.

Anna Stout is the executive director of AlignCare Health, Inc. The aforementioned access-to-care organization researched, developed, and stood up with nearly $12 million in Maddie’s Fund money at the University of Tennessee before spinning out as an independent nonprofit in 2023. Her professional role and her livelihood are the organization; the organization’s business is connecting families to veterinary care through participating veterinary practices. And its founding funder is the Duffield philanthropic network.

Heather Cammisa is a member of the leadership team of Open Door Veterinary Collective. The organization that received a $1.9 million grant from the Dave & Cheryl Duffield Foundation this February, whose stated mission is building a national network of veterinary care, whose programs exist to change how veterinary practices handle money, and whose PetHelpFinder directory routes pet owners to veterinary clinics.

None of this is hidden. I didn’t hire a private investigator to determine this, I found it in public records, company websites, press releases and LinkedIn profiles. These are honest-to-goodness professionals doing publicly documented work. …which is the problem.

The question is not whether Anna Stout or Heather Cammisa concealed anything. The question is how a person whose organization is funded to work with veterinary practices, on veterinary economics, in veterinary care delivery, satisfies a statute requiring “no financial or professional association with the veterinary profession.” Not a minimal association. Not a disclosed association. No association.

If you’re inclined to argue that running a veterinary access nonprofit is not an association with the veterinary profession, the profession itself disagrees with you — as of this week. At the AVMA’s annual convention in Anaheim, currently underway, Open Door’s director of veterinary services, Dr. Kristin Jankowski, presented continuing education to veterinarians in a symposium on delivering accessible care. An organization that employs a director of veterinary services, who teaches veterinarians at the profession’s flagship convention, has a professional association with the veterinary profession by this author’s very-much-not-a-lawyer reading of the statute.

And the statute is written in the present tense, “have no association,” which makes this a live eligibility question, not a quibble about paperwork from appointment day. It is live at the worst possible moment. Because since January 1, the board these two seats sit on has been doing the single most consequential thing it has ever done: writing the rules for the Veterinary Professional Associate — the scope, the supervision, the training standards, the credentialing — for the first program of its kind in the country.

The public’s two seats at that table are occupied by the well-connected, well-funded foundations funded by the Duffield family.

Prognosis

And here’s the damn-near-heretical words of a seven-day-a-week clinician: veterinary care costs more than it used to because it’s better than it used to be.

In 1987, when a miniature schnauzer named Maddie was comforting a stressed software entrepreneur, routine veterinary pain management remained rudimentary. In a veterinary teaching-hospital population treated between 1983 and 1989, only 40% of dogs received any postoperative analgesia, fewer than one in five received it for more than eight hours, and nearly half of the dogs believed to be in moderate or severe pain received none. During approximately the same era, anesthetic-related deaths were reported in one of every 870 healthy ASA I–II dogs; a major study conducted in the early 2000s would place that risk at approximately one in 1,849. Parvovirus, meanwhile, could carry mortality approaching 90% when treatment was unavailable, even though vaccination and intensive supportive care had begun to make survival possible.

By 2005, when Oracle wrote its ten-billion-dollar check, board-certified specialists were still exotic outside university towns. By 2012, Workday’s IPO year, they were not.

And by 2025, a dog like Maddie could receive chemotherapy protocols, advanced imaging, ICU ventilation, and dentistry that would have been science fiction across most of her actual lifetime.

Much of it, credit where due, advanced by shelter medicine research the Duffields themselves funded.

That is what the 8.9% is buying. When household veterinary spending rises from $514 to $560 in a year, the comfortable explanations are greed. Greedy veterinarians, greedy consolidators, sentimental owners exploited at their weakest. And whatever share of the truth those explanations hold, the larger truth is less comfortable and more hopeful: the care is simply better, and better often costs more. The rate of improvement in what we can do for an animal has outrun inflation for decades.

Which, I think, reframes the access-to-care crisis in a way the apparatus described above seems constitutionally unable to consider. If care is expensive because it is good, then the two available answers are: find ways to help more people afford good care or manufacture a cheaper tier of practitioner and call the output care.

The Duffield network’s Colorado experiment chose the second answer. The Veterinary Professional Associate is, by design, a person with less training, less education, and less clinical depth than a veterinarian, authorized to diagnose and cut under a “supervision” whose definition is being written right now by a board whose public-member seats you have already toured.

I also want to not make one of the most admirable philanthropists in the world out to be an evil genius or some kind of supervillain. I believe David Duffield to be a man trying to do what he believes to be good in the world.

There is no evidence that the Duffields directed any of the inside baseball in Colorado. There’s no memo from Incline Village, no instruction from Pleasanton, no phone call anyone has produced. I don't claim it and I don't believe it's needed. That is, in a way, the whole problem. Nine hundred million dollars doesn't require anyone's bad intent to bend a small profession's regulatory machinery; it only requires momentum and organization. Money at that scale changes the climate of the entire profession. Grantees anticipate, institutions align, and boards fill with the people the ecosystem has already made prominent, because those are the names anyone can find.

And if you want proof that this dynamic operates without villains, look no further than the Duffields' greatest triumph. The no-kill movement saved millions of animals. And, along the way, its signature metric, the live-release rate, taught a generation of shelters to manage their statistics as attentively as their animals: restricting intake so the denominators stayed clean, warehousing dogs in kennels for months because an outcome delayed is not an outcome counted, keeping the number sacred even when the metrics and the welfare of the animal in run 14 had quietly diverged.

Maddie's Fund itself has spent recent years recasting its mission away from the very framework it built. A tacit acknowledgment, from the people who would know best, that a well-funded goal can outrun its own purpose. None of that required bad intent either. It required only money, metrics, momentum, and an imperfect system.

The question this record raises isn't whether anyone conspired. It's whether anyone on the shore of Lake Tahoe has the faintest idea what their money is doing in Denver beyond making veterinary care less expensive. And whether a promise made to a schnauzer was ever supposed to include rewriting another state's laws and regulations.

And here is where I stop being able to write around the schnauzer.

Maddie was the beloved dog of a family that was, at the time, not yet rich. She is the founding story of this entire philanthropic empire. Maddie is the promise on the living room rug, the lighthouse in the storm.

So ask the question the empire never seems to: what would have become of Maddie under the system now being built in her name? If the affordable option available to a leveraged, uncertain young family had been not a veterinarian but the discount tier of medicine provided by someone who never attended veterinary school, operating under whatever “supervision” a captured rulebook eventually permits?

The Duffields’ promise was that other families would have what they had, and what Maddie Duffield had was a veterinarian.

That is the future this money is purchasing, a thousand miles from home, with the best of intentions: real medicine for families like the Duffields became, and something less for families like the Duffields used to be. There’s a word for building a two-tier system and marketing the lower tier as equity, and the word is not access.

The haves get veterinarians. The have-nots get what the rulebook allows. And the animals, who never had a vote in Colorado or anywhere else, inherit the difference.

Mr. Duffield and I have something in common: we made a promise to help animals. He to Maddie and I to all of them.

And how we navigate the next months and years will answer whether we keep it, or break it.

I would’ve made this a Letter to the Editor at JAVMA, but I’m outside the response window for such a letter. Having doubled children and hospitals in the time since the paper was published online, I’m a bit behind in my journal reading. Still, the study rankled. Instead of bothering Dr. Fortier with this one, I’m sharing with my Substack audience.

This is one of those statistically pedantic little rants to which my wife is occasionally subjected. On the upside, this one is short.1

Drs. Durocher2 et al3 take on a genuinely unsettled question, and their documentation of supportive-care and antimicrobial practice across 137 pyometra cases is a useful contribution. The study’s central claim, the one carried by both its title and its clinical-relevance statement that immediate surgery “may not be superior to delayed surgical intervention”, is not supported by the reported data and, in at least two respects, is contradicted by them.

The stated primary outcome was survival to discharge. One of 137 dogs died, there was an overall mortality of 0.73% (exact 95% CI, 0.02%–4.00%), and the single death fell in the immediate-surgery arm, while the delayed arm recorded none.

A zero numerator is weak evidence of safety: the rule of three shows that observing no deaths remains compatible with a nonnegligible true event rate.4 No survival comparison between timing groups is reported, and with a single event none would be informative; for an evenly divided sample, the two-sided Fisher exact P is approximately 0.4.

Power for a binary outcome is governed by the number of events, not the sample size. Detecting even a large absolute reduction, for example from 5% to 1%, the upper figure consistent with reported ranges for surgically managed canine pyometra5 would require roughly 570 dogs, more than four times the enrolled sample. And detecting more modest, still-clinically-meaningful differences would require well over a thousand. “No definitive advantage” therefore describes the study’s statistical power, not any property of the treatment. Which is nonsignificant (read: unable to be computed) difference is not evidence of equivalence.6

The interpretation is further compromised by worryingly nonrandom treatment allocation. The authors report that every dog with septic peritonitis, and the sole mortality, underwent immediate surgery. The sickest pyometra patients were eliminated at the outset! The immediate group was thus systemically and systematically sicker (a textbook instance of confounding by indication7)and comparable outcomes despite greater baseline severity, if anything, argue for early source control rather than against it. The direction of any timing effect on survival cannot be inferred from these data.

Finally, the conclusion that immediate surgery “may not be superior” is a noninferiority claim, and establishing noninferiority requires a prespecified margin and adequate power to exclude a clinically meaningful difference8, neither of which is provided by the paper’s observational analysis. The title’s scope is also inaccurate: feline patients were excluded by design, so the findings pertain to dogs alone, not “small animals.”9

Jokes and statistical pedantry aside, there’s value to this paper. And none of what I’ve written diminishes the value of the descriptive work or the importance of the question.

But the data support only a narrower, hypothesis-generating statement: in this cohort no survival difference was detected, in a sample far too small and too confounded to detect one. Presented as evidence that emergent surgery “may not be imperative,” the conclusion risks being read as license to delay intervention in a disease that can still deteriorate rapidly. A prospective, adequately powered, ideally noninferiority-designed canine study is needed before that inference is drawn.

So what’s this all mean?

Less than the title would indicate. The fact is that this study doesn’t move the timing decision in either direction, and that is the point.

It didn’t show that operating immediately helps and it equally didn’t show that waiting is safe. It showed that 137 dogs and a single death can’t tell you much about survival at all. The decision lands back where it always was: with the patient, not the abstract.

And the patient still matters in the old, unglamorous, pathophysiologic ways. There’s a difference between stabilizing and delaying, and the paper muddles the two. Spending a few hours correcting hypovolemia, electrolytes, and acid-base derangements to turn a septic dog into a better anesthetic candidate is good medicine. Reading “no definitive advantage” as license to slot an unstable pyometra into tomorrow’s elective lineup is a dramatically different thing and the study offers absolutely no cover for it.

The distinction bites hardest exactly where the study is quietest: the authors sent every septic-peritonitis case, and their only death, to immediate surgery.

That isn’t evidence that delay is safe; if anything, it’s a faint reminder that source control is what keeps the sickest patients alive. That tells me that their doctors have excellent judgment in deciding which pyometra patients to cut immediately and which ones to sleep on.

The stable, open-cervix dog can probably wait for a proper resuscitation and a daytime table. The closed-cervix dog with a climbing lactate and a flagging blood pressure is the patient the “surgical emergency” framing was built to define and address.

If there’s a portable lesson, it isn’t about uteri. It’s that “no difference” and “no difference we could detect” are very much not the same thing. The space between them is where a the wrong kind of clinical folklore gets conceived. When a paper reports a null, the first question shouldn’t be “what did they find?” but rather “could they have found it if it were there?”

Here, with one event, the answer was no. So measure the creatinine (in mg/dL), stabilize the patient, and make the call about the dog in front of you.

Subscribe now

Recently I had an opportunity to take a leadership evaluation. I follow a number of leadership groups and one of them considers me a kind of “friend of the firm” and asked me to be one of their guinea pigs for their evaluation of a test in development.

When the scores came back, well, I hadn’t scored as well as I would’ve liked. Resisting (however narrowly) the urge to be sour about it, I gave it some thought instead. And I realized that I had zero frame of reference for this test.1 Being a clinician, I decided to learn more about the diagnostic itself. Being kind of intense about every most many some things, I learned a lot.

I learned that the methodologists inside the leadership-assessment field have been saying for years that the practice has drifted from its psychometric foundations.

I learned that the empirical structure of 360-degree ratings is dominated by rater variance, rather than ratee variance. Which means that a 360-score is, statistically, more a portrait of the people doing the rating than it is of the person being rated.

I learned that the affective machinery of negativity bias is real, well-documented, and exactly the kind of asymmetry for which the rating instruments do not compensate.

I learned that the leader-categorization literature predicts that raters carry different prototypes into different contexts. What looks like a single rating of a single person is often two or three different ratings averaged together without anyone noticing.

I learned that surgeons, pilots, anaesthetists, and the U.S. Army have all, independently, built leadership instruments that rate behavior inside the operational context rather than in aggregate across all contexts.

I learned that surgeons, pilots, and the U.S. Army have all, independently, built leadership instruments that rate behavior inside the operational context rather than in aggregate across all contexts.

It’s all struck me as pretty interesting, so I thought I’d share it.

Subscribe now

What the Diagnostic Is Claiming to Measure

The assessment I took is part of a class of instruments sometimes called valence-based 360s. “Valence” in this context means that the ratings have a built-in evaluative direction, the tools ask raters across the leader’s working life to score the leader along a continuum from, for example, “creative” or “developmentally advanced” behavior on one end to “reactive” or “developmentally constrained” behavior on the other.

The instrument aggregates ratings from supervisors, peers, and direct reports into a single score profile, then maps that profile against an underlying theory of adult development. To continue the above example, higher creative scores are interpreted as evidence of maturity whereas higher reactive scores are interpreted as evidence of incomplete development. The thinking being that maturity is a positive for leaders and incomplete development is a negative. These tools and the reports they’re used to produced are marketed and used as developmental tools, but they also get deployed for selection, succession planning, and other consequential decisions that affect careers.

It’s worth noting that the instrument is doing something coherent, and I’m not suggesting it’s nonsense. The questions are crafted with care, the underlying developmental theory is intellectually serious, and the people who built the instruments are not charlatans. The question I started asking isn’t whether the instrument measures something. It’s whether what it measures is what it claims to be measuring — which is, in clinical terms, whether the diagnostic has the construct validity to support the interpretation being placed on its results.

That distinction matters because the methodologists inside the field already know there’s a problem. Bracken and Church,2 writing a focal review in Industrial and Organizational Psychology, document what they call the “devolution” of 360-degree feedback: three decades of deployment in which the practice has drifted away from its psychometric foundations even as it’s been more widely adopted.

Fischer, Tian, Lee, and Hughes, in a “fundamental rethink” of the abusive-supervision literature published in The Leadership Quarterly, argue that an entire subfield of leadership assessment has been confusing what followers perceive about leaders with what leaders do.3 The instruments are deployed at scale anyway, because the demand exists for something to help us understand leadership better. And in fields less scientific than veterinary medicine, which is most of them, the market will sometimes rewards confident assessment over careful assessment.

So we learn that the question isn’t whether the field of leadership is aware (it is), and we ask the next logical question: what’s been done about it?

A Microscope without a Speedometer

I hated my veterinary histology course, but there’s no denying that a microscope is one of the most important instruments in medical history. But hand one to a veterinarian and ask them to assess a patient’s current velocity and they’ll ask if you’re feeling alright. There’s nothing wrong with the microscope, but because the microscope isn’t built to clock meters per second. Sadly, I’ve not had the opportunity to use a microscope with a built-in speedometer, and so there’s nothing about the instrument that can answer that question put to it.

A valence-based 360 is, in this sense, a microscope. It’s built to detect a particular kind of interpersonal pattern, and it does that detection by aggregating rater observations of the leader’s behavior. The raters fill it out by reporting what they have observed of the leader, and what the raters observed is real. The raters aren’t mistaken. The raters (probably) aren’t biased in the sense of having an axe to grind. The raters are reporting their experience of the leader, and the instrument is capturing that experience accurately, in its register.

And here’s the quiet part out loud: what the instrument is measuring is distinct from what it is being claimed about what it measured. If it claims to be measuring the leader’s stable developmental pattern, for example, which is supposed to be the leader’s interpersonal pattern as a property of the leader, then we are asking the microscope to report a velocity.

This is the construct-validity question in its leadership diagnostic clinical form. A diagnostic test has construct validity when the thing it measures is the thing it claims to be measuring. A troponin assay has construct validity for myocardial injury because elevated troponin has been demonstrated to reflect the underlying tissue damage it claims to detect. The confluence of BUN, creatinine, and urine specific gravity measurements has construct validity for renal function because the values it measures correlates with the physiology it’s used to assess. The signal and the interpretation line up and have been empirically demonstrated to do so.

The signal a valence-based 360 produces is real. But in some (read: many) cases, the interpretation being placed on it (i.e. that the score is a property of the leader rather than a property of the encounter between the leader, the raters, and the contexts in which they encountered each other) is the microscope reporting a speedometer reading. The instrument is reporting it confidently, but the instrument doesn’t actually have the capacity to measure what it’s reporting.

A Portrait of the Raters

In 2000, three industrial-organizational psychologists named Scott Scullen, Michael Mount, and Maynard Goff published a paper in the Journal of Applied Psychology with a deceptively boring title: “Understanding the latent structure of job performance ratings.” What they did was straightforward. They took two large samples of managers (2,350 in one, 2,142 in the other) each rated by multiple supervisors, peers, and direct reports on multiple performance dimensions, and they decomposed the variance in those ratings into its sources. How much of the variation in scores is attributable to actual differences between the people being rated? How much is attributable to systematic perspective differences across rater groups? And how much is attributable to the idiosyncrasies of individual raters? things like the rater’s own personality, mood, history with the ratee, internal scale calibration, whatever each unique rater brings uniquely to the rating?

The headline finding has, in the twenty-five years since, become one of the more uncomfortable facts in organizational psychology. In their two samples, idiosyncratic rater effects accounted for 62% and 53% of the variance in ratings. Actual ratee performance, the thing the instrument is supposed to be measuring, accounted for only 21% and 25%.

I want to be exceedingly clear about this, because it’s crucial and also because it makes my eye-twitch: In a 360-degree rating, the largest single source of variance is not the person being rated, it’s the rater.4 By a factor of roughly two and a half to three.

Hoffman, Lance, Bynum, and Gentry later complicated this picture in a useful direction: systematic perspective differences across rater groups (bosses see things differently from peers, who see things differently from subordinates) are real and meaningful, and cannot be dismissed as merely statistical noise. The two findings, irritatingly, coexist: perspective matters and the dominant source of variance is still the individual rater. Both can be true and are, so far as we can tell.

Take that finding into a clinical frame and the diagnostic implications are stark. If we ran a test of liver values where 53% to 62% of the result was attributable to the person who drew the blood or the person who restrained the patient, rather than the patient? Not the technique they used, but how they felt doing it? I would come to and hold the conclusions of that diagnostic very differently. I might use it to measure technicians instead of liver values. I might keep it around for some other purpose, in a specific and tightly defined context. But I would not act like it was telling us only (or even mostly) about the patient.

Subscribe now

Asymmetry and Blind Spots

So what is the rater bringing?

In 2001, Dr. Roy Baumeister and three colleagues published a sweeping review in Review of General Psychology under one of those satisfyingly pithy paper titles that needs little explanation: Bad Is Stronger Than Good.

You already get it. Across everyday events, major life events, close relationships, first impressions, learning processes, and social judgments, the authors documented the same asymmetry: negative experiences register more strongly than positive ones, negative memories persist longer, and negative impressions form faster and resist disconfirmation harder. All in all? Negative information dominates positive information of equivalent weight, sometimes by ratios approaching five to one, when the two are combined in judgment.

It ain’t subtle and it’s one of the most replicated findings in social psychology. Dr. Baumeister himself revisited the work two decades later and concluded, dryly, that bad is still very much stronger than good. The mechanism would seem to be evolutionary: organisms that paid disproportionate attention to threats outlived organisms that didn’t, but whatever the deep cause, the empirical fact is settled. The negative carries more weight in human perception than the positive does.

Now consider what this means for an aggregate rating instrument. A subordinate rating a leader on a developmental 360 is asked to integrate, across some period of time, the totality of their observations of that leader.

The instrument treats the integration as if it were neutral, as if the rater’s memory averaged positive and negative interactions with equal weight and utmost rationality before producing a score. But the rater’s memory does no such thing.

The negative interactions are encoded much more deeply, retrieved more readily, and weighted more heavily in the synthesis. A single sharp directive in a high-stakes moment will sit in memory longer and louder than ten patient explanations on quieter days. The instrument receives and processes the rater’s synthesis and reports it as if it were a balanced summary, but we know, empirically, that it isn’t.

This is not a critique of the rater. This is a human rater is doing what human memory and human judgment do. But it is a critique of an instrument that treats human memory and human judgment as if they were balanced when forty years of research, Rozin and Royzman’s “negativity bias, negativity dominance, and contagion” framework formalized the four sub-mechanisms, say plainly that they are not.

So now we know that a 360 instrument is not measuring the individual’s behavior or performance, rather it is measuring the raters’ negativity-weighted recollection and experience of the individual’s behavior and performance. These are not necessarily the same things, and no amount of careful question construction at the front of the instrument can fix what’s happening at the back of the raters’ minds.

The raters’ experience of a colleague, superior, or subordinate can have value in circumstances where we are evaluating someone based on or in terms of, say, the raters’ experience of a colleague, superior, or subordinate. But if we want to evaluate a person’s performance in terms of, say, their performance? We still have a problem.

Subscribe now

How Other Fields Solved It

There’s one more layer of the rater problem worth naming (and I promise I’ll stop here) before getting to what’s actually been done about it.

The implicit-leadership-theory literature, going back to Lord, Foti, and De Vader in 1984, establishes that every rater carries their own “cognitive prototype” of what a leader is supposed to be, and that the prototype activated in any given moment depends on what context the rater is mentally inhabiting at the time.

A rater thinking of the operating room activates a different prototype than the same rater thinking of the conference room. Same leader, same rater, two different implicit standards. And I haven’t found a 360 instrument that has a way of discerning which prototype was active when the rater filled out each question on the form, or one with a way of parsing ratings made against one prototype from ratings made against another. All of them accumulate and more or less average the findings and report the result.

And hat’s the diagnostic problem in full: rater variance dominates, negativity bias loads the variance asymmetrically, and the prototypes underneath the ratings are shifting from context to context without the instrument’s awareness.

The question I wanted answered was a simple one: Has anyone solved this? Because in clinical medicine, when we discover that a diagnostic test has this kind of measurement instability, we don’t keep using it the same way. We bound its claims, restrict its use case, and sometimes go looking for a better instrument.

It turns out that several professions, seemingly independently, working from inside their own fields’ safety cultures, have arrived at the same answer: bounded context.

Aviation seems to have arrived there first. In the wake of decades of accidents in which steep cockpit authority gradients prevented junior crew from challenging captains emerged the instrument called NOTECHS, Non-Technical Skills for pilots, and it rates crew leadership behavior inside the cockpit, not in the pilots’ lives generally. A strictly bounded frame by design.

Surgery followed. The Scottish surgical-safety community, led by Rhona Flin and Steven Yule, developed NOTSS, Non-Technical Skills for Surgeons, explicitly modeled on the NOTECHS approach. As NOTECHS evaluated things within the cockpit, NOTSS rates surgical leadership intraoperatively. Not surgeon-as-person, not surgeon-as-departmental-colleague, not surgeon-as-mentor, not surgeon-as-charmer; just Surgeon-in-the-OR. Henrickson Parker, Flin, McKinley, and Yule later built the Surgeons’ Leadership Inventory along the same lines: a taxonomy and rating system specifically for intraoperative leadership behavior.

The U.S. Army has the Multi-Source Assessment and Feedback (MSAF) program for officer development, and the program’s design choices are revealing: MSAF results are not shared with supervisors and not used for evaluative or selection decisions. After decades of work on multisource feedback for officers, the military’s institutional position, as evidenced by current practice,5 is that the data is not robust enough to support consequential decisions, only developmental conversations. Even that limitation, the Army deemed such assessments valuable, and built the Commander 360 variant for battalion and brigade commanders, though again context-bounded by command level.

Three fields, aviation, medicine, and warfare, with radically different instrument-design traditions, across different decades of development work, and the same answer: bound the context. Feels more like more than mere coincidence to me. Feels like convergence on a methodology that worked.

The conclusions were simple. Don’t aggregate behavior across the operating room and the holiday party. Don’t aggregate behavior across the cockpit and the captains’ lounge. Don’t aggregate behavior across deployment and garrison. And, this is the part that matters most, don’t treat interpersonal-perception ratings as a reliable evaluation of key performance in the first place.6

The high-stakes, high-responsibility fields figured out something the general leadership-assessment industry has not yet metabolized and manifested: when the stakes are high, and you’re evaluating the performance of the responsible individual: you bound your instrument. You restrict what it claims and you don’t make conclusions on things it can’t actually measure.

What This Means for Veterinary Medicine

I’m very much aware that this could sound like a long-winded defensive justification from a guy who didn’t do as well on an evaluation with professional stakes slightly higher than an online quiz about “Can we predict your favorite color based on how many Star Wars characters you can name?”

But it served to highlight an ongoing challenge for me, and I think for others, in veterinary medicine. And I’ll admit without hesitation that I’ve been considered a poor performer in jobs where I’ve performed quantifiably and technically well and people didn’t like me.7

Veterinarians and technicians are often asked to be both affable, charismatic, well-liked leaders and demonstrate exceptional technical competence. It’s difficult to go from being in the strict and well-defined hierarchy of the operating room — where the work is time-sensitive, the cost of error is measured in life, and a directive register is utterly appropriate to the context — and shift, sometimes within the hour, to the collaborative, collegial work of hospital management, where the register experienced by a technician, the register that kept a patient alive at 10 AM, is going to be experienced as off-putting at best and abusive at worst during a lunchtime meeting. And the instruments we use to measure that same leader has no way of distinguishing that affective spillover.

And I think this all matters more than the inside-baseball methodology might suggest.

We are increasingly evaluating veterinarians, like physicians and surgeons and pilots and officers before them, with instruments that aggregate interpersonal perceptions into single scores and then use those scores to make consequential decisions about careers: things like partnership tracks, specialty appointments, corporate and practice leadership, Board service, promotions, and terminations. If the scores are, as the empirical literature consistently indicate, dominated by rater variance and asymmetrically loaded by negativity bias and built on top of shifting implicit prototypes the rater isn’t aware of activating, then we are making meaningful consequential decisions on the basis of diagnostic tools that cannot reliably support those decisions.

The fix, I think, is not to abandon multisource feedback. The high-reliability fields didn’t abandon it; they bounded it. They restricted what it claimed and built tools that rated what they could actually rate and told folks exactly what it meant. Namely that behavior in a defined context, observed through a defined frame, against a defined standard, is what’s being measured. They distinguished technical performance from interpersonal perception, and they declined to let perception ratings drive the technical evaluations. They were, in the most literal sense, careful diagnosticians of their own diagnostics.

We can absolutely do the same. We can distinguish, in our own field, between clinical and technical performance. We can distinguish that which can be measured against outcomes, against medical-record audits, against complication rates and case management quality from interpersonal perception, that which can be measured but which should not be allowed to overdetermine the evaluation of a technically excellent clinician. We can build feedback structures that are bounded by context: the OR is its own context, the exam room is another, the hospital management meeting is a third, and the holiday party is a fourth. A veterinarian’s behavior across different domains can be discussed, but the discussion needs to be honest about which context is being evaluated, by whom, and against what standard.

The most important thing we can do is the thing clinicians have always done with their diagnostics: evaluate the tests before we trust the result. Ask what the instrument is actually measuring, what it can support and what it can’t, ask whether the claim being placed on the score exceeds the measurement’s actual capacity to bear that claim, and ask whether the diagnostic is fit for the purpose to which it’s being put.

A microscope is an excellent instrument, but it isn’t a speedometer. And the cost of mistaking one for the other — especially in clinical medicine, and increasingly in clinical leadership — is paid by the people who get evaluated on the wrong things. By understanding the diagnostics better, we can do better for the people around us.

A little more than three weeks ago, your little sisters were born.

They were delivered in simultaneous states of Washington and prematurity, so they were both far from home and combined to only barely weigh what you did when you were born.

Your mom flew across the country to greet them when they arrived, while I stayed home with you and the animals. The decision was not an easy one, but it was the one available to us. Too many kids, too many animals, and too many hospitals to allow us to both travel across the country for it. While I vividly remember (and have colorfully described) the circumstances of your entry into the world, I will not be able to do that for your sisters.

Your sisters, Brenna and Hawley, have required a bit of extra care since they arrived. Being early, they found themselves ever so slightly unprepared for the world. Not enough brown fat to burn for warmth, not enough energy stored in their muscles to easily sustain themselves, reflexes not yet refined, and even holes in their hearts that have not finished closing. So they’ve stayed in the hospital, in what’s called the neonatal intensive care unit. Your mom, of course, has stayed with them.

For almost three weeks now, she’s spent roughly 20 hours a day in a NICU 3,000 miles from home. Your Aunt Kathy went out to help, and we’re indescribably grateful to her for that.

Your dad opened another veterinary hospital just a few months before your sisters were born. Doubling children and hospitals in a two-month span, and working all of the days of the week since. I doubt you’ll remember any of this by the time you’re old enough to read it. And, frankly, I hope you don’t remember the absence.

The things I’ll remember, and the things I hope you will too, are the good things. I hope you’ll remember the mornings, when I’d come to your room, you shout “Daddy!” You tolerate a hoist and a hug from me, before you eagerly launch into a plan, a list of demands, or an explanation on a topic about which I must be immediately informed. You’ll emphatically insist on playing soccer, watching cars, wearing a certain shirt, going downstairs, going outside, or having a donut, a cookie, an apple, or mac and cheese. Sometimes some of those things, sometimes all of those things, sometimes all at once. You smile big, you demand “uppies,” and then tell me to come play with you or eat with you or watch your favorite show with you or read a book with you.

Sometimes I’ll take you to daycare, sometimes I’ll hand you off to a sitter. Sometimes you’re happy to send me on my way — “Daddy go work!” — and sometimes you hang on to me with the strength of, well, a much larger primate. On days I pick you up from daycare, you’ll ask if we’re going to the park, the playground, or the hospital. It always hurts my heart to leave you, and always delights to see you.

Last night, I was out long past your bedtime performing an emergency surgery on a puppy. This Father’s Day, I am, predictably, working. Today I don’t feel like a man who couldn’t be there for the birth of his daughters, who couldn’t spend time with his son, who couldn’t spend time with his father. Today I only feel like a man who wasn’t, a man who didn’t.

Still, every day, even those when we’re short on time, there is joy to be found. And that is where I will focus.

You woke up with your outrageous enthusiasm, with a fire that lights the world. You nearly succeeded in launching yourself out of my arms to grab the Lightning McQueen shirt you wanted to wear. You were excited for cars and donuts and animals. We video chatted with your mom and sisters. Later today, your Aunt Tess will pick you up and take you to see your grandparents, and while there’s no small regret that I won’t be there, I am heartened to know that you’ll see your grandfather, because if I am a good father to you, Will, it is because he taught me how.

Lately, I’ve been thinking of the days when when your grandfather would take me to practice or train. We did that a lot when I was growing up. Sometimes he wouldn’t tell me how far we had to go or how many more reps or sets we’d do. He’d just say, “one more.”

It wasn’t really soccer games or races we trained for. Days like these, days that can feel long and relentless, are what he was preparing me to meet. When I think of him and those workouts, especially on days like these, I can’t help but smile. Because I don’t think of limits, of fatigue, or of weakness. I think: one more.

You were named, William Jay, for both of your grandfathers, for your father, for your great-grandfather, and your Norman ancestors. Today, and especially today, I’ll remind you not only of the names you carry, but all the delight and joy and love as well.

We will do our best. We will delight in what is, find joy in all the moments we have together, and love every second.

Your mom, your sisters, and I will be home soon.

Love,


Dad

There was no patient waiting on the answer, there was no product, there was a question from a guy who didn’t know something. The heat-stable organism, Thermus aquaticus, he identified from that microbial mat led to the discovery of Taq polymerase and it’s the reason a COVID test works, the reason a crime lab can match a DNA sample, the reason we can read a human genome at all.

Thomas Brock’s grant, had you read the proposal at the time, would have looked like a man using tax dollars to splash around in pond scum.

An uncomfortable and frequently inconvenient pattern is found throughout the history of scientific inquiry: the research that turned out to matter most almost always appeared , at the moment someone had to fund it, to be a poor bet, a ridiculous proposal, or, at best, an eclectic notion, often delivered by someone possessing the tenuous, tenacious sanity of the slightly mad. Research often came from the wrong places, studied the wrong organisms, embarrassed the wrong people, and took far longer than anyone in charge wanted to wait. Suffice to say that Ayn Rand would not approve.

On May 29, the Office of Management and Budget published a sweeping 412-page proposed rule, formally known as the Regulation for Federal Financial Assistance, that rewrites the government-wide rules governing how nearly every federal research dollar is awarded and spent (2 CFR Part 200). The comment period closes July 13. Most of the coverage has framed it as a fight about politics and I, a registered Independent, think it’s much more important than the color of my baseball cap.1

I want to do something narrower and, I think, much nearer to my narrow realm of expertise: walk through the specific provisions that touch science, explain in plain terms what each one does, and show the kind of discovery each one would have kept out of the world.

No political ideology here, just mechanism and consequence from a person who has spent the majority of his waking hours in the study, practice, and application of science.

Subscribe now

The Result is the Point (§200.461)

Start with the least-discussed provision, because it may be the most quietly destructive. Under the proposed §200.461, publication costs like page charges, open-access fees, the “article processing charges” that journals levy to make a paper free to read, become unallowable under a federal award. They can be paid only if a statute specifically requires it, or if the agency signs off on a particular paper, case by case.2

That matters because for a huge share of federally funded research, the paper is the deliverable. A clinical trial that finds a cheap generic drug works as well as an expensive one has produced exactly one tangible thing of value: the published result that lets a veterinarian in Ohio act on it. Knowledge that doesn’t get disseminated didn’t, in any practical sense, get produced.

Open-access fees at major journals now run from roughly $1,000 to north of $12,000 per article. Tell a researcher their grant won’t cover that, and one of three things happens: they publish in a less visible venue, they publish less, or the finding sits in a drawer. The provision contains a small, worrying line: that a general requirement to make results public “must not be construed as authorizing publication costs.” The government may compel you to share what you found and decline to pay for the act of sharing it.

Why does this matter to me? Because I have successfully treated patients based on peer-reviewed research with an n=6 for a disease that I, as a six-day-a-week-clinician, has seen but once in more than a decade. Because of that open access paper, that’s never so much as been cited elsewhere, I was able to save a patient’s life.

The people who feel this first are the ones with the least cushion: a scientist studying a rare disease whose entire patient community lives behind the paywall that an open-access fee would have removed. Less getting published means that less reaches the clinic. Cures can exists and no one can read about it. I would have one less patient.

Someone has to be Impressed (§200.205)

The proposed rule installs a new step before any discretionary grant goes out the door. It calls this “pre-issuance review” (§200.205): a “senior appointee” must personally review and approve awards, weigh whether they “advance the President’s policy priorities” and meet a standard the rule names “Gold Standard Science,” and, this is the eyebrow-raising clause, must “not ministerially ratify or routinely defer to” the recommendations of the scientists who reviewed the proposal. Peer review is explicitly demoted to “advisory.”

I’m a snob about expertise. I routinely remind myself (and anyone in earshot) that I’m an extraordinarily well-educated expert in a VERY limited sliver of knowledge. I spent a lot of my day relying on two decades of education and experience that makes my pattern recognition appear almost prescient. Meanwhile, I couldn’t tell you the difference between a carburetor and a transmission if you threw them at me.

And that’s what expertise is for! We go to extreme lengths to certify and justify and regulate the application of expert knowledge. Not just anybody can cut your hair, spay your dog, register your car, or check your blood pressure. The reason we let scientists evaluate other scientists’ proposals is that you frequently cannot tell, from the outside, the difference between a crank and a pioneer. To a lay person? They can look identical until the data comes in.

In the 1990s an endocrinologist named John Eng went looking for useful compounds in the venom of the Gila monster, a desert lizard with a name from a fantasy world’s bestiary. The work was hard to fund because, it sounds, in no small way, like a waste of resources. The compounds he found launched the entire class of GLP-1 drugs that now includes Ozempic and Wegovy, currently rewriting the treatment of diabetes and obesity.

Nobel Laureate3 Katalin Karikó spent two decades on messenger RNA while grant after grant was rejected because reviewers couldn’t see the application. She was demoted for it. Her work was so specific that even other experts struggled to see its value. Her work also became the basis for COVID mRNA vaccines.

A subject-matter expert can sometimes see the promise in lizard venom. A political appointee, who is more likely to be a lawyer or administrator than a scientist, is far less likely to do so. And the rule specifically instructs that appointee not to rely on the other experts who might. We do not get the Gila monster drugs from a reviewer optimizing for proposals that look obviously sensible today.

Science is a Conversation (§§200.432, 200.454)

Two smaller provisions, taken together, reach something foundational.

Under the proposed §200.432, the cost of attending a scientific conference is allowable only if the agency has “expressly approved” it and written it into the award’s terms. Under §200.454, a researcher can no longer use grant funds to subscribe to the journals in their own field. Those costs are made flatly unallowable.

When you strip away the accounting language you see what’s being rationed.

Conferences are where results are presented months before they’re published, where a graduate student gets the comment that reshapes a project, where two labs realize they’re each holding half of the same answer, where the connections between scientists and researchers are made.

In 1975 the scientists working on recombinant DNA gathered at Asilomar and wrote their own safety guidelines. The field governed itself, in a hotel conference room. And journals are simply how a scientist knows what is already known; you cannot meaningfully add to a literature you are not permitted to read.

What both provisions hand the funding agency is the power to decide in which conversations a funded scientist may participate and which ones are even allowed to happen.

This is no longer about whether the science is good, but whether the gathering is sanctioned. In the same way that decisions are made by the people who show up, science advances by argument among the people allowed in the room. These provisions decide who is allowed in the room and which rooms are allowed to exist.

Being Wrong is the Job (§200.206)

The proposed §200.206 expands the list of factors an agency may use to judge the “risk” an applicant poses. Among the new ones: a record of “discredited or non-replicable studies.”

Sit with that as a screening criterion, because it inverts how discovery actually works. Most hypotheses are wrong. Science isn’t about “getting an answer,” science is a way of “asking a question.”

Many studies that turned out to be important were, for a while, contested, doubted, or flatly disbelieved by the consensus of the field. A scientist who has published something later overturned is not a bad actor; that is the ordinary shape of a working career at the edge of the known.

Stanley Prusiner proposed that a protein with no genetic material could transmit disease. Not a toxin, not a living or non-living organism. A protein.

This was an idea the field of biology considered tantamount to heresy for years.

It explains mad cow disease and Creutzfeldt-Jakob. And it won Dr. Prusiner the Nobel Prize in 1997.

Barry Marshall insisted that bacteria, not stress, caused stomach ulcers, was dismissed by the medical establishment, and finally drank a flask of the bug himself to prove it. He was awarded the Nobel Prize in 2005.

Both men had, by the standard of their respective moment, a documented history of “discredited” ideas.

Screen out the people who have ever been wrong in an interesting way, and you screen out precisely the people willing to take the risks that move the world.4

Knowledge Keeps its Own Schedule (§§200.202, 200.340)

A phrase recurs throughout the proposed rule: the “national interest.”

Awards must serve it, foreign collaborations are restricted unless they advance it, and an agency may terminate a grant already underway if the work no longer effectuates “Federal agency priorities, or the national interest as they exist at the time of the termination“ (§200.340).

The trouble is not the phrase, rather, it is that the phrase is undefined and, by the rule’s own wording, movable. It’s pegged to whatever the priorities happen to be on the day someone decides to pull the plug. Science runs on a longer timeline than that; Dr. Brock’s pond scum took twenty years to become PCR.

History has a number of clear, non-partisan verdicts on what happens when a government makes its own current sense of the national interest the test of acceptable science.

In the Soviet Union, Trofim Lysenko’s rejection of genetics (because Mendelian inheritance was deemed incompatible with state ideology) became official policy. Real geneticists were silenced or purged. Scientists fled the country in the dead of night. While the Soviet Union made advances in other fields, they trailed in genetics. The result was failed harvests, diminished food production, and a national biology set back generations, all in the name of the “national interest.”

And the deeper problem is, to me, that we cannot know what will serve the national interest. Science built to solve for the national interests. It’s built to solve for advancing knowledge, which is an inherent service of the national interest.

Take the CRISPR gene-editing revolution, which began as pure investigation as to how bacteria fight off viruses. A question borne of curiosity with no obvious payoff whatsoever. A “national interest” filter applied at the front end would not have funded “how does a pond bacterium defend itself,” and we would not have gene editing.

Some Things Can Only Be Studied Where They Are (§200.202)

The proposed §200.202(e) builds in a “domestic-first” framework for research-and-development awards: funding flows to U.S.-organized entities, foreign recipients are barred except in narrow cases, and any “international element” in a project must be affirmatively justified as serving the national interest.

This collides with a basic fact about how knowledge is made: a great deal of the most important science cannot be done on American soil, because the thing being studied isn’t on American soil.

You study a disease where the disease is.

The vaccine that now exists for Ebola was tested during an active outbreak in West Africa, because that is the only place Ebola transmission can be observed; you cannot (and should not) run that trial in Kansas.

The single most important finding in HIV prevention, that putting a patient on treatment cuts transmission by roughly 96 percent, came out of a trial conducted across several countries, and it reshaped HIV care everywhere, including the United States.

You study a population where the population is: Iceland’s unusually complete genetic records produced the discovery of a gene variant that protects against Alzheimer’s.

You study high-altitude adaptation in the people who live at high altitude.

And the best instruments are not always domestic. The facility in Switzerland where physicists found the Higgs boson is also, not altogether coincidentally, the place where a researcher invented the World Wide Web.

A rule that treats research conducted abroad as suspect by default treats the locations of the world’s diseases, populations, and instruments as an inconvenience. It’s not an inconvenience, it’s the simple reality that answers and solutions are generally found where there are questions and problems.

Inconvenient by Design (§200.332(i))

Of all the provisions, the one I would flag for sheer and terrifying arbitrariness is §200.332(i).

It requires that organizations passing federal funds to subrecipients ensure those subrecipients don’t take actions that could “significantly damage the reputation” of the pass-through, the agency, or the federal government, and if they do, the funding can be pulled.

“Significantly damage the reputation” is left entirely undefined, which is a problem, because producing findings that embarrass powerful people and institutions is not a malfunction of independent research. It is frequently incidental and sometimes the entire point.

Suppose a federally funded study finds that a popular pastime, say, a lifetime of weekend golf, with its cumulative sun exposure, is associated with a meaningfully higher rate of melanoma. That is a genuinely useful finding; it is something every dermatologist would want every golfer to hear.

Now suppose someone influential, somewhere in the chain, decides the finding reflects poorly on a pastime associated with someone powerful. Under a standard this vague, the safe move is to make the problem go away by pulling the funding of such research. And millions of golfers and their doctors never learn about a risk they could have managed with a hat and some sunscreen.

Think of how many fewer people would be afflicted with melanoma. Think of how much less strain that puts on doctors, hospitals, insurance companies, and families. The world just got that much safer, people suffered less, and we all spent way less money treating a disease that could’ve been significantly curtailed.

And that’s not hypothetical at the margins! It’s the recurring shape of public-health history.

Clair Patterson’s geochemistry revealed that leaded gasoline had contaminated essentially the entire planet. That’s a finding fought bitterly by industry and downplayed by government for years before it produced one of the great public-health victories of the century.

In Flint, Michigan, it was research by a local pediatrician and an outside engineer that exposed the lead in the water. And officials’ first instinct was to attack the researchers.

Those findings “damaged the reputation” of someone in power. Every one of them needed to be done for the health and safety of individual Americans. A standard that lets reputational discomfort justify defunding is a standard that risks suppressing exactly the research we most need to keep independent.

The Cure isn’t in the Reagents

Research has two kinds of cost. There are direct costs, things like the reagents, the specific salaries, the equipment for one project. And there are indirect costs: the unglamorous infrastructure that makes any of it possible, things like the buildings, the freezers running around the clock, the biosafety officers, the animal-care staff, the ethics boards, the high-containment labs, the imaging and genomics cores shared across dozens of projects.

These “facilities and administrative” rates are negotiated, and they are considerable. The cure for a disease does not come out of the reagent budget alone, but also out of the whole ecosystem the reagent budget sits inside.

An earlier attempt to slash those rates to a flat 15 percent was blocked by the courts as inconsistent with law. The proposed rule, notably, declines to reopen rate-setting and tells commenters not to raise it.

The back door was shut, but the front door is wide open, buried in the pre-issuance review criteria is an instruction that “preference for discretionary awards should be given to institutions with lower indirect cost rates” (§200.205). Same destination, different route. You don’t have to cap the rates if you simply steer the money toward the places with the least infrastructure.

The catch is that institutions with high indirect rates tend to have excellent reasons for those rates.

Institutions that maintain the expensive machinery the hardest science requires. The lab equipped to safely study a dangerous emerging pathogen, the kind of containment facility that lets researchers characterize a new coronavirus, is more expensive to keep running than one that doesn’t, and that expense lives in the indirect rate.

Prefer the low-overhead institution and you are, with the best of intentions, steering federal research dollars away from the places actually built to make the next major discovery.

The Throughline

I’ve deliberately kept away from political ideology, because the deepest issue here is not partisan. Vaccines don’t work better if you’re a Democrat or a Republican, antibiotics don’t care who you voted for, and science doesn’t care if the legislator that signed a funding bill knows what DNA stands for.

None of this is to pretend the people who wrote the rule are fools, or that every concern behind it was invented or imagined.

The reproducibility crisis is real; a great deal of published work does not hold up, and a funding system that never asked whether its findings replicated would deserve suspicion. Collaboration with adversarial governments carries genuine risk and it’s why the Wolf Amendment exists, and serious people defend it for serious reasons. Indirect-cost rates are frequently opaque, and "trust us" has never been a legitimate budget justification. The instinct underneath these provisions, that money taken from the public ought to answer to the public, is not a purely politically motivated vice. Rather, I think, it’s a virtue pointed in the wrong direction.

And I can say that without contradiction, because for all my protestations of neutrality, I have been defending an ideology this entire essay. Not a political one, because it has no party, and it does not care who you voted for. But an ideology all the same: the conviction that the way to challenge the world is through disciplined thinking, careful evaluation, and patient study, and that any claim worth making is one made in a form that others can review, test, and overturn. That is not the opposite of accountability. It is simply among the oldest and most ruthless forms of accountability we’ve built.

Science does not ask to be trusted, it asks to be challenged. A rule that hands the checking to whoever holds power this year rather than to the evidence and to the people trained to read it has not made science more accountable. Instead, it has only replaced the kind of accountability that works.

All eight of these provisions answer a single question: who decides what is worth knowing, and on what basis. And these provisions answer the question in a way that runs against the grain of how discovery actually happens.

The research that changed the most often appeared the least promising at the outset . It looked like pond scum and lizard venom. It came from people who had been wrong before in interesting ways about things like prions and ulcers. It was done in the wrong places like West Africa, Iceland, and Switzerland. It embarrassed the powerful like leaded gasoline and Flint’s water. And frequently takes longer than any administration’s attention span or term limit, as in the case of the two decades between a hot spring to PCR. It depended on infrastructure that wouldn’t be paid for under the proposed rules.

Those are not incidental features of great science, rather they are the point and purpose and power of science. They are, very often, where the value comes from, and, sadly, they are precisely the features these rules would file down.

A grant system optimized to fund only what looks sensible today from people with unblemished records at well-behaved institutions on questions aligned with this year’s national interest risking nothing that could embarrass anyone is a system optimized to miss the next Taq polymerase, GLP-1, COVID vaccine, and antibiotic therapy entirely.

The comment period on the rule closes July 13.

Whatever one thinks of the politics, that is the practical window in which any of this can still be argued on the record and the question worth putting on the record is a simple one. Eighty years of American discovery happened under a system willing to bet on the strange, the unproven, the inconvenient, and the slow. The next pond-scum enzyme and the next lizard-venom drug are out there somewhere, in proposals that look, right now, like a poor bet. The only real question this rule raises is whether we will still fund the bet.

I found this pre-print paper from Schuloff et al (2024), and I used Claude to turn it into a catalogue of prompting techniques. They turned it into a SaaS business.

The utility of my AI writing, I’ve long (dis)claimed, is not that I’m inventing the techniques or technology, only that I’m collecting and presenting them to a mostly veterinary audience. What I’ve done here, or at least tried to do, is provide a prompt that you can copy and paste into your preferred large language model (which should probably be Claude2) and it will help guide you through the technique that should work the “best” for accomplishing your goal.

This isn’t something to try for every situation, as the LLMs have gotten so good that you usually get the output you want. This is more for the times when you have a reasonably complex task and you’re struggling to make the software do what you need.

You can also read the paper, or the LLM-produced catalogue, just for the sake of learning more about prompting. It’s a dated paper (by LLM standards), not peer-reviewed, and neither paper nor catalogue are exhaustive, but there was enough good in there for me to think “hey, that’s interesting” or “hm, that’s useful” often, and to use it with sufficient frequency, to decide to publish it on Substack.

The Prompt Report

I’ve linked the article above, and I’m posting the output mostly unedited. I’ve only removed the last paragraph where Claude offered to use this to help with other projects.3

This bit is, by design, a reference dump. More cheat sheet than encyclopedia. You don’t need to — and shouldn’t — read it as prose. It’s meant to be copy and pasted.

This is “The Prompt Report” by Schulhoff et al. (2025) — the most comprehensive systematic survey of prompt engineering techniques to date. It came out of UMD, with contributors from Stanford, OpenAI, Vanderbilt, Princeton, and others. The paper used a PRISMA-based systematic review process, starting from ~4,800 records and distilling down to 1,565 relevant papers, from which they catalogued 58 text-based prompting techniques and 40 multimodal/multilingual techniques.

What makes it genuinely useful rather than just encyclopedic is the taxonomy — they don’t just list techniques, they organize them into functional categories that help you understand when to reach for each one. They also include a real-world prompt engineering case study on suicide crisis detection that’s worth the read on its own, because it shows how messy and non-linear the actual process is.

Let me lay out the full catalogue.

---

TEXT-BASED PROMPTING TECHNIQUES (58)

1. In-Context Learning (ICL)

The foundational category — providing exemplars and/or instructions within the prompt so the model can perform tasks without weight updates.

Few-Shot Prompting Design Decisions: exemplar quantity, ordering, label distribution, label quality, format, and similarity to test instances all matter significantly. On some tasks, exemplar order alone can swing accuracy from below 50% to above 90%.

Few-Shot Techniques:

• KNN — selects exemplars similar to the test input using k-nearest neighbor retrieval

• Vote-K — proposes unlabeled candidates for annotation, then uses the labeled pool for few-shot prompting while ensuring diversity

• Self-Generated ICL (SG-ICL) — uses the LLM itself to generate exemplars when training data is unavailable

• Prompt Mining — discovers optimal prompt template formats by analyzing which patterns appear most frequently in large corpora

Zero-Shot Techniques:

• Role Prompting (aka Persona Prompting) — assigns a specific role to the model (”Act as a travel writer”)

• Style Prompting — specifies desired tone, style, or genre

• Emotion Prompting — incorporates psychologically relevant phrases (”This is important to my career”) to boost performance

• System 2 Attention (S2A) — asks the LLM to rewrite the prompt removing irrelevant information, then answers from the cleaned version

• SimToM — for multi-person/multi-object questions, establishes what facts one person would know, then answers based only on those facts

• Rephrase and Respond (RaR) — instructs the LLM to rephrase and expand the question before answering

• Re-reading (RE2) — adds “Read the question again:” plus a repetition of the question; surprisingly effective on complex reasoning

• Self-Ask — the LLM decides whether it needs follow-up questions, generates them, answers them, then answers the original

2. Thought Generation

Techniques that prompt the model to articulate reasoning steps.

Chain-of-Thought (CoT) — the foundational technique: providing exemplars that include reasoning paths before the final answer.

Zero-Shot CoT variants:

• Zero-Shot CoT — appending “Let’s think step by step” (or similar thought inducers) with no exemplars

• Step-Back Prompting — first asks a high-level question about relevant concepts before diving into reasoning

• Analogical Prompting — auto-generates exemplars that include CoT reasoning; improves math and code generation

• Thread-of-Thought (ThoT) — uses “Walk me through this context in manageable parts step by step, summarizing and analyzing as we go” as the thought inducer; works well with large, complex contexts

• Tabular CoT (Tab-CoT) — makes the LLM output reasoning as a markdown table, improving structure

Few-Shot CoT variants:

• Contrastive CoT — includes both correct and incorrect reasoning exemplars so the model sees how not to reason

• Uncertainty-Routed CoT — samples multiple CoT paths, takes majority if above a confidence threshold, otherwise samples greedily

• Complexity-Based Prompting — selects complex examples for exemplars and uses majority vote among longer reasoning chains

• Active Prompting — asks the LLM to solve training questions, calculates uncertainty, then has humans rewrite the highest-uncertainty exemplars

• Memory-of-Thought — pre-computes CoT on unlabeled training data, then retrieves similar instances at test time

• Auto-CoT — uses Zero-Shot CoT to automatically generate chains-of-thought, then assembles them into Few-Shot CoT prompts

3. Decomposition

Breaking complex problems into simpler sub-problems.

• Least-to-Most — prompts the LLM to break a problem into sub-problems without solving them, then solves sequentially, appending each answer to the prompt

• DECOMP (Decomposed Prompting) — teaches the LLM to use specific functions (string splitting, internet search, etc.) via few-shot examples, then lets it route sub-problems to those functions

• Plan-and-Solve — an improved Zero-Shot CoT: “Let’s first understand the problem and devise a plan to solve it. Then, let’s carry out the plan and solve the problem step by step”

• Tree-of-Thought (ToT) — creates a tree-like search by generating multiple possible “thought” steps, evaluating progress, and deciding which branches to continue

• Recursion-of-Thought — when a complex sub-problem arises mid-reasoning, it sends that sub-problem to a separate LLM call, then inserts the answer back

• Program-of-Thoughts — generates code as reasoning steps, then executes via a code interpreter

• Faithful CoT — combines natural language and symbolic language (e.g., Python) reasoning in a task-dependent fashion

• Skeleton-of-Thought — generates an answer skeleton, then solves sub-parts in parallel for speed

• Metacognitive Prompting — a five-part prompt chain mirroring human metacognition: clarify the question, preliminary judgment, evaluate response, confirm decision, assess confidence

4. Ensembling

Using multiple prompts and aggregating results.

• DENSE (Demonstration Ensembling) — creates multiple few-shot prompts with distinct exemplar subsets, aggregates outputs

• MoRE (Mixture of Reasoning Experts) — uses different specialized prompts for different reasoning types (retrieval for factual, CoT for math, generated knowledge for commonsense), selects best by agreement

• Max Mutual Information — creates varied prompt templates, selects the one maximizing mutual information between prompt and output

• Self-Consistency — prompts the LLM multiple times with non-zero temperature for diverse CoT paths, then takes majority vote

• Universal Self-Consistency — like Self-Consistency but uses an LLM to select the majority answer rather than programmatic counting (useful for free-form text)

• Meta-CoT (Meta-Reasoning over Multiple CoTs) — generates multiple reasoning chains, inserts all into one prompt, generates a final answer

• DiVeRSe — creates multiple prompts, runs Self-Consistency on each, scores reasoning paths step-by-step

• COSP (Consistency-based Self-adaptive Prompting) — runs Zero-Shot CoT with Self-Consistency on examples, selects high-agreement outputs as exemplars for a final prompt

• USP (Universal Self-Adaptive Prompting) — generalizes COSP to all tasks using unlabeled data and a more complex scoring function

• Prompt Paraphrasing — rewording the original prompt while preserving meaning; effectively data augmentation for ensembles

5. Self-Criticism

Having the LLM evaluate and improve its own outputs.

• Self-Calibration — answers a question, then asks the LLM “Is this answer correct?” to gauge confidence

• Self-Refine — iteratively: generate answer → get LLM feedback → improve answer → repeat until stopping condition

• Reversing CoT (RCoT) — asks the LLM to reconstruct the original problem from its answer, then compares for inconsistencies

• Self-Verification — generates multiple CoT solutions, scores each by masking parts of the original question and testing if the LLM can predict them

• Chain-of-Verification (COVE) — generates an answer, creates verification questions, answers them, then produces a revised final answer

• Cumulative Reasoning — generates potential reasoning steps, evaluates each (accept/reject), checks if the final answer is reached, repeats if not

6. The case study technique:

• AutoDiCoT (Automatic Directed CoT) — automatically generates CoT explanations, then combines them with contrastive (incorrect reasoning) exemplars. Developed during the paper’s own suicide crisis detection case study.

---

PROMPT ENGINEERING TECHNIQUES (automated optimization)

• Meta Prompting — prompting an LLM to generate or improve a prompt

• AutoPrompt — uses backpropagation to optimize “trigger tokens” in a prompt template (soft-prompting)

• APE (Automatic Prompt Engineer) — generates multiple Zero-Shot instruction prompts from exemplars, scores them, creates variations of the best

• GrIPS — like APE but uses deletion, addition, swapping, and paraphrasing operations

• ProTeGi — passes outputs through a criticism prompt, generates new prompts from criticisms, uses a bandit algorithm to select

• RLPrompt — uses reinforcement learning (Soft Q-Learning) to optimize prompt templates; often selects grammatically nonsensical but effective text

• DP2O — the most complex: combines RL, custom scoring, and LLM conversations to construct prompts

---

MULTILINGUAL TECHNIQUES (14)

CoT extensions:

• XLT (Cross-Lingual Thought) — six-instruction template including role assignment, cross-lingual thinking, and CoT

• CLSP (Cross-Lingual Self-Consistent Prompting) — constructs reasoning paths in different languages for the same question

ICL extensions:

• X-InSTA — aligns in-context examples with input via semantic alignment, task-based alignment, or both

• In-CLT (Cross-lingual Transfer) — uses both source and target languages for in-context examples

• PARC — retrieves exemplars from a high-resource language for low-resource targets

Translation-specific:

• Translate First Prompting — translates non-English input to English before processing

• MAPS (Multi-Aspect Prompting and Selection) — mines knowledge from source text, generates multiple translations, selects best

• Chain-of-Dictionary (CoD) — extracts words, looks up meanings in multiple languages, prepends to prompt

• DiPMT — similar to CoD but only source and target language definitions

• DecoMT — divides source text into chunks, translates independently, then combines with contextual information

• Interactive-Chain-Prompting — generates sub-questions about translation ambiguities for human resolution

• Iterative Prompting — creates draft translation, then refines with automated retrieval or human feedback

---

MULTIMODAL TECHNIQUES (26+)

Image:

• Prompt Modifiers — appending words to change generated images (medium, lighting, etc.)

• Negative Prompting — numerically weighting terms so the model considers them more/less

• Paired-Image Prompting — shows before/after transformation pairs, then presents a new image

• Image-as-Text Prompting — generates textual descriptions of images for inclusion in text prompts

• Multimodal CoT — extending CoT to image inputs (”Solve this step by step” with a math problem image)

• DDCoT (Duty Distinct CoT) — extends Least-to-Most to multimodal, creating and solving subquestions

• Multimodal Graph-of-Thought — extends GoT to multimodal with image captioning for visual context

• Chain-of-Images — generates images (as SVGs) as part of the reasoning process (”Let’s think image by image”)

Audio, Video, Segmentation, 3D: The paper notes these are early-stage but growing, with techniques for text-to-video generation, video editing, semantic segmentation, 3D object synthesis, surface texturing, and 4D scene generation.

---

AGENT TECHNIQUES

Tool Use:

• MRKL System — LLM router with access to multiple tools (weather, date, calculator)

• CRITIC — generates response, self-criticizes, then uses tools to verify/amend

Code Generation:

• PAL — translates problems directly to code, executes via Python interpreter

• ToRA — interleaves code and reasoning steps iteratively

• TaskWeaver — transforms requests to code with user-defined plugins

Observation-Based:

• ReAct — think → act → observe → repeat, with all history in the prompt

• Reflexion — adds introspection layer to ReAct: evaluates success/failure, generates reflections for working memory

Lifelong Learning:

• Voyager — proposes tasks, generates code, saves actions to long-term memory (Minecraft)

• GITM (Ghost in the Minecraft) — recursive goal decomposition with structured text actions and external knowledge base

RAG:

• Verify-and-Edit — generates multiple CoTs, retrieves external info to edit selected ones

• Demonstrate-Search-Predict — decomposes questions, uses retrieval for sub-questions, combines

• IRCoT — interleaves CoT and retrieval, each guiding the other

• Iterative Retrieval Augmentation (FLARE, IRP) — generates temporary sentences as search queries during long-form generation

---

KEY FINDINGS FROM THEIR BENCHMARKING

They tested techniques on MMLU with GPT-3.5-turbo. Few-Shot CoT performed best at 0.692 accuracy. Surprisingly, Zero-Shot CoT dropped below plain Zero-Shot (0.547 vs 0.627). Self-Consistency helped Zero-Shot CoT but didn’t improve Few-Shot CoT. The spread between best and worst phrasing variants within a single technique was large, reinforcing how sensitive these models are to prompt wording.

The suicide crisis detection case study is particularly instructive — 47 development steps, ~20 hours, going from 0% to 0.53 F1. The most interesting finding: accidentally pasting an email twice improved performance, and removing the duplicate made it worse. Anonymizing names in the email also hurt. This underscores their point that prompt engineering remains partly a “black art” where inexplicable details can matter enormously.

I’ve included this because I think sometimes you might want to reference the paper specifically, and sometimes you might only need a catalogue refresher to help you prompt it.

Bluntly, if you get the hang of five to eight of these techniques, you are probably now LLM-capable enough to do pretty much anything you need. If you’ve got few-show prompting, chain-of-thought, decomposition, self-refinement, and role/persona prompting in your toolbox? You almost certainly don’t need 50 more to get high-level functionality out of your LLM.

Still, every once in awhile — and often enough for me that I’m sharing it — I find this useful.

The Catalogue Prompt

Here’s where this becomes a tool instead of just a neat idea.4 You can copy the prompt below, paste it into your LLM of choice, and include the catalogue or article alongside it.

Describe the problem, and it will recommend different techniques and way.

You are an expert prompt engineering consultant. You are thoughtful and thorough, insightful and diligent. You are patient with you explanations but will not sacrifice accuracy for sycophancy. I’m going to give you two things:

1. A comprehensive catalogue of every known prompting technique (the article pasted)

2. A description of my specific problem

Your job is to:

- Analyze my problem’s characteristics (type of task, available data, complexity, whether I have labeled examples, latency/cost constraints, whether accuracy or recall matters more)

- Recommend and rank 1-3 prompting techniques from the catalogue that are the best fit

- Explain WHY each technique fits my problem

- Provide a concrete example of how I would structure my prompt using each recommended technique

- Flag any technique-specific pitfalls or design decisions I should be aware of (e.g., exemplar ordering effects, answer extraction challenges)

- If my problem would benefit from combining techniques (e.g., Few-Shot CoT + Self-Consistency), explain how to layer them and/or provide a step-wise approach

Here is the catalogue of prompting techniques:

[PASTE THE FULL ARTICLE HERE]

Here is my problem:

[DESCRIBE YOUR PROBLEM HERE — include: what task you’re trying to accomplish, what data you have available, what model you’re using, what “good” looks like for your use case, and any constraints on cost/latency/complexity]

The better you describe your problem, the more context you provide, the better the output will be. This seems to work fairly well inside of Claude’s Projects feature, but it’s also been effective in Incognito Mode.5

The other thing the meta-prompt doesn’t include that might be useful is a note about what you’ve already tried and why it didn’t work. I omit it here because I expect most folks (and, often, me too) will just copy and paste it as is as a first try.

Conclusion

Honestly, this was intended for my own utility, but I’ve been using it enough to believe it’s worth sharing.

Obviously, the point of it is very much not to memorize a taxonomy, rather it's to have a reference tool for when the model isn't doing what you need and you're not sure why. You don’t need to memorize 58 knots and hitches, you need to know a bowline and a square knot exist, and know when to use each one.

Most of the time, you won't need this mini reference. Most of the time, you’ll have internalized the techniques already and, when you haven’t, the models are good enough that plain language covers you anyway. But for the other <5%ish of the time, when the task is complex or the stakes are real or you're three rephrased prompts deep and still getting boiled garbage, knowing that a technique exists can be the difference between giving up and getting it done.

Subscribe now

And, like anyone else with a modest professional interest, I sent a dozen or two Colorado Open Records Act requests to a variety of Colorado state agencies and institutions.1

In November 2024, the people of Colorado voted to create the Veterinary Professional Associate. In doing so, they placed the subsequent work of turning a ballot measure into a regulatory framework in the hands of the Colorado State Board of Veterinary Medicine. This is the work that determines much of what the Veterinary Professional Associate will be.

That work is deciding what training qualifies a graduate to enter the profession, what the supervision requirements look like, what scope of practice the role carries. That work is not glamorous, and I say that as a guy who has spent a lot of time in unglamorous jobs. The rulemaking is conducted through stakeholder meetings and redline cycles and motions seconded at hearings most members of the public will never attend. It is the work that turns democratic will into rules with real consequences for animals, for owners, and for the veterinarians and technicians who will share clinics with this new tier of practitioner.

While unglamorous, the work is important and consequential. It is work of public trust.

The legislative declaration of the Colorado State Board of Veterinary Medicine, from the Veterinary Practice Act, Title 12, Article 315:

“This article 315 is enacted as an exercise of the police powers of the state to promote the public health, safety, and welfare by safeguarding the people of this state against incompetent, dishonest, or unprincipled practitioners of veterinary medicine. It is hereby declared that the practice of veterinary medicine is a privilege conferred upon persons possessed of the personal and professional qualifications specified in this article 315.”

“Safeguarding the people of this state” is a powerful phrase. It’s not a trivial matter. The board's sole and exclusive statutory purpose is consumer protection under the state's police powers, safeguarding the public from incompetent, dishonest, or unprincipled practitioners. Not promoting the profession. Not supporting workforce expansion. Not industry development. And licensure is framed as a privilege, not a right.

Further, a regulatory body like the State Board of Veterinary Medicine functions only to the extent that the public, and the regulated profession, are able to have confidence that the rulemaking serves the people’s interests rather than the interests of the rulemakers or any other constituency.

Colorado law and DORA’s own internal policies recognize this directly. The Board’s Code of Ethics, DORA Policy 80-30, requires members to:

“Decline to deliberate, participate, or otherwise attempt to affect the outcome of any matter before the regulatory program when doing so may result in a conflict of interest or the appearance of a conflict of interest.”

The standard is appearance, not just actuality. While that may raise the eyebrows a bit, I think the standard is high because the standard has to be high. A board whose members participate in matters for which their neutrality is reasonably in doubt is not a board upon which the public can rely, regardless of whether the appearance of conflict cashes out in a documented financial transaction reported on one’s taxes.

Meanwhile, I’ve gotten some productions in response to my CORA requests. And what’s absent is as interesting as what’s included.

I love America. I really do.

I don’t know how many countries in the world2 allow just any old citizen to send a letter requesting document from a government agency, and that agency just sends it along, but I’m glad I live in one that does.

CORA No. 169-2026 (their numbering system) was filed by yours truly on April 04, 2026, received by Colorado’s Department of Regulatory Agencies on April 24, and the responsive production was emailed to me on April 29. The documents discussed below are public records, because I asked for them and a Regulatory Coordinator for the Colorado Department of Regulatory Agencies sent most of what I requested.

Government is so cool.

I practice in Pennsylvania, and have no financial interest in any hospital or clinic in Colorado. I have, I suppose, a conflicting interest in the existence or eradication of a credentialed mid-level practitioner. As a veterinarian and a pet owner, I think the VPA will not have a positive impact on the profession or on patient care. As a practice owner, I have to acknowledge that, per the assertions of its supporters, hiring a mid-level practitioner may prove to be cheaper — and thus make a hospital I own more profitable — than hiring a veterinarian. That’s a fairly broad interpretation of “conflict,” I admit, but I think it’s right to acknowledge it.

The people of Colorado voted, however narrowly, for the establishment of the VPA position. It’s not my place or purview to say that they’re wrong to do so, only that I disagree.

I’m not going too much into the nature of my disagreement with the establishment of the mid-level practitioner, rather, I’m going to discuss what I believe to be a significant ethical problem for the Colorado State Board of Veterinary Medicine as it pertains to the rulemaking of the VPA position.

Subscribe now

The Colorado State Veterinary Board is made up of nine members. Five licensed veterinarians,3 two veterinary technicians,4 and two public members.5

The board members, as they stand today:

• Carolyn E. Karrh, DVM (President) — term expires 6/30/2027

• Dean Vicksman, DVM (Vice President) — 6/30/2026

• Catherine Kuntz, DVM (Secretary) — 6/30/2027

• Tracey Jensen, DVM — 6/30/2029

• Martha Jo Myers, DVM — 6/30/2029

• Lynn Ferguson, RVT — 6/30/2027

• Rebecca Rose, RVT — 6/30/2027

• Heather Cammisa (Public) — 6/30/2029

• Anna Stout (Public) — 6/30/2028

The public members of the board, Anna Stout and Heather Cammisa, notably have professional and financial association with veterinary medicine.

Ms. Stout, a registered lobbyist, is the executive director of AlignCare Health, Inc, a company which describes its mission as “improv[ing] access to veterinary care for families in need by aligning community resources and activities.”

Ms. Cammisa is a member of the leadership team of Open Door Veterinary Collective, a 501(c)(3)company which “giv[es] veterinary clinics tools to expand access to care so more pets get care AND clinics get paid for their services.”

The statutory language, from § 12-315-106(2)(c) of the Colorado Veterinary Practice Act:

"(2) The governor shall appoint: ... (c) Two members to the board from the public at large who have no financial or professional association with the veterinary profession."

It’s hard to imagine a definition of “no financial or professional association with the veterinary profession” that does not preclude the board membership of people who lead companies with direct financial and professional associations with the veterinary profession.

And it’s not as though Ms. Stout or Ms. Cammisa hid this information. It’s on their respective LinkedIn profiles, on company websites, in trade publications and marketing materials. There’s no reasonable case to be made that Ms. Stout or Ms. Cammisa made any effort to obscure their association with the veterinary profession.

So this information is easily verifiable by the folks who are, presumably, supposed to check before they appoint someone to a State Board. In the previous sentence, Governor Jared Polis is “folks.”

It is the job of the board members to protect the people of Colorado, to safeguard.

Can the people of Colorado trust that Ms. Cammisa will protect their interests if a veterinarian employed by the Open Door Veterinary Collective’s biggest client receives a board complaint?

What if a veterinarian in Colorado is asked by AlignCare Health, Inc. to take on a difficult procedure, a fraught patient, or to accept an alternative or lesser payment? Would they dare decline knowing that the executive director sits on the State Board?

My questions are rhetorical, but they should also be irrelevant. On the plain language reading of the statute, it is difficult to see how either Ms. Cammisa or Ms. Stout satisfy the statutory restrictions applicable to the Board’s public members.6

One of the newest appointees to the Colorado State Board of Veterinary Medicine is Tracey Jensen, DVM, with her term expiring on June 30th, 2029.

Dr. Jensen is, by every public record I can find, an accomplished and well-regarded veterinarian. She is a past-president of the American Animal Hospital Association and Wellington Veterinary Hospital, which she co-owned with her husband, won AAHA Practice of the Year in 2012. She was named Petplan’s “Veterinarian of the Year” in 2015.

As recently as February 2026, it’s reported that Dr. Tracey Jensen is the spouse of Dr. Wayne Jensen. Dr. Wayne Jensen is the architect of Colorado State University’s Veterinary Professional Associate program.

§ 24-18-108.5(2), C.R.S. is a provision of Colorado legislation specifically concerning board members like Dr. Tracey Jensen, it says,

[A member] “shall not perform an official act which may have a direct economic benefit on a business or other undertaking in which such member has a direct or substantial financial interest.

I believe that two aspects of the above described circumstances could be seen to rise to a credible claim of conflict of interest.

First, “an official act with direct economic benefit on a business or other undertaking.” Under the post-January 1, 2026 amendments the Colorado State Veterinary Board is now required to register VPAs, set qualifications, adopt scope-of-practice rules, define supervision, and approve credentialing organizations and training programs (§ 12-315-106(5)(a)–(b), § 12-315-201, etc.).

Those acts will have substantial impact on the value of the CSU VPA program, as yet the only one of its kind in the country. Acts favorable to a permissive scope, broad supervision ratios, and/or rapid credentialing pathways will facilitate the increase of enrollment, tuition revenue, faculty positions, and external funding for CSU’s new program.

Second, “a business or undertaking in which the member has a direct or substantial financial interest.” Dr. Wayne Jensen’s professional standing, position, and compensation as the architect of a decade-long institutional initiative are tied to the program’s regulatory success.

Colorado treats a spouse’s financial interest as the officer’s own across multiple ethics regimes. The constitutional gift ban at Article XXIX, Section 3(2) of the Colorado Constitution attributes to the public officer any gift or thing of value over the statutory threshold given to that officer’s spouse, prohibiting receipt “either directly or indirectly as the beneficiary of a gift or thing of value given to such person’s spouse or dependent child.”

The Colorado Revised Nonprofit Corporation Act, at C.R.S. § 7-128-501(5), defines a “party related to a director”, for purposes of conflicting-interest-transaction analysis, to include “a spouse”, alongside lineal relatives and entities in which the spouse holds an interest.

I find it particularly telling that the Colorado Office of Legislative Legal Services, in its published ethics guidance, offers as its first illustrative scenario a legislator whose spouse “is a well-known and well-recognized research scientist who heads a special institute for biotechnological research at one of our state’s leading research universities,” voting on legislation that would assist that institute, identifying it as a textbook conflict-of-interest situation requiring disclosure.

The statute that governs Dr. Jensen’s situation, § 24-18-108.5(2), prohibiting an “official act which may have a direct economic benefit on a business or other undertaking in which such member has a direct or substantial financial interest”, does not separately define “financial interest” in isolation from the rest of Colorado’s ethics framework. The framework consistently attributes the spouse’s financial interest to the officer. Admittedly, I’m not a lawyer, but as a nearly intelligent layperson, I have a hard time reading § 24-18-108.5(2) any other way.

Set the statute aside for a moment. One spouse builds the program; the other sits on the board that decides whether the program qualifies. That's the situation on its face.

Can the people of Colorado trust that Dr. Tracey Jensen will act exclusively in their interest when she rules on matters of the VPA program? A program her spouse spearheaded, for years, at Colorado State University?

Another rhetorical question, sure, but it has a real answer. This is why there’s a recusal mechanism for board members with potential conflicts, even for those with the appearance of conflict. The system was set up to absorb and manage a situation like this.

According to more than 400 pages produced in response to a CORA, one that named this specific conflict of interest, there is documentation that Dr. Tracey Jensen has recused herself from other matters before the CSBVM, but never from the VPA rulemaking.

Subscribe now

In its production response, the Department of Regulatory Agencies (“DORA”) cited a single CORA exemption: § 24-72-204(2)(a)(VII), the provision that protects “electronic mail addresses, telephone numbers, or home addresses” provided to a state agency for purposes of future communication.

That exemption permits the redaction of personal contact data points. It does not permit the withholding of substantive records, like, I imagine, ethics and conflict declarations.

Item 6 of the request I submitted asked for "all conflict-of-interest disclosure forms, recusal notices, or ethics filings" submitted by Dr. Jensen, "including any disclosures related to the VPA program, the MS-VCC program, or Colorado State University."

The production contains exactly two ethics-policy documents from Dr. Jensen: a generic Code of Conduct Acknowledgment Form she signed on July 14, 2025, and a generic FY26 Annual Board Member Training attestation dated April 7, 2026. Neither contains any specific disclosure of a conflict and neither references her husband, the MS-VCC program, the VPA, or Colorado State University.

And I asked for it, specifically. Item 8 of my CORA request sought “All communications between DORA staff or the Board and Tracey Jensen, DVM, regarding any potential conflict of interest arising from any professional or personal relationship between Dr. Tracey Jensen and Dr. Wayne Jensen, DVM, PhD, MBA, and his role in the MS-VCC program at CSU.” The production contains zero such communications.

This is consistent with one of two readings. Either Dr. Jensen filed no conflict-of-interest disclosures in connection with her service on the Board, and there were no communications between her and DORA about the conflict. Or such records exist and were withheld.

The exemption DORA cited does not authorize withholding ethics filings, so if the second reading is correct, the withholding seems unsupported by the cited statutory authority.

If the first reading is correct, the absence of disclosure is the entire problem.

Helpfully, even if Dr. Jensen didn’t flag her own conflicts of interest, a member of the public did. Before rulemaking occurred, there was an opportunity for public comments on the matter.

On Monday, July 14, 2025, at 9:10 AM, Sarah Carter, DVM, sent an email to the rulemaking inbox. Her message:

“I am writing in with concerns of conflict of interest/selection bias for the two newly appointed veterinarians on the Colorado state board of veterinary medicine, appointed by Governor Polis. Tracey Jensen DVM was appointed by Governor Polis. She is posted on the website ‘All Pets deserve Vet [Care]’ as an endorsing veterinarian of prop 129 and creation of a VPA. . . . Her husband, Wayne Jensen, who was formerly the department head at CSU CVM, is intimately involved in developing this VPA role despite majority of colleagues, professional groups, state VMAs, state veterinary technician associations, and the AVMA being opposed. . . . So I ask, is appointing Tracey a conflict of interest?”

Dr. Carter’s email was received by DORA. It is in the production. The Board therefore had, at least thirty-one days before the August 14 hearing, formal written notice that the conflict had been flagged.

The CORA production contains no documented response to Dr. Carter’s email. No reply from DORA staff. No memo from the Senior Assistant Attorney General assigned to the Board. No notation in subsequent meeting minutes. No recusal protocol triggered.

The question of conflict was raised, by a veterinarian, in writing, on the official record. The record is silent on what, if anything, the Board did to answer the question.

On August 14, 2025, the Board convened the Permanent Rulemaking Hearing. It was Dr. Tracey Jensen’s first board meeting. Roll call lists her as present.

The minutes record two motions on Agenda Item 1, the VPA rulemaking. The first was Anna Stout’s motion to adopt the rules, seconded by the second new Polis appointee, Dr. Martha Jo Myers (more in a moment). It carried.

The second was a motion by Heather Cammisa to designate the American Association of Veterinary State Boards (AAVSB) as the national credentialing entity for the VPA. It was seconded by Dr. Tracey Jensen. Thus, even though the public records I’d requested were not produced, I believe it’s safe to conclude that Dr. Jensen did not recuse herself. Two veteran members, Dr. Dean Vicksman and Lynn Ferguson, RVT, voted no. The motion carried.

The minutes record no recusal by Dr. Jensen on either motion.

The Board handles recusal in general. The same minutes, in the same meeting, record that Ms. Lynn Ferguson “recused from participation in this case and departed the meeting room” on a disciplinary matter, and that Dr. Carolyn Karrh recused on another. In meetings since, Dr. Jensen is recorded as having left the room, recusing herself from other disciplinary matters.

The recusal practice at this Board is to leave the room and have the recusal noted in the minutes. On the rulemaking that will have a tremendous impact on determining whether her husband’s program qualifies as a path to the credential, the new appointee made the choice to remain present, to vote, and to second the contested motion at her first meeting.

The motion Dr. Jensen seconded matters in its own right, not because the designation it made is necessarily wrong, but because it is exactly the kind of designation that the rulemaking process exists to deliberate openly.

The rulemaking process exists so that designations of this consequence get that deliberation and scrutiny. The meeting minutes do not show such scrutiny.

The second new appointee is Martha Jo Myers, DVM, term expiring June 30, 2029. Her name also appears on the published endorser list at allpetsdeservevetcare.com.

She is a paid veterinary advisor and consultant to Vetster, a commercial veterinary telehealth platform; this is documented on Vetster’s own site and in trade press. The VPA category, as designed, expands supervised veterinary practice capacity in ways structurally compatible with telehealth-platform business models.

Dr. Myers, like Dr. Jensen, made an affirmative motion contribution at the August 14 hearing when she seconded Ms. Stout’s motion to adopt the VPA rules. The minutes record no recusal for Dr. Myers either.

Dr. Carter’s July 14 comment, sent to DORA, named Dr. Myers explicitly as a second potential conflict, and the comment is in the production, and the record is similarly silent on any documented response.

At least four of the nine members of the Colorado State Board of Veterinary Medicine appear, to my eye, to have seemingly undisclosed conflicts of interest and/or direct professional or financial interests that seem, according this layman’s reading of Colorado law, should preclude them from rulemaking on the matter of the Veterinary Professional Associate or, in some cases, membership on the State Board.

Colorado law does seem to afford a presumption of integrity and impartiality to those serving in quasi-judicial capacities. Mountain States Telephone & Telegraph Co. v. Public Utilities Commission, 763 P.2d 1020 (Colo. 1988), says so plainly, and I have no quarrel with the principle. But the same opinion seems to make clear that the presumption is rebuttable, and that it holds only in the "absence of a personal, financial, or official stake in the decision evidencing a conflict of interest." That doesn’t sound like a blanket immunity, it sounds like a condition. And the facts here seem to fail it.

Then there’s Goebel v. Benton, 830 P.2d 995 (Colo. 1992), holds that "mere allegations, opinions, or conclusions unsubstantiated by facts" are insufficient to mandate disqualification. And I think that makes sense, too, but Goebel is a case in which the Colorado Supreme Court ordered the judge to recuse. The court held that when the totality of the facts demonstrates a "bent of mind" or an appearance of bias, disqualification is required as a matter of law. A spousal relationship to the person who spearheaded the creation of the only qualifying academic program is not a “mere allegation,” it's a matter of public record. A prior public endorsement of the ballot measure is not an unsubstantiated opinion about bias, it's a documented position.

And Venard v. DOC, 72 P.2d 446 (Colo. App. 2003), provides that recusal is appropriate when a board member's impartiality "could reasonably be called into question", which, to this layperson, seems to describe the situation before the Board with some precision.

Two were publicly aligned with the ballot measure before their appointment and they presently work in roles that deal directly with the veterinary profession. One has a documented family connection to the state’s only academic training program the rulemaking will determine to qualify or not qualify, and another has a documented commercial advisory connection to a category of practice the rulemaking expands.

A member of the public raised the conflicts in writing thirty-one days before the first vote. The Board’s records contain no disclosure forms, no recusal notices on the VPA, no ethics filings, no communications specifically addressing the conflict — none of which are protected by the public records act exemption that DORA invoked.

The first vote both new appointees participated in was the rulemaking on the matter that the conflicts touched. Both seconded motions on it. Neither recused.

The State Board of Veterinary Medicine has a Code of Ethics that requires recusal where the appearance of a conflict exists. The standard the Board sets for itself is a standard the record does not show was met. A regulatory body that does not visibly meet its own ethics standard is a regulatory body the public has no way to verify is serving its interests.

You might say the rules would've passed anyway, waving off the concerns. That they could’ve redone it and we’d still be in the same stage of things. That it wouldn’t have mattered.

But in the August 14th, 2025 meeting Motion 1 (rules adoption) was moved by Ms. Stout and seconded by Dr. Myers. Motion 2 (AAVSB designation) was moved by Ms. Cammisa and seconded by Dr. Jensen. That meeting was attended by eight members of the board, with Ms. Rose absent. As I’ve described above, Two of the eight present should not have been seated at all under § 12-315-106(2)(c); two more Board members should have recused under § 24-18-108.5(2) and the Board's own Code of Ethics. So, doing the math here, if the members with conflicts of interest recuse and two more are statutorily ineligible? They don't have a quorum to pass either motion anymore. And if they did vote along the same lines, what was a 6-2 vote in the affirmative would be tied 2-2. That substantially changes the climate of the Veterinary Professional Associate in Colorado.

The structural questions about AAVSB, about Standard 6 of the AVMA Council on Education’s Accreditation Policies and Procedures, about CSU’s $50 million for a Veterinary Health and Education Complex appropriation under HB 24-1231, about the curriculum overlap Dr. Wayne Jensen himself documented, those exist independently and will be addressed in their own time. They do not depend on the outcome of the conflict-of-interest question.

What does depend on it is whether the people of Colorado can have confidence that what was decided at that August 14 hearing was decided in their interest.

And here’s another reason I love America: there’s a way to fix all of this.

A board appointment is a deal: the Governor offers a seat; the appointee accepts a set of rules. The rules are written down. The Code of Ethics is on DORA’s website. The Practice Act is on the legislature’s website. The recusal mechanism is in use, the Board’s own meeting minutes of the same show two members of this Board recusing themselves on the same day on disciplinary matters. Published minutes from a later meeting show that Dr. Tracey Jensen recused herself on certain disciplinary matters. The system works. It just requires the people who volunteered to participate to use the system that's been built.

The remedy for an undisclosed conflict is disclosure. The remedy for a missing recusal is recusal. The remedy for a vote taken without either is reconsideration of the vote with proper procedure. A motion can be reconsidered. A regulation can be redrafted. A board’s reputation, once a problem is named honestly and addressed visibly, can be reestablished.

The path forward is not difficult, though it may be uncomfortable for some involved. Dr. Tracey Jensen and Dr. Martha Jo Myers can file conflict-of-interest disclosures specifying the relationships, the financial interests, and the prior public positions that bear on the matters before the Board. Both can recuse from further VPA-related rulemaking. The Board members without conflicts can reconsider, with full disclosure on the record, the motions taken on August 14.

A 52.7-47.2 vote of the Colorado electorate authorized the creation of the Veterinary Professional Associate, it did not authorize a particular board composition, and it certainly did not authorize the suspension of § 12-315-106(2)(c) or § 24-18-108.5(2). Voters don't get to vote conflicts of interest into legality, and a Governor doesn't get to appoint around statutory eligibility requirements because a measure passed. The statute means what it says, I would think, regardless of whether it passed by a single vote or two million.

DORA can produce any disclosure forms, recusal notices, ethics filings, attestation responses, and conflict-related communications that exist for any Board member during the rulemaking period.

If those records exist and were merely overlooked, they will surface. If they do not exist, that absence is itself the answer to the trust question, and the answer requires a remedy.

While rhetorical questions litter this piece, the question of who is this rulemaking actually serving is not a rhetorical question. I believe it has an answer, and that answer is in the record.

I will keep sharing the record as it comes in. Recordings and minutes from the April 2026 Colorado State Veterinary Board meeting are not available as of the time of this writing.

As Mr. Mark Cushing frequently ends his articles: much more to come.

This is the first letter I’ll write you, and not the last. It was, briefly, a source of raised-eyebrows-in-my-direction from your mother that I wrote one letter to both of you, instead of one each. But I have not met either of you yet, and so I address our twins a pair, just for now.

A few weeks ago a missile hit a girls’ school on the other side of the world. It was a Saturday, the start of the week in Iran, so the classrooms were full of girls aged seven to 12. News websites show photographs of blood staining small, pink backpacks. One hundred and sixty-eight of those girls never made it home, and far more than 168 fathers around the world are trying to make sense of it. I’m certain that none of us can.

I don’t have politics. I have daughters.

You have not arrived just yet, still, I hope, a few weeks away from being born. As your dad, I am already possessed of a deep, desperate, almost debilitating desire to protect you, but I see photographs on the news, and I’m not sure I know how to do that in a world where flying robots fire missiles at girls’ schools.

But, just as for your brothers, it is not right for me to protect you from everything in the world. Instead I will take this letter, delivered on Mothers’ Day, to tell you about the women who came before you. I will tell you stories of their kindness and courage, of their love and dedication, of their strength and gentleness. The world will try to tell what those words mean, but I think you should learn it from your family first.

I’ll tell you about your great-grandmother, Polly, who ran her own business for decades, a very big deal for a woman born in the 1920s. This is the first of many stories you’ll hear about her:

The woman who told me this story had gotten divorced and was getting married again, another very big deal for a Catholic woman in the 1970s. And on the day of her wedding she’d come to Polly’s beauty shop to “get her hair did,” as they said in that part of the world. The bride-to-be was distraught, inconsolable. She’d just learned that her daughter had been diagnosed with leukemia. She was going to call off her wedding, postpone the ceremony.

Polly was sympathetic, of course, and said the girls (the women who worked for her) would do this woman’s hair and makeup anyway, “just for practice.” “The works,” said your great-grandmother. The girls went to work, spending hours doing this woman’s hair and makeup and nails and whatever else “the works” entailed.

The bride-to-be was, she told me, so overwhelmed by how she looked and the kindness she received that she decided to get married that day. She was happily married for more than four decades. And her daughter beat leukemia.

When I asked your great-grandmother about it, she didn’t remember. And not because time had started to rob her of memories, but because she did things like that so often that they no longer stood out to her. It was just another day at the office for her.

I will tell you about your grandmother too. She managed the processes and safety of chemical and petroleum operations all over the world for decades. It's hard to quantify1 just what kind of impact a mind as brilliant as hers has had on the world, but I’m confident in asserting that her work has saved even more lives than mine. In her retirement she has turned that formidable intellect to watercolor painting, and she is, genuinely, exceptionally good at it. A woman who spent her career making sure that things didn't explode now spends her afternoons making sure that the sunlight on a hillside looks just so. It’s not that hers is now a smaller life, it’s that she was never only one thing.

I will tell you about your aunts, Diane, Grace, and Tess. Their impact on your lives, I suspect, will have less to do with what I tell you about them and more to do with what you hear from them. It may sound unfair to group my aunt and yours into a single unit, but they share a rare and lovely gift: they laugh. At holidays, at meals, on trips, always with - and sometimes at - the world. They do so irrepressibly, irresistibly, and you cannot help but join them. I delight to know that you will find yourselves laughing before you understand why. While they are each of them thoughtful and generous in their gifts, it is this one, I think, to be the most generous they ever offer: that ability to walk into a room and make everyone in it enjoy the time a bit more.

I will tell you of your mother too, of course. Who is the bravest, strongest, and most loving person I’ve ever known. To paraphrase and contradict Oscar Wilde, I will tell you about the importance of being earnest, but I will also tell you that you could do far worse than growing up to become your mother.

One of the things I love most about your mother is her curiosity. About everything. She asks questions the way other people breathe, not merely as a habit, rather as a means of sustaining life.

I love that she’s patient in a way that I deeply admire and rarely replicate.

I love that she delights in the smallest of things: a dog who has learned a new trick, a fuzzy creature she hasn’t seen before, or your brothers telling her something they learned at school.

I love that she loves animals with a profound depth and serious affection, and she extends that same ferocious, diligent care to her boys, and, already, to both of you.

I write to you girls for many of the same reasons I write to your brother, because it is imperative that you hear stories of love and courage. While love is familiar, I think courage can be more nebulous. It is too easy to mistake demonstration of courage for volume, for spectacle, or for the application of force. The world too often believes that strength is a clenched fist, a loud voice, or a drone circling above a schoolyard. But your mother’s courage is never any of those things. Rather it is demonstrated in the decision to show up again after the worst day. It is in the choice to be gentle with the people she loves even when the world has not always been gentle to her. She proves every day that grit, determination, and diligence are every bit as formidable as strength of arm or as flying robots.

I think of the big world you two will enter, and I do worry. But then I think on the smaller world into which you come. I think of how, when your mother and I learned that we were having twin girls, we told our family and friends. And, every time, without exception, their eyes went wide. Then their mouths opened, jaws hanging. And then, and I swear this is true, every one of them, every brave and brilliant and capable person in our lives, every person who saved lives, who fought disease, who laughed as they took on the world, those people looked just a little bit frightened.

Your mother and I are not so different from our family and friends. We have conversations that are, approximately, only a thin level of logistical and emotional sophistication above “how the hell are we going to do this?” And then we smile, laugh, and know we’ll find a way.

I don’t tell you that to guilt you or to worry you. I tell you that because I think it is remarkable, and worth recording, that even before you were born, when you were but a few cells and an aching hope, all of these strong, brave, brilliant, and loving people were absolutely and already intimidated by the two of you. You two had our whole world shaking in their boots before you even arrived.

So I worry about rockets and flying robots just like I worry about pressures and societies and a world that too often falls well short of doing right by girls and women. But I think, upon reflection, that I could worry a little less about two girls coming into a world that is already intimidated by you.

And, I’m happy to tell you, every one of those wide-eyed, open-mouthed, slightly frightened, and genuinely affectionate people simply cannot wait to meet you. Your brothers talk about you every day.2 Your mother and I daydream about what you might look like, about your respective personalities, about who you’ll be, and taking you all for bike rides. We’ll do it as we assemble cribs and acquire pink baby toys and pick colors for nail polish so we can tell you two apart early on.

In the meantime, you two stay put for a bit longer, please, and your brothers and I will wish your mom a happy Mother’s Day3 for you both.

I love you, Hawley. I love you, Brenna.

Love,

Dad

But, well, I read things. And so do you. And that’s what brings us here.2

This week I read the decision of Chiles v. Salazar, decided by the Supreme Court on March 31, 2026.3 And I believe it has something important to say to every licensed medical professional in the United States, including, of course, veterinarians. Perhaps especially veterinarians.

Subscribe now

The Decision

In 2019, the Colorado Legislature passed a law banning licensed counselors from engaging in “conversion therapy” with minors. The law defined “conversion therapy” as “any practice or treatment that attempts to change an individual’s sexual orientation or gender identity.” Enter Kaley Chiles, MA, LPC, LAC, a licensed mental health counselor in Colorado. Ms. Chiles uses only talk therapy. As one would expect from someone with her credentials, there are no physical interventions, no medications, just words.

She argued that the law, as applied to her, violated her rights protected by the First Amendment.

The Supreme Court agreed, overwhelmingly, 8-1.

Justice Neil Gorsuch authored the majority opinion, finding that Colorado’s law doesn’t merely regulate a type of medical treatment, it picks sides. The statute explicitly permits counselors to provide “acceptance, support, and understanding” for identity exploration and to assist patients “undergoing gender transition.” And, further, it prohibits medical and mental health professionals from speaking in ways that would help a patient move in another direction.

They could talk a patient towards one conclusion, but not another. That, the Court held, is “viewpoint discrimination,” which “represents an even more ‘egregious form’ of content regulation from which governments must nearly always ‘abstain.’”

It’s absolutely crucial to note that the Court was emphatic, not in protecting “conversion therapy,” with its history of dangerous, abusive, even torturous interventions, but rather in protecting the right to freedom of speech found in the First Amendment.

The Coalition

I was surprised to learn that an 8-1 decision is, in itself, unusual at the Supreme Court. Roughly a third of the Court’s decisions are unanimous, and near-unanimous outcomes show up regularly on statutory interpretation questions that could, I suspect, have a sedative effect on even the most diligent law clerk.

But this was not one of those cases. This case is flashpoint in an ongoing culture war. It affects conversion therapy, LGBTQ+ rights, and the scope of state regulatory authority. The eight Justices of the majority, Roberts, Thomas, Alito, Sotomayor, Kagan, Kavanaugh, Barrett, and Gorsuch, are not, shall we say, frequently found on the same side of these sort of cases. Justices Sotomayor and Kagan make up the liberal wing of the Court, and they sided with Justices Thomas, Alito, and Gorsuch, who, decidedly, do not.

Justice Kagan wrote a concurrence, joined by Justice Sotomayor, that agreed that the Colorado law was unconstitutional because it discriminates on viewpoint. The concurrence also noted that a law drafted differently, one that was content-based but viewpoint-neutral, would raise “a different and more difficult question.”

When Justices Sotomayor and Thomas agree on the outcome of a politically charged case, it's worth paying attention to why. According to the Empirical SCOTUS Stat Review for the 2023-2024 term, Justices Thomas and Sotomayor agreed on the outcome of their cases only 50.85% of the time, tied with Justices Thomas and Kagan for the lowest agreement rate of any pairing on the bench. The entire term produced just two 8-1 decisions. Justice Through the 2023-2024 term, Thomas has been the lone dissenter in 8-1 decisions thirty-six times.

The fact that Justice Jackson was the lone holdout, and read her dissent from the bench - a step justices often reserve for cases where they believe the majority has committed a serious error - tells you something about how clearly not only Justice Jackson, but also the other eight justices, saw the question.

What This Article Is and Isn’t

This isn’t an article about conversion therapy. It’s not an article about talk therapy. It’s not an article about hirudotherapy. It’s not an article about any kind of therapy.

This is an article about how the First Amendment and the right to freedom of speech is treated when it comes to medical professionals in the United States.

It’s about whether your doctor, your therapist, your veterinarian, should or can be governed by licensing boards staffed by experts in their field, or by statutes passed by legislators who, by and large and with all due respect, are not.

It’s about the difference between what a professional says and what a professional does. And about why that distinction matters even and/or especially when the line between the two starts to blur.

It’s about the uncomfortable but essential truth that just because the medical establishment believes something today doesn’t mean that it will believe the same thing tomorrow. It’s a feature of medicine, of science. And it’s one of the foremost reasons why we shouldn’t be trying to etch today’s medical consensus into law.

Colorado House Bill 19-1129 and What It Actually Did

Colorado House Bill 19-1129 was signed into law in 2019. The law did several things. As mentioned, it defined “conversion therapy” as “any practice or treatment that “attempts or purports to change an individual’s sexual orientation or gender identity.” Further, it classified conversion therapy with a minor as “unprofessional conduct,” which made it grounds for fines, probation, and/or loss of professional license.

Whatever you think about conversion therapy, that is, legally, a very broad definition of conversion therapy. The term “conversion therapy” connotes abusive and harmful acts, it conjures thoughts of electroshock “therapy” and practices that would seem more like torture than medicine. The Court’s decision was quickly condemned by the likes of Amnesty International and The Trevor Project. The law also contained very specific carve-outs, explicitly allowing acceptance, support, and understanding for identity exploration and development as long as the counseling did not seek to change sexual orientation or gender identity. The law also explicitly permitted assistance to a person undergoing gender transition.

Reading it all together, absent the breathless political and cultural lenses of headline journalism, and you start to get a sense of the problem the Court identified. A licensed counselor in Colorado, i.e. Kaley Chiles, could talk about with a minor patient about their sexual orientation or gender identity, but only in one direction.

The law permitted counselors to assist a patient exploring or transitioning to a new identity while prohibiting counselors from assisting a patient who wanted to move in the other direction. As the Supreme Court’s syllabus put it, “the law permits Ms. Chiles to speak in ways that encourage a client undergoing gender transition, but prohibits her from speaking in ways that help a client realign their identity with their sex.”

Writes Justice Gorsuch:

On matters of sexuality and gender, Ms. Chiles’s clients, including minors, come to her with different goals in mind. Some “are content with” their sexual orientation and gender identity and seek assistance only with “social issues, family relationships,” and the like. Id., at 207a, 213a–214a. In cases like those, Ms. Chiles does not try to persuade her clients to “change their attractions, behavior, or identity,” but aims instead to help them address their stated goals. Id., at 214a; see also id., at 207a. Other clients, however, come to her hoping to “reduce or eliminate unwanted sexual attractions, change sexual behaviors, or grow in the experience of harmony with [their] bod[ies].” Ibid. And in these cases, too, Ms. Chiles seeks to help her clients reach their own stated objectives. Ibid. In doing so, she does not prescribe any medicines, perform any physical treatments, or engage in any coercive or aversive practices. Id., at 205a–207a. All Ms. Chiles offers is talk therapy. Ibid.

If a minor client asked their counselor to help support their transition to a new gender, the counselor could help. If a minor client changes their mind, as humans sometimes do, the counselor is now forbidden by law from supporting them.

The asymmetry is the crux of the decision authored by Justice Gorsuch. The Justice held that this isn’t just content-based regulation of speech, but viewpoint-based regulation of speech.

A content-based regulation is a law that restricts speech based on what is being said, rather than how, when or where it’s being said. A classic example, and the one Justice Gorsuch walks through in his opinion, is Cohen v. California (1971). The State of California charged a man with disturbing the peace for wearing a jacket that said “Fuck the Draft” in a courthouse.4 The disturbing the peace statute mostly regulated conduct, like brawling or causing a scene, but, as applied to Mr. Cohen, the State of California was punishing him for the content of his expression, not the manner of it. And the Court held that was a First Amendment problem (5-4).

A viewpoint-based regulation is a law that doesn’t just restrict a topic, it picks a side within the topic. The government decides which perspective on a subject is permissible and which isn’t.

One of the landmark cases for viewpoint-based discrimination is Rosenberger v. Rector and Visitors of University of Virginia (1995). The University of Virginia funded student publications but refused to fund a Christian magazine. The Court held that this was viewpoint discrimination (5-4), as the university allowed student speech about philosophy, culture, and ethics, but denied funding to one particular perspective on those subjects. So the topic was permitted, but the viewpoint wasn’t. And that’s also a First Amendment problem.

Colorado’s law fits the Rosenberger pattern. The topic, therapeutic conversations about sexual orientation and gender identity with minors, was permitted. But the law allowed counselors to speak in ways that supported identity exploration and gender transition while prohibiting speech that moved in the opposite direction. Same topic, with one viewpoint allowed, the other banned.

Justice Gorsuch’s opinion treats this as a straightforward application of existing doctrine. Viewpoint discrimination, he writes, represents “an egregious form” of content regulation, one from which governments must “nearly always abstain.” He acknowledges the difficulty of the underlying policy question, which is how best to help minors struggling with issues of identity and orientation is a subject of what Justice Thomas once called “fierce public debate.” But the opinion’s conclusion is unequivocal: “The First Amendment stands as a shield against any effort to enforce orthodoxy in thought or speech in this country. It reflects instead a judgment that every American possesses an inalienable right to think and speak freely, and a faith in the free marketplace of ideas as the best means for discovering truth.”

That last line matters for where we’re going as we careen down this research rabbit hole. The Court isn’t saying conversion therapy is good. It isn’t saying conversion therapy is bad.5 It’s saying the government doesn’t get to decide which side of a conversation between a licensed professional and a willing patient is legally permissible. A prevailing standard of care, Gorsuch notes, “may reflect what most practitioners believe today, but it cannot mark the outer boundary of what they may say tomorrow.”

The judgment of the Tenth Circuit was reversed and the case remanded, which means that the lower courts must now evaluate Colorado’s law under strict scrutiny, the most demanding standard of constitutional review and one the government rarely survives. The law remains on the books, but its judicial oomph has all but evaporated.6

Justice Jackson’s Dissent and Its Importance

Justice Jackson opens her dissent with a hundred-year-old quotation: “there is no right to practice medicine which is not subordinate to the police power of the States.” That’s from Lambert v. Yellowley, decided in 1926 (5-4). “This was true 100 years ago,” she writes, “and it should be true today.”

It’s a powerful line, as Justice Louis D. Brandeis was wont to pen, and the tradition she’s invoking is real. States have regulated the practice of medicine for a very long time7 through licensing, through malpractice regimes, through standards of care. And they’ve done so without the First Amendment standing in their way. That history is not in dispute. What’s in dispute is whether that tradition stretches to cover what Colorado did here.

Justice Jackson’s central argument is that the majority draws a constitutionally meaningless line between treatments administered through physical acts and treatments administered through words. Talk therapy, she writes, “is a medical treatment.” If a state can ban a harmful physical procedure, she reasons, it can ban a harmful verbal one too. The restriction on the counselor’s speech isn’t the purpose of the law but a byproduct. The state is regulating medical conduct, and the fact that the conduct happens to consist entirely of speech doesn’t change the analysis. She calls the majority’s contrary reasoning “maddeningly circular.”

There’s an internal logic to this that deserves acknowledgment. If you accept that talk therapy is treatment, and that the state can regulate treatment, then the state can regulate talk therapy. The speech restriction is, in Justice Jackson’s framing, “incidental,” a word that does a lot of heavy lifting in First Amendment law. Under prior precedent, particularly NIFLA v. Becerra (2018), restrictions on speech that are merely incidental to the regulation of professional conduct don’t trigger the heightened scrutiny that restrictions on speech-as-speech do.

But here’s where Justice Jackson’s dissent starts to raise concerns that extend well beyond conversion therapy, and well beyond this case, regardless of where you come down on the underlying policy.

Justice Jackson concedes that Colorado’s law discriminates based on viewpoint. She writes that “heightened scrutiny is not warranted” even while acknowledging the viewpoint asymmetry in the statute. Her position is that viewpoint neutrality is simply the wrong framework when the state is regulating the provision of medical care. As she puts it: “Neutrality is not — and cannot be — the touchstone of the laws that govern the quality of care that professionals provide to patients.” Standards-based regulation, she argues, exists because the state has a view about safety and efficacy. The entire point of regulating medical practice is to pick a side, to say this treatment is acceptable and that one is not.

And the justice has a point. Under Justice Jackson’s framework, a state legislature could identify a prevailing medical consensus, codify it into statute, and then prohibit licensed professionals from even discussing alternatives with patients, and all without triggering any heightened constitutional scrutiny. The professional’s speech isn’t being restricted as speech, it’s being restricted as treatment. The fact that the patient walked in the door asking for that conversation is, constitutionally, irrelevant.

Justice Jackson is candid about the stakes she sees going the other way. She warns that the majority’s decision “opens a dangerous can of worms” and could make speech-based medical treatments “effectively unregulatable.” She lists state regulations that require professionals to promote a patient’s “best interests,” to provide care that is not “cruel,” to “assure client welfare and protection,” and argues all of them are now potentially vulnerable because they involve the state dictating the content of a provider’s speech. “In the worst-case scenario,” she writes, “our medical system unravels as various licensed healthcare professionals start broadly wielding their newfound constitutional right to provide substandard medical care.”

That’s a real concern, and I dug into this research expecting to agree with her. But eight justices looked at it and came out the other way. And their reasoning, I think, matters more than their vote count.

The majority isn’t saying that the state can never regulate what a medical professional says. It’s saying the state can’t regulate which side of a conversation a medical professional is allowed to take. There is a line between a state saying “you must meet certain professional standards” and a state saying “you may express this viewpoint but not that one.” The first is how licensing boards and malpractice law have always worked. The second is what Colorado did. And the distinction between the two is, constitutionally and medically, the whole ballgame.

Justice Jackson, of course, has a point. If talk therapy is therapy, and the state can regulate therapy, then the viewpoint discrimination framework starts to look like a category error in which we're applying a speech doctrine to what is, functionally, a medical intervention. Her logic is consistent, and her warning that the majority's reasoning could be weaponized against legitimate medical regulation isn't hypothetical hand-wringing but a real vulnerability. But eight of her colleagues looked at the tension between the state's power to regulate medical practice and the First Amendment's prohibition on viewpoint discrimination, and decided which one bends.

The answer, for this Court, is that the First Amendment does not.

Not because regulating medical care is beyond the purview of the state, but because the moment you allow a state legislature to dictate which side of a professional conversation is legally permissible, you've created a tool that will not always be wielded by people who are “right.” Justice Jackson is probably correct that this decision will be cited in ways that make her, and me, uncomfortable. But the majority decided that the cost of dulling the First Amendment is higher than the cost of whatever comes next. History, and the Constitution, suggests they're right.

Subscribe now

What It Could Look Like for Veterinarians

Moving from the Supreme Court to the veterinary hospital sounds like a significant shift in setting, but I’ll offer an example of what such a law, stripped of its cultural connotations, might look like for a practicing veterinarian.

A client walks in with a four-year-old Labrador Retriever named Bruno. Bruno, is a not-so-slim 113-pounds of rowdy Labrador, and he’s limping on his right hind leg. You’d know that particular limp from 20 yards, but you’re a professional, so you do your physical examination, find a clear-as-day drawer sign, and the radiograph of the stifle shows a bucket of effusion.

The best surgery to repair a ruptured cranial cruciate ligament is the tibial plateau leveling osteotomy. It’s the most common orthopedic surgery done on the hindlimb in the United States, accounting for something like 75% of all surgical interventions for the torn CrCL. By and large, the outcomes are excellent.

But the recovery can be tough. It’s at least eight weeks of strict exercise restriction in a dog who hasn’t been simultaneously conscious and stationary for five consecutive seconds since he was born. And “strict” means no stairs, no slick surfaces, no running in the yard, no jumping on furniture, no unsupervised or off-leash movement. We live in a pen or a kennel (with a non-slip mat) for eight weeks at least. This is a major osteotomy on a tibia that takes a lot of force from Bruno, and if the owner is noncompliant (or just unlucky), we’re looking at implant failure, a revision surgery, or even an amputation.

My clients are a 78-year-old couple who live in a split-level house with lots of stairs. Their means of exercising Bruno is lobbing a tennis ball off the back deck, watching Bruno careen down the stairs, snag the tennis ball, and launch himself back up. He’s hardly ever walked on a leash and weighs almost as much either of his owners. Their closest family members live 45 minutes away and works full time.

The logistics of a safe TPLO recovery for a patient like Bruno, in a home like this, are effectively impossible.

If I’m doing my job right, that is to say I’m exercising the professional judgment my Pennsylvania veterinary license exists to certify, I’m talking to them about alternatives. Maybe we discuss a lateral suture technique, which is less invasive and has a shorter, less demanding recovery. Sure, the outcomes aren’t as good, but they’re better than nothing. Maybe we discuss their financial situation, and perhaps the cost of orthopedic surgery is prohibitive. Then maybe we talk about medical management and physical rehabilitation. Even lower success rates, but still better than nothing for Bruno.

I need to be allowed to have all of that conversation, not some of it. Medicine is invariably contextual. An exercise regimen that would result in effective fitness and weight control for me would probably mean death for an 70-year-old heart patient. The caloric restriction for one person to lose weight results in a fatal hepatopathy for another. The medication and dose that cures one disease exacerbates another. Professional judgment necessarily exists because the abstract “best practice” and the actual “best practice” are rarely the same thing.

Enter the state legislature, full of people who are, demonstrably and fortunately, not veterinarians. But one of those well-meaning legislators reads the same peer-reviewed literature I have and concludes, correctly, that the best way to address a torn cruciate in a canine is the tibial plateau leveling osteotomy. They’re a well-meaning legislator, and they love their dog, and they introduce and successfully pass a law stating that any licensed veterinarian who recommends anything besides a TPLO for a torn cruciate has engaged in “unprofessional conduct” and is subject to fines, probation, and loss of license.

The science is clear. The prevailing professional opinion is overwhelming. We must protect the public, and not allow veterinarians to practice substandard care, so it’s the TPLO or else.

And now the options for my little old clients are:

1. They pursue the surgery, on the advice of their veterinarian, despite the high cost and high risk to recovery.

2. Do nothing. They can’t afford the single option available to them.

3. Drive two miles to the next state without an enthusiastic and well-meaning legislature and get more appropriate advice from a different veterinarian.

Frankly, I’m rooting for Option Three, and “hoping they find advice in a state that allows doctors to practice medicine” is not a reassuring option.

But that’s effectively what Colorado did. They didn’t do it to veterinarians and it wasn’t about knee surgery, but the principle is the same. The state identified prevailing professional consensus, wrote it into law, passed it, and made it illegal for a licensed professional to have a conversation with any minor patient who wanted to move in a direction the state decided was impermissible.

And we note Justice Jackson’s vigorous dissent, that the speech is just incidental to the regulation of conduct, where viewpoint neutrality isn’t the touchstone, and the state - not the doctor or the client - gets to decide the side it thinks is correct. There’s nothing stopping any state legislature from doing the same thing to any medical professional in any field on any topic. Up to and including Bruno’s knee surgery.

Subscribe now

Eppur Si Muove

Everything, so far, rests on a premise I’ve left undefended, and one that the Court largely ignored: that the prevailing medical opinion is not a sufficiently reliable foundation upon which to build legislation. Not because medical consensus is wrong, it’s often right, but because science and law operate on fundamentally distinct and frequently incompatible epistemological architectures.

Science is provisionally, conditionally true. In The Structure of Scientific Revolutions, Thomas Kuhn argued that science doesn’t progress in smooth, predictable, linear accumulations of facts. It works within paradigms, those shared frameworks of assumptions that define what questions are worth asking and what answers count as valid. Much the time, we do “normal science,” solving puzzles inside the paradigm, refining what we already believe to be true. But anomalies accumulate, data that doesn’t fit, and, eventually, the weight of what the paradigm can’t explain triggers a crisis, then a revolution. The old framework collapses and a new one takes its place. Kuhn named it a paradigm shift, and was careful to point out that it’s not a failure of the system, it is the system. The epistemology is intentionally iterative.

The law is authoritatively, institutionally true. In the United States, that’s not an accident. If Kuhn explained why science must change its mind, James Madison taught us why the law must not. In Federalist No. 10, Madison warned that the greatest threat to a republic is the majority faction. That group, united by some common passion or interest, that captures the machinery of government and uses it to impose its will on everyone else. Madison’s solution was structural: a republic large enough and diverse enough that no single faction could easily assemble the power to dominate, with institutions designed so that changing the law requires sustained consensus across multiple bodies over time. The bicameralism, the presentment, the separation of powers? All of it exists because Madison understood that when a prevailing opinion commands a legislative majority, the minority needs something more durable than the majority's good intentions to protect it. Statutes don't say, "this is our best understanding," they say, "this is what is." Laws create fixed points, a binary of legal and illegal, that stay where they're put until a future legislature affirmatively undoes them. The epistemology is deliberately declarative.

The tension isn’t new either. In 1633, the same decade the colony of Virginia started regulating the practice of medicine, Galileo Galilei was at last forced by the Roman Inquisition to recant his heretical support of Copernican heliocentrism. On threat of execution, Galileo recanted, though, its said, after he finished recanting, he muttered, “eppur si muove.” Italian for, “and yet it moves.”8

The Earth did not care what the Church declared or about its interpretation of scripture. The Earth moved anyway. The evidence didn't wait for the institution to catch up. It took the Catholic Church until 1992, three hundred and fifty-nine years if you’re counting, to formally acknowledge that Galileo Galilei had been right.

And if you think such lag time is the exclusive purview of the Catholic Church, I’d like to refer you to the first edition of the Diagnostic and Statistical Manual of Mental Disorders, the DSM, published in 1952 by the American Psychiatric Association. The prevailing medical opinion at the time classified homosexuality as a “sociopathic personality disturbance”9 and was not the least bit controversial for doing so at the time. From 1952 until 1968, homosexuality remained, officially, a sociopathic personality disturbance.

In 1968, when DSM-II was published, homosexuality had be reclassified as a “sexual deviation.”

It wasn’t until 1974 that homosexuality was no longer classified as a mental disorder in the DSM.

As recently as 1989, six years after Sally Ride went to space and four years after the first .com domain was registered, the United States government was still blocking the entry of homosexuals into the country on the basis of their “mental illness.”

That’s a forty-year arc from “sociopathic personality disturbance” to “normal variation of human sexuality.” Forty years during which the profession argued, published, replicated, revised, and eventually changed its mind. The profession didn’t arrive at its current understanding because someone flipped a switch. It got there through a structural and institutional willingness to say, “we now know better than we did” That process, the willingness to be wrong, to look at the evidence again, to revise an opinion, is perhaps the single most powerful feature of medicine as a discipline. It is not a failure of the system, it is the system working exactly as designed.

And what if we apply this framework to the Colorado framework to that timeline?

If, in 1965, a state legislature had looked at the prevailing psychiatric consensus and passed a law requiring licensed therapists to treat homosexuality as a mental disorder, because that’s what the DSM said, because that’s what the professional organizations endorsed, because that’s what the science supported, that law would have had the full weight of medical authority behind it. It would have been, by medical standards, evidence-based. And any therapist who believed otherwise, who sat with a patient and said “I don’t think there’s anything wrong with you,” would have been in violation of the law. Not just bucking convention or risking a licensing board complaint, but breaking the law.

The professionals whose evolving views eventually changed the consensus, the researchers who challenged the underlying data, the clinicians who saw their patients clearly, the activists within the APA who pushed for declassification, would have been legally prohibited from practicing in accordance with their own professional judgment. The very speech that moved the profession forward would have been the speech the state had made illegal.

That’s what happens when you freeze a provisional truth into a declarative one. You break the thing that made it trustworthy in the first place. Not because the consensus is right or wrong today, but because medicine doesn’t progress on “probably.” It progresses by questioning, by testing, by allowing room for the possibility that what we believe now isn’t what we’ll believe in one, two, or ten years.

The mismatch isn’t about whether any particular opinion’s correctness, but rather it’s about what happens when it changes. Licensing boards are epistemologically compatible with science because they’re composed of practitioners, they update standards as the evidence evolves, they can discipline bad actors without freezing the entire profession in place. They can absorb new evidence without a legislative session or public majority. A statute is, in general, epistemologically incompatible with science because it lacks the mechanism for self-correction that makes science what it is.

Laws can’t do that and aren’t designed to do that. Laws are meant to stay where we put them. Laws require a legislature to convene, a bill to be introduced, a committee to hear it, a vote to be taken. The lag time between a scientific consensus shifting and a statute catching up can be measured in decades. And in the meantime, every professional who sees the evidence more clearly than the legislature did is practicing under threat of sanction for saying what they believe to be true.

Justice Gorsuch, in the majority opinion, put it this way: a prevailing standard of care “may reflect what most practitioners believe today, but it cannot mark the outer boundary of what they may say tomorrow.” And earlier: “Medical consensus is not static; it evolves and always has.”

He’s right. Medical opinion describes what most practitioners believe today, but it cannot define what they’re permitted to say tomorrow, or next year. And that principle doesn’t depend on what you think about conversion therapy, or homosexuality, or any particular clinical question. It depends on whether you believe that the practice of medicine should be governed by institutions capable of changing their minds or by statutes that cannot.

The moment you codify provisional knowledge into declarative law, you’ve taken the one of core and powerful aspects of medicine, its willingness to change, and made that feature illegal.

What This Article Still Isn’t

I’ve spent thousands of words talking about conversion therapy and I haven’t told you what I think of conversion therapy. And that is very deliberate.

It doesn’t matter what I think about conversion therapy, and it didn’t matter to the Justices of the Supreme Court either.

My opinion is about mechanisms and systems. It’s about how we govern the people we trust to take care of us. Licensing boards exist for a reason, composed of practitioners who understand the nuances of the field they regulate. A state veterinary board does not need a legislative session to update its understanding in and discipline professionals who harm patients.

That is a profoundly different instrument than a statute. A licensing board is a living document. A statute is a photograph. And a photograph of medicine, taken at any given moment, will always be at least partially wrong, because that’s how medicine works.

The 8-1 margin in Chiles v. Salazar suggests this isn’t a left-right issue, even though it will inevitably be covered as one. Justices Sotomayor and Kagan did not suddenly discover a shared judicial philosophy with Justices Thomas and Alito. What they share, or at least what all eight justices in the majority share, is the recognition that the government cannot appoint itself the arbiter of which side of a medical conversation is legally permissible. That principle drew support from across the Court’s ideological map.

Subscribe now

Why We’re Better Off

Colorado’s legislature looked at the prevailing medical consensus on conversion therapy, decided it was correct, and enacted it into law. They may well have been right about the medicine. Most of the major professional organizations agree with them.

The manner in which Colorado tried to accomplish its goal, however, crossed a line that violated the First Amendment rights of medical professionals.

The Court said so, and I think we’re better off for it.

Not because conversion therapy is good. Not because conversion therapy is bad. Not because any particular medical opinion is right or wrong. But because the freedom for professionals to exercise judgment, to be governed by expert bodies rather than political ones, and to revise their views as evidence develops is the engine that makes medicine improve. It’s the reason we no longer classify homosexuality as a sociopathic personality disturbance. It’s the reason we no longer bleed patients to balance their humors. It’s the reason veterinary orthopedics looks different today than it did twenty years ago and will look different again in twenty more.

A licensing board can absorb a paradigm shift in ways that a statute cannot. A licensing board can say, “the evidence has changed and so have we.” A statute says, “this is what the legislature believed in 2019, and it will remain the law until a future legislature believes otherwise.” The lag between those two moments, the moment when the science moves and the moment when the law catches up, is where patients get hurt. It’s where professionals get silenced. It’s where progress is frozen.

Galileo recanted under threat of execution. The Earth moved anyway. It took the Catholic Church 359 years to admit it. The APA needed forty. Colorado’s legislature needed seven, and a spanking from the Supreme Court.

The tools that shape our world are, largely, agnostic. The same constitutional architecture that was used to pathologize homosexuality for decades is the same one that eventually protected the rights of LGBTQ people to exist openly. We don’t get Obergefell v. Hodges without an apparatus protecting individual rights against state power. We don’t get Lawrence v. Texas without substantive due process. The same First Amendment protects a counselor with a BA in Biblical Studies and the activists and professionals who changed the DSM. If we dull the tools when we don’t like the direction they cut, those tools might not be there when you need something sharp yourself.

The lesson isn’t that consensus is reliable or unreliable. The lesson is that the best medical opinion is dynamic rather than fixed. The institutions we build to govern medicine need to be capable of honoring that distinction. We need the recognition that the law was designed to do something else entirely, and it does that thing well. But when you ask a tool built for permanence to govern a discipline built on revision, you get exactly what Colorado got: a well-intentioned statute that the Constitution wouldn’t tolerate.

Just because we believe it today doesn’t mean we should believe it forever. It’s one of the most powerful aspects in the advancement of medical knowledge and the justices who decided Chiles v. Salazar helped make sure we get to keep it.

An essay appeared last fall in Today’s Veterinary Business declaring that there exists a “Path to Greatness” for veterinarians in the age of AI.

It’s a pleasant enough piece, featuring a hypothetical new graduate named William who is given an omniscient AI mentor and must decide whether to lean on it or learn from it.1 The article concludes, with the confidence of a man who has found the trail and is waving you toward it, that greatness can only be reached in resisting the temptation of convenience. Use AI, but not too much. Take the stairs sometimes. Be a mountain climber, not a button-presser.

Before I go on to describe why I take exception to this article, I want to say that I’m genuinely glad that there are publications like TVB that publish conflicting points of view. It’s good for the profession that we’re allowed to disagree in discourse.

The first problem I have with the essay is the premise itself: that there is a path to greatness. Moreover, I take issue that a columnist and veterinarian, however well-intentioned and well-credentialed they are, however many podcast downloads and subscribers they have, can describe the path or define greatness for anyone else.

It’s a seductive idea, and it sells well. “If you pay me to do so, I shall reveal the little known path to greatness in a pleasing fashion!” But such messages are, at their core, false.

I wouldn’t consider this kind of message to be a lie, with connotations of malice or intent. In fact, I believe the author believes in the message he wrote, but the message is nonetheless not true. It’s akin to the freely given and benevolent advice of the aging practice owner, the one who bought his clinic in 1968, telling new graduates already facing hundreds of thousands of dollars in debt to “just go buy a practice” and build it like they did, because it worked for them, once.

This business model of generic advice is built on two equally erroneous premises: first, that greatness is a fixed destination with a known route rather than a quality that emerges, unpredictably and often painfully, from the particular, peculiar, and unrepeatable circumstances of your own life. Second, the notion that there exists a set of specific instructions you can follow to reach it.

Antonio Machado understood this more than a century ago: Se hace camino al andar, he wrote. You make the road by walking it. There is no path until you’ve walked it, and by the time you have, it’s yours and nobody else’s.

The veterinarian in rural Appalachia pulling a calf at three in the morning is not on the same path as the oncologist in Manhattan, and neither is on the same path as the fresh graduate trying to figure out how to tell a young couple that their golden retriever is suffering. And while neither the calf, the cancer, nor the couple cares if you used AI in the meantime, it’s more important to be cognizant that these are different lives, different walks, different versions of whatever greatness means. Nobody standing outside those lives gets to define it for you.

The second problem is one of intellectual rigor, or rather its absence. The essay leans heavily on what it calls “a MIT Media Lab study” to argue that AI use causes cognitive atrophy. This is presented as settled science, a terrifying finding that should alarm any “professional thinker.” What it does not mention is that the study in question, Kosmyna et al., “Your Brain on ChatGPT”, is an arXiv preprint.2

As a preprint, it has not passed peer-review. It has not been published in a scientific journal. Its sample size is fifty-four participants. The lead researcher herself has publicly cautioned against the kind of overblown interpretation that the essay enthusiastically provides, going so far as to ask journalists not to use words like “brain rot” or “damage” in describing her preliminary findings.

I don’t bring this up to dismiss the research, which asks important questions worth pursuing. I bring it up because if you’re going to tell veterinarians and veterinary students how to think about AI, you owe them the courtesy of doing so with actual peer-reviewed evidence rather than something short of it. Citing a preprint as though it were established fact, without noting its status or limitations, is not a minor omission in a publication aimed at professionals. It is a failure of the very intellectual discipline the essay claims to champion. If we are going to be the mountain climbers of the mind, then we should know the difference between a summit and a base camp.

There are a number of examples of how automation results in skill atrophy. But Kosmyna et al isn’t one of them, not yet. Noted examples from engineering, aviation, and radiology could have been used to make the point. But there’s an exquisite, and, to some, particularly delightful, irony in the founder of the Uncharted Veterinary Conference writing an article titled “The Path to Greatness” failing to use the example of well-documented negative impacts on spatial memory from the habitual use of GPS on self-guided navigation. That is the peer-reviewed evidence you’d want for that piece, the metaphor is elegant and self-evident.

The third problem, and the one that might matter most to students reading this, is the casual paternalism of it. The essay tells you that most people will choose “the path of least resistance” and that only “some of us” will opt for the harder, nobler road. The passage is “narrow and steep, and few will choose it.” This is not guidance so much as a sorting ceremony. It divides the profession into the enlightened few who resist AI’s siren call and the comfortable masses who succumb. It tells you what greatness looks like before you’ve had much of a chance to discover it for yourself.

Veterinary students are the audience of this piece, I’ve only met a few of them, but I’m certain every one deserves better than that.

Consider a different hypothetical: Dr. Hannah Jay graduated last spring with two hundred and twelve thousand dollars in debt and a quiet, persistent worry of being unready. On her first solo emergency, a dyspneic French bulldog late on a Tuesday night, she used an AI tool to double-check her radiographic interpretation, ran a differential she’d already been building in her head through VIN’s AI Mode, and caught a subtle pneumothorax she almost second-guessed herself out of seeing. Nobody told Dr. Jay that she’d taken the path of least resistance, that she should’ve used a textbook instead of a search engine, or that she ought to have done jumping jacks while reading the radiograph to increase the degree of difficulty. Nobody docked points from her greatness score. The patient walked out of the hospital later that week and went home.

Dr. Jay went home too, a little more confident. And then she came back the next day and did it again, continuing on her path. She did not become a lesser doctor for using every tool at her disposal to treat her patient. She became a better one; stronger and more capable in the ways you want your doctors to be: precise, effective, and unwilling to let pride stand between her patient and an answer.

Students and our younger colleagues deserve to be told the truth. And the truth is that nobody knows what a great veterinarian looks like in 2026 because greatness is not defined by a single rigid rubric for the more than 100,000 veterinarians in the United States or anywhere else. The tools are new. The pressures are new. The economics, the case complexity, the client expectations, the information landscape, and more: all of it is shifting under our feet. Anyone who tells you they have the map is no more helpful than an astronomer who believes the Earth to be fixed in place as the cosmos move about it.

Use AI or don’t. Use it a lot or a little. Read the preprints and the peer-reviewed papers and know the difference between them. Make your own mistakes, then make all new ones, follow your own curiosities, sit with your own discomfort in the exam room when a client asks you a question that has no right answer and no dataset to search. Find mentors and guidance, listen to their advice and insight, and disregard some of it. Discomfort is not a failure of preparation, it is not a fall from grace or greatness. It is simply part of the practice of medicine. It is the walk most of us will choose to walk. And while I maintain that there is no path, I also take care to remind you that you need not walk alone.

There is no path to greatness. There is only the walking, and no one can do it for you, and no one, no matter how large their platform or adorned their trophy case, gets to decide whether or not your walking counts as you make your own path.

Se hace camino al andar.

And it would seem that Ms. Halioua2 and her team are onto something. In February 2026, Loyal announced a $100 million Series C led by age13, with participation from Baillie Gifford, bringing total investment in the company to over $250 million. They’ve got a number of drugs in development to justify that funding.

LOY-001 targets large and giant breed dogs, which are known to have shorter lifespans. It’s a long-acting injection administered every three to six months by a veterinarian, targeting the overexpression of insulin-like growth factor 1 (IGF-1) believed to drive accelerated aging in large dogs. In 2023, the FDA’s Center for Veterinary Medicine accepted Loyal’s Reasonable Expectation of Effectiveness application for LOY-001. A first for any longevity drug, in any species.

LOY-002 is a daily, flavored tablet for senior dogs that targets metabolic dysfunction as an underlying mechanism of aging. In February 2025, the FDA also accepted the RXE for LOY-002. In December 2025, the FDA accepted the Target Animal Safety submission. That’s two of three major technical sections complete for Expanded Conditional Approval. The company’s STAY study enrolled over 1,300 dogs across 70 independent veterinary clinics and is expected to last four years. It is, according to Loyal, the largest clinical trial in the history of veterinary medicine by enrollment. The company describes the drug as a caloric restriction mimetic.

LOY-003 targets large dogs via a daily oral pill. Loyal identified it as CRN019414 in a blog post announcing their partnership with Crinetics Pharmaceuticals. It’s a nonpeptide somatostatin receptor type 2 agonist, targeting the same GH/IGF-1 pathway as LOY-001, different delivery vehicle.

Interestingly, they’ve never disclosed the active pharmaceutical ingredient for either LOY-001 or LOY-002. Even the molecular targets have been kept somewhat vague. LOY-003 is the only compound they’ve named.

They raised nine figures, now a quarter of a billion dollars, without publicly revealing the drug and only hinting at the mechanism of action? For somebody like me, the opacity raises the eyebrows a bit. Which is to say that it sparks a dopamine-rich hyperfocus for being a fun puzzle to solve.

Subscribe now

Disclaimer

I’ve spent some time trying to uncover or figure out the nature and identity - or approximations thereof - of Loyal’s LOY-001 and LOY-002.

I have no insider knowledge, I have no financial interest, and no connection besides a free pair of socks they gave away at a veterinary conference.5 What I do have is intense curiosity, a veterinary degree, an internet connection, and dubious self-regulatory mechanisms.

But I want to make it absolutely clear at the outset that everything that follows is speculation. I own a small animal hospital in Chadds Ford, Pennsylvania. I have no consulting relationship with the company, I have no access to proprietary data, and, other than Dr. Brennan McKenzie, whose reputation in veterinary medicine precedes his connection with Loyal, I doubt that I could reliably pick any member of their leadership team out of a lineup.

This article is an intellectual exercise. It is a statement of opinion, not fact, though I’ve tried to base it on a collection of facts. I’m just a nerd who went on a research bender, trying to assemble the pharmacological puzzle from publicly available sources like patent filings, peer-reviewed literature, FDA regulatory documents, the company’s own press releases and published research, and the leadership’s public writings and interviews.

I am very much not claiming to know what LOY-002 contains. I’m reasoning, in print, towards a hypothesis.

Loyal has stated that they cannot disclose the active ingredient prior to FDA conditional approval for intellectual property reasons. I respect that decision. On the outside chance I’m right6 about what follows, it changes nothing about the regulatory process or clinical evidence they’re generating.

Nothing in this article constitutes or should be interpreted to constitute medical advice, investment advice, or a recommendation to prescribe, compound, or administer any medication on- or off-label to any animal.

There, now we’ve finished the CYA bit. So down the rabbit hole we go.

LOY-001: Quick and Easy

I’m going to focus on LOY-002, since it poses the more interesting mystery. LOY-003 has been disclosed for years. LOY-001 is strongly suggested to be a somatostatin-pathway drug. Loyal has confirmed it targets the same pathway as LOY-003, which is CRN01941, a somatostatin receptor agonist.

I think LOY-001 is something near to octreotide, along with a wrinkle that makes the dosing extremely long-acting and/or possibly a novel formulation or depot technology. It might be an antisense oligonucleotide that knocks down IGF-1 directly. The three-to-six-month dosing interval probably means it’s not plain octreotide.

See? Done. This could’ve been a much shorter essay.

Loyal CEO, Celine Halioua

I didn’t set out to profile the company’s CEO, but a substantial amount of communications about the company come directly from her personal pages. And I wanted to include a few paragraphs about her because it proved more important, and interesting, than I’d anticipated.

CEO Celine Halioua7 is precise, deliberate, and, for the most part, candid in her public persona. She shares company milestones alongside deeply personal disclosures on Twitter/X. She talks about her OCD medication, her rescued dog’s abuse history, and her Series B funding failures.8 And it’s obviously meant to work as a marketing ploy, but, looking closely, I’d have a hard time condemning her for it.

She says she’s spent hundreds of hours studying Silicon Valley communication strategy at what amounts to a mechanical level and she openly admits to engineering her public persona. That might give some folks pause, but I think that’s just what a good leader does in that position. My skin crawls at the leaders who try to make the world believe that they’re naturals. That the facility they have in something was the result of a deity, cosmic beneficence, or a magic wand. That strategy is nothing more than fairy-tales-as-emetics for me.

Ms. Halioua, by contrast and to her credit, is candid about the effort, revealing what I’ll call a precisely measured amount of genuine humility. She studied strategies and mechanisms in order to develop a skill crucial for her job? She learned and practiced to get better at something? How. Dare. She.

Moreover, she writes well, and I say that as a person who spends a lot of time trying to write well. I’m admittedly inclined to overweight the value of that, but being intentional about what you put in print is something for which I both endeavor and advocate. She’s written “I write to synthesize my thoughts” and “don’t build boring shit” in her blog posts, which cover monoclonal antibodies, personal trauma, communications, and startup operations. I’m not sure I could recognize Ms. Halioua if she sat next to me on a park bench, but in the writing I see a familiar mind: intense, curious, and purposeful.

This is all to say that after reading a lot of what this person has written, I’ve come to respect them a great deal.

The Pharmacological Fingerprint of LOY-002

LOY-002 is where things get interesting, because Loyal has simultaneously told us a great deal and almost nothing. What they have disclosed, across interviews, blog posts, press releases, and published research, creates a delightfully tight set of pharmacological constraints.9

Think of it as a game of “Guess Who?” or one of those logic puzzles you might’ve done as a kid. What your nerdier friends may refer to as a constraint satisfaction problem. Except instead of being given a list of statements, you get to compile them yourself.

Here’s what I’ve found:

The drug is a caloric restriction mimetic. Ms. Halioua has described it as “delivering the benefits of caloric restriction through a pharmaceutical drug, without weight loss or appetite suppression.” She’s also said, in a tweet responding to questions about GLP-1 agonists, that “you can hit further downstream the caloric restriction pathway to get the metabolic aging benefits without appetite suppression - that’s our focus for now.”

The word downstream is helpful in those sentences. It say the drug isn’t acting on appetite or food intake, it’s acting on whatever the cell does after caloric restriction signals arrive. That’s more than marketing copy, it’s an insight into the mechanism.

Ms. Halioua has explicitly ruled out rapamycin and metformin. Those are the two most prominent longevity candidates in public discourse. Crossing them off narrows the field considerably.

She’s also written that, “to maximize benefit, aging drugs need to become boring.”

LOY-002 has been described as targeting “a mechanism that, in mice and humans, has shown notable improvement in cognitive function and all-cause mortality.” While this is not a direct quote, it comes from an interview published in a legitimate trade publication and it is a precise statement. And it’s one on which I’m going to lean, because it not only requires that LOY-002 has shown improvement in cognitive function and all-cause mortality, but also that the compound has rodent data and human data. It’s not necessarily from aging-specific trials, but from somewhere in the published literature.

LOY-002 uses “a slightly different mechanism” from LOY-001 and LOY-003. Since both of those target the GH/IGF-1 axis (LOY-001 via injection, LOY-003 via oral somatostatin receptor agonism), this definitively separates LOY-002 from the somatostatin/GH/IGF-1 pathway and, I think, places it in the metabolic and energy-sensing domain.

In a dvm360 interview, Ms. Halioua characterized LOY-002 as “a little bit more of a reversal mechanism. It’s not going to make a 10-year-old dog a puppy, but we’ve seen a rescuing in age-related functional decline in dogs that have been treated with this drug.”

And the safety data: no clinically significant adverse events at 1×, 3×, and 5× dosing in a standard Target Animal Safety study, plus field safety data from over 400 senior companion dogs in the STAY trial. The FDA accepted this package in December 2025. That’s an unusually clean safety profile for a chronic-use drug.

The drug is formulated as a daily beef-flavored oral tablet for dogs 10 years and older, weighing at least 14 pounds. It is, per the company’s description, intended for virtually the entire senior dog population aside from the very smallest patients.

So here’s the constraint set. Whatever LOY-002 is, it must be:

1. A caloric restriction mimetic

   1. Stated

2. Active downstream in the CR pathway, not via appetite or intake

   1. Stated

3. Not rapamycin, not metformin

   1. Stated

4. Not a somatostatin analog or GH/IGF-1 axis drug, a different mechanism from LOY-001/003

   1. Stated

5. Supported by cognitive function data, in rodents and/or humans

   1. Inferred secondhand, stated by journalist in interview

6. Supported by all-cause mortality data in humans

   1. Stated

7. Safe at five times the intended dose in dogs, with no clinically significant adverse events in 400+ dogs

   1. Stated

8. An oral tablet, likely generic or off-patent

   1. Stated, and consistent with the company’s IP strategy

9. Not causing weight loss or appetite suppression

   1. Stated

That is actually a pretty narrow pharmacological window. And it only took reviewing a few dozen blog posts, tweets, interviews, and press releases to get there.

The PPAR Trail

In February 2026, coinciding with Loyal’s Series C announcement, Rapamycin Longevity News published a post stating that “industry analysis of the drug’s physiological effects points to the Peroxisome Proliferator-Activated Receptor (PPAR) pathway, specifically involving PPAR-γ modulation or Selective PPARα Modulator (SPPARMα) activity.” This fairly briefly article helped me to narrow the search considerably.

I’d describe that article as industry analysis, not insider disclosure. Their post is at least as speculative as my article. But it does align remarkably well with the pharmacological evidence, and it narrows our search to a specific family of nuclear receptor agonists.

Peroxisome proliferator-activated receptors (PPARs) are nuclear transcription factors that function as lipid sensors and metabolic regulators. There are three isoforms: PPARα (predominantly expressed in liver, heart, kidney, and muscle; governs fatty acid oxidation), PPARγ (adipose tissue, immune cells; governs insulin sensitivity and inflammation), and PPARδ/β (broadly expressed; governs fatty acid catabolism and energy homeostasis). They are master regulators of metabolic gene expression.

Think of these three as three regional managers for a company that runs fuel logistics.

PPARα is the operations manager at the power plant (liver, heart, muscle). When fuel shipments arrive (fatty acids), the PPARα decides how fast to burn them. High demand? Light the furnaces. Low supply? Burn the reserve fuel stocks.

The PPARγ handles warehousing and human resources. How much fuel to store, where to put it, and keep the immune cell workforce from striking or rioting. When this is working, your inventory is organized and labor relations are smooth; that’d be your insulin sensitivity and inflammation control.

PPARδ/β is the roving efficiency consultant who works across every department. They’re less specialized, but everywhere at once. They’re tweaking fuel usage and make sure no division is wasting fuel or running a deficit.

The connection between PPARα and caloric restriction (CR) is not speculative. It is established in the peer-reviewed aging biology literature. A foundational finding, one that anyone doing serious work in this space would know, is that approximately 19% of all hepatic gene expression changes induced by caloric restriction are PPARα-dependent. In PPARα-null mice, those CR-mediated metabolic adaptations are abolished entirely. Caloric restriction upregulates PPARα expression in heart and liver tissue, increasing lipid utilization and enhancing fatty acid β-oxidation. A 2020 review in Cells stated the relationship explicitly: “Ligands of PPARα have been proposed to serve as mimetics of CR.”

That’s not “may be associated with,” that’s “a proposed class of caloric restriction mimetics.”

And Loyal’s own published research keeps pointing at the same metabolic targets. McKenzie et al. (2025) demonstrated that insulin increases, adiponectin decreases, and free fatty acid accumulation all correlate with frailty in 451 companion dogs. And explicitly noted that adiponectin signals through PPARα to increase fatty acid oxidation and reduce inflammation. Peloquin et al. (2024) showed that saturated fatty acid concentrations predict insulin sensitivity in dogs. McKenzie et al. (2024) demonstrated that a high-fat diet induces metabolic changes mimicking natural aging.

These papers all highlight biomarkers like hyperinsulinemia, dyslipidemia, elevated free fatty acids, decreased adiponectin. And those biomarkers all squarely in the therapeutic wheelhouse of a PPARα agonist.

Where the Path Led: Fenofibrate Esters

If you’re a veterinarian reading this, you’ve might’ve already guessed where I’m going. If you’re not, let me introduce you to a drug that’s been sitting quietly on pharmacy shelves since the 1970s.

Fenofibrate is a fibric acid derivative with fenofibric acid as its active metabolite. It’s a PPARα agonist, the most widely prescribed one in human medicine. Fenofibrate is used to treat hypertriglyceridemia and mixed dyslipidemia, typically as adjunctive therapy with statins. It’s been generic since 2005, available from over 30 manufacturers worldwide, in tablets and capsules ranging from 35 to 160 mg.

It is also already used, off-label, in veterinary medicine. It is a drug with which the veterinary community has experience, on which the FDA has decades of safety data, and that VCA Animal Hospitals lists as a standard veterinary medication for canine hyperlipidemia.

Fenofibrate is the most commercially available and familiar delivery vehicle among multiple possible prodrugs. Choline fenofibrate already received its own FDA approval as a distinct product. Lv et al. (2020) synthesized seven fenofibric acid ester prodrugs (JF-1 through JF-7).

All of the evidence I'll present below applies to fenofibric acid as the active metabolite. Most of the published data comes from studies using fenofibrate because that's the prodrug with forty years of clinical history. But the pharmacological logic of the CR mimetic activity, the cognitive data, the mortality signal, and the canine safety profile is driven by fenofibric acid at the PPARα receptor, not by which ester shell delivers it. Whether LOY-002 is fenofibrate itself, a novel ester with superior bioavailability, or something else in the fenofibric acid family, the constraint analysis points to the same place.

Criterion #1: Caloric restriction mimetic

Technical: PPARα ligands, including fenofibrate, have been explicitly proposed as CR mimetics in reviews. The mechanistic overlap with caloric restriction is well-established at the transcriptomic level: mediated through shared AMPK, SIRT1, and mTOR signaling cascades. A separate patent filing (US20160186257, unrelated to Loyal) identified bezafibrate, a pan-PPAR agonist in the same fibrate class, as the most effective CR mimetic in comprehensive gene expression analysis.

Translation: Fenofibrate tricks the body into thinking it’s on a diet, at the cellular level, without actually reducing food intake. The molecular machinery it activates overlaps so heavily with what caloric restriction activates that researchers have literally published papers calling drugs in this class “caloric restriction mimetics.” A different patent, unrelated to Loyal, even identified a close chemical cousin of fenofibrate as the single best pharmaceutical match for the gene expression signature of caloric restriction.

Criterion #2: Downstream mechanism, no appetite suppression

Technical: Fenofibrate acts at the nuclear receptor level, modulating gene expression in hepatocytes and other target tissues. No mention of the appetite circuits. It doesn’t suppress food intake. It activates the same downstream metabolism that caloric restriction activates (namely fatty acid β-oxidation, ketogenesis, insulin sensitization) without the upstream signal of reduced caloric intake.

Translation: Fenofibrate works inside the cell, flipping metabolic switches in the liver and other organs, not in the brain's hunger center. It never tells the dog to stop eating. Instead, it goes straight to the cellular machinery that caloric restriction eventually activates anyway, burning fat, making ketones, improving insulin response, by direct mechanism. All of the metabolic benefits, none of the begging at an empty bowl.

Criterion #3: Not rapamycin, not metformin

Technical: It is neither of those things. I am offering no citations for this.

Translation: Duh.

Criterion #4: Different mechanism from LOY-001/003

Fenofibrate operates through PPARα-mediated transcriptional regulation. It has nothing to do with somatostatin receptors or direct IGF-1 suppression. It does, however, attenuate IGF-1 receptor signaling downstream, cancer biology studies have shown it suppresses IGF-1R-mediated cellular responses through inhibition of IRS-1, ERK, Akt, and GSK3β phosphorylation. This is a mechanistically distinct pathway from the direct IGF-1 level reduction that LOY-001 and LOY-003 pursue. Different mechanism but with similarly broad relevance to aging.

Translation: Fenofibrate doesn't work anything like Loyal's other two drugs. LOY-001 and LOY-003 reduce the amount of a growth hormone signal (IGF-1) circulating in the body, effectively turning down the output at the source. Fenofibrate leaves the signal alone but makes the cells less reactive to it, like putting earplugs on the listener instead of muting the speaker. Same net effect on aging pathways, completely different mechanism, which is more or less exactly what Loyal said LOY-002 would be.

Criterion #5: Cognitive function data in rodents

If the fenofibric acid hypothesis is correct, we would expect to find evidence of the cognitive benefit in published literature. And we do.

At least four rodent studies demonstrate fenofibrate’s cognitive benefits. It prevents post-ischemic memory impairment (dose-dependently reducing infarct volume), preserves hippocampal neurogenesis, reverses high-fat-diet-induced cognitive deficits via the PGC-1α/irisin/BDNF pathway, and improves post-operative cognitive dysfunction after isoflurane anesthesia. The Alzheimer’s Drug Discovery Foundation’s Cognitive Vitality report attributes fenofibrate’s neuroprotective properties to anti-inflammatory, antioxidant, and vascular barrier protective effects through PPARα activation.

An interview with Ms. Halioua published in a trade publication characterized LOY-002 as targeting a mechanism with demonstrated improvement in cognitive function and all-cause mortality in mice and humans. That phrasing is the journalist's, not a direct quote, so I'm not treating it as a hard constraint. But it is consistent with what the PPARα mechanism would predict.

Translation: Fenofibrate doesn't just help metabolism, it protects the brain. At least four studies in mice and rats show it prevents memory loss after stroke, keeps the hippocampus generating new neurons, reverses the cognitive fog caused by high-fat diets, and sharpens thinking after anesthesia. The Alzheimer's Drug Discovery Foundation has specifically flagged it as neuroprotective.

Criterion #6: All-cause mortality data in humans

A Korean cohort study of 10,114 diabetics found fenofibrate use associated with all-cause mortality with a hazard ratio of 0.437. That’s a 56% relative risk reduction, with longer use showing even stronger effects. A separate Korean general-population study of 555,672 individuals with high triglycerides showed all-cause mortality HR of 0.826 for fenofibrate add-on therapy, dropping to 0.618 with a year or more of use. Post-hoc analysis of the ACCORD Lipid trial found significant mortality reduction (HR 0.57) in early-diagnosed hyperlipidemia. The ACCORDION legacy study showed sustained benefit (adjusted HR 0.65, P = 0.02). A Korean national cohort of 110,723 matched pairs found fenofibrate users had HR 0.716 for all-cause death.

Translation: people who take fenofibrate die less, and the studies are considerable. Across multiple large human populations, tens of thousands to over half a million patients, fenofibrate users had significantly lower rates of death from all causes. The biggest signal showed a 56% reduction in mortality among diabetics. The consistent finding across every study is the same: take it longer, do better. These aren't just rodent models anymore. These are real humans, in real healthcare systems, over years.

Criterion 7: Safety in dogs at up to 5× dosing

This is where fenofibrate separates from the field.

Miceli et al. (2021) published a prospective randomized trial of 124 dogs with severe hypertriglyceridemia, treated at 10 mg/kg once daily. The result: 85.9% triglyceride normalization. Adverse effects: 3% experienced mild diarrhea. No significant adverse effects. Munro et al. (2021) at UC Davis studied micronized fenofibrate in 10 hyperlipidemic dogs over a dose-escalation protocol. Quiet demeanor and firm stools in one dog were the only reported adverse reactions. Crakes et al. (2021) demonstrated safety in diabetic dogs with maintained body weight and improved gut barrier function. O’Kell et al. (2025) did the same with naturally occurring diabetes mellitus. The fatal toxicity dose in dogs exceeds 4,000 mg/kg, providing an enormous therapeutic margin.

Loyal’s product brief states the active ingredient has “extensive” prior safety data across studies lasting four weeks to one year in over 400 dogs, with no clinically significant adverse events at 1×, 3×, and 5× expected doses.

Those study durations (four weeks, thirteen weeks, one year) are the standard regulatory toxicology study intervals required for original FDA drug approval in any species. Which means that, whatever this drug is, it went through the full gauntlet of FDA-mandated preclinical toxicology in dogs when it was first approved for human use. Fenofibrate, as a human drug approved in the 1970s, would have required exactly these studies. Those datasets already exist. They would constitute the kind of “extensive evidence” that a startup could reference, even a well-funded one, without having to generate all of it de novo.

I’ll talk more about the alternatives later, but this is where fenofibrate separates from the pack.

Translation: Fenofibrate is almost comically safe in dogs. Multiple studies, over a hundred dogs in one trial alone, show it works beautifully for its intended metabolic effects with virtually no side effects. The worst reported reaction across all the studies was one dog who got quiet and had poops. The lethal dose is so astronomically high it's practically irrelevant. Moreover, because fenofibrate was originally approved for humans in the 1970s, the FDA already required extensive safety testing in dogs as part of that process. Those studies, four weeks, thirteen weeks, one year, already exist in a filing cabinet somewhere. A startup with good lawyers wouldn't need to reinvent that wheel. They'd just need to patent a new use for it.

Criterion 8: Oral tablet, generic, off-patent

Fenofibrate is available as an oral tablet. It’s been off-patent for over two decades, and it’s manufactured by more than 30 generic companies worldwide. This makes it ideal for a method-of-use patent strategy: you patent the novel veterinary indication (canine lifespan extension) rather than the compound itself.

Translation: You can't patent fenofibrate, that ship sailed in the 1900s. But you absolutely can patent a new use for it, like "this old cholesterol drug extends dogs' lives." That's your intellectual property claim. And it explains the secrecy reasonably well: the moment you tell the world what's in the pill, every veterinarian in the country can prescribe it off-label before your patent is even enforceable. If you've raised a quarter billion dollars, you probably want to keep your mouth shut until the paperwork is locked down.

Criterion 9: No weight loss or appetite suppression

This is, I think, my hypothesis’s weakest point.

In obese beagles, fenofibrate caused a small (3%) but statistically significant weight loss over 15 days that reversed upon cessation. In rodent models, higher doses clearly reduce food intake and body weight. However, in diabetic dogs at ideal body condition, “all dogs maintained their body weight and ideal body condition score” during treatment.

The resolution to this contradiction is likely dose-dependent. LOY-002 is intended for metabolic modulation in aging, not aggressive lipid-lowering in hyperlipidemic patients. A lower dose optimized for metabolic signaling rather than triglyceride normalization might produce clinically negligible weight effects. Loyal’s claim may be precisely calibrated to the dose they’re using.

Translation: Fenofibrate can cause minor weight loss at higher doses, obese beagles lost about 3% in one study, and rodents on big doses clearly ate less. But dogs at healthy weight held steady. The likely explanation is straightforward: crank up the dose and you get weight effects; dial it down to the sweet spot for anti-aging metabolic tuning rather than aggressive cholesterol-lowering, and the weight needle probably doesn't move. Loyal would know exactly where that line is. They've had 1,300 dogs to figure it out.

The Paper Trail

Cellular Longevity Inc. was granted U.S. Patent Application 18/621,541 on December 10, 2024, titled “Method of increasing lifespan in mammals.” It was filed under Track One prioritized examination. The full claims remain inaccessible through free patent databases—likely due to indexing lag for recently granted patents, or possibly because Loyal exploited non-publication options to keep the filings confidential until grant.

CBInsights’ automated topic classification reveals five keywords assigned to this patent: Fatty acids. Prodrugs. Diabetes. Vegetable oils. Nutrition. Now, that’s probably just a piece of machine learning software10 assigning keywords based on patent data and metadata. But it’s remarkable how well it fits.

Now consider the following:

1. Fenofibrate is a textbook prodrug. It has zero pharmacologic activity in its administered form. It is completely hydrolyzed by tissue and plasma esterases to its active metabolite, fenofibric acid. No unchanged fenofibrate is detectable in plasma. The “prodrugs” tag alone eliminates pioglitazone (active parent compound), acarbose (not a prodrug), octreotide (not a prodrug), and most other candidates.

2. Fenofibric acid is a fibric acid derivative. Literally a propanoic acid ester, a fatty acid. The “fatty acids” tag matches the drug’s chemical classification, its mechanism (PPARα-mediated fatty acid β-oxidation), and the metabolic pathway it modulates.

3. Fenofibrate esters are extremely lipophilic. Fenofibrate’s log P is 5.24. Its aqueous solubility is 0.263 µg/mL, which is essentially insoluble. Commercial fenofibrate formulations require lipid-based excipients to achieve adequate bioavailability using things like soybean oil, soya lecithin, and related vegetable-derived carriers are already standard in existing formulations. The “vegetable oils” tag reads like an excipient disclosure for a lipid-based oral delivery system. For a beef-flavored daily tablet, you’d need this kind of formulation work.

4. Diabetes fits fenofibrate’s established clinical role managing diabetic dyslipidemia.

5. Nutrition aligns with its labeling as adjunct to dietary therapy and Loyal’s framing of LOY-002 as a pharmaceutical alternative to caloric restriction.

Three total patents have been filed by Cellular Longevity Inc. The top topics across all filings, diabetes, companion dogs, prodrugs, all seem reinforce the fenofibrate hypothesis rather than diluting it.

The Elimination Round

This wouldn’t be worth much if I only argued for fenofibrate. This will review the reasons why I concluded on fenofibrate and not something else.

1. Octreotide and other somatostatin analogs are almost certainly LOY-001’s mechanism and definitively LOY-003’s (Crinetics’ CRN01941). Having LOY-002 also be a somatostatin analog would be redundant and directly contradicts Ms. Halioua’s statement about a “slightly different mechanism.” A 2005 study by Robben et al. found octreotide didn’t suppress basal GH levels in healthy fasting dogs—undermining the core rationale. Chronic somatostatin analogs carry a 10–63% incidence of gallstones and 57–68% GI adverse event rates in oral formulations (Mycapssa Phase 3 data). That’s incompatible with “no clinically significant adverse events” at 5× dosing. And oral peptide octreotide requires specialized liquid-filled capsule technology with approximately 0.7% bioavailability, fasting administration, and doses 200× the injectable equivalent. Good luck making that into an affordable beef-flavored tablet.

2. Acarbose deserves more respect than it usually gets in these conversations. It’s the only drug to show robust lifespan extension in the NIA Interventions Testing Program—16–22% median lifespan increase in male mice across multiple sites. It’s an oral tablet, generic, off-patent, and already used in veterinary medicine for diabetic dogs. Those are real credentials.

   1. But acarbose is not a PPAR agonist. It has no “Prodrugs” or “Fatty acids” patent tag relevance. Its cognitive function data is indirect and limited to combination studies and hepatic encephalopathy patients rather than direct neuroprotection. And its GI side effect profile in dogs is a pharmacological inevitability of its mechanism: block carbohydrate absorption, and undigested carbohydrate reaches the colon, where it ferments. Nelson et al. (2000) found that 60% of dogs developed semisoft to watery feces. You cannot dose-optimize your way around the fundamental mechanism of action. Loyal’s “no clinically significant adverse events at 5× dosing in 400+ dogs” is very hard to reconcile with a drug that gives three in five dogs loose stool at therapeutic doses.

3. Pioglitazone targets PPARγ rather than PPARα, is not a prodrug, caused cardiac hypertrophy and thoracic effusion in dogs during preclinical safety testing, and carries a black box warning for congestive heart failure in humans. It also commonly causes weight gain—which is the wrong direction for a longevity intervention. It’s currently in the NIA ITP 2022 cohort, with results pending. If those results are spectacular, I’ll revisit. Until then: no.

4. Pemafibrate is the SPPARMα specifically named in the Rapamycin News analysis. It’s an elegant molecule, but it has no published dog safety data whatsoever, is approved only in Japan, and its landmark PROMINENT cardiovascular outcomes trial—the one designed to validate the SPPARMα concept in human clinical endpoints—failed its primary endpoint. That’s not a death blow for the molecule, but it means there’s no published mortality or cognitive benefit data to match Loyal’s stated criteria.

5. Bezafibrate is a pan-PPAR agonist in the same fibrate class and was actually identified as the most effective CR mimetic in comprehensive gene expression analysis. It’s a credible molecule. But it has less canine safety data than fenofibrate and isn’t marketed in the United States, making it fit the marketing claims less precisely. For a company planning to launch a commercial veterinary product through U.S. distribution channels, sourcing a non-U.S.-marketed API adds unnecessary complexity.

Of Course, I Could Be Wrong

A constraint satisfaction problem only works if I steelman my own arguments.

1. The lifespan gap. The strongest evidence against fenofibrate is the absence of formal lifespan extension data in normal aging mice. Fenofibrate has never been tested in the NIA Interventions Testing Program. No published study demonstrates extended median lifespan in healthy wild-type mice. If Ms. Halioua’s interview-but-not-quoted reference to improvement in all-cause mortality in mice refers to a formal longevity study, this would need to be Loyal’s proprietary data, which weakens the hypothesis by requiring an assumption of unpublished evidence.

   1. Alternatively, the “mice” reference may encompass disease-model mortality reductions; fenofibrate reduced mortality by 80% in a Japanese encephalitis model and extended survival in cachexia models. The language is ambiguous enough to accommodate either interpretation.

2. The weight question. I addressed this above, but it remains a genuine vulnerability. If any well-powered canine study shows significant weight loss at LOY-002’s dose, the hypothesis has a problem.

3. The secrecy question. If LOY-002 is just fenofibrate, a generic drug any veterinarian could prescribe off-label tomorrow, then why the elaborate secrecy? Two possible answers.

   1. First, the method-of-use patent strategy requires maintaining confidentiality until the patent is granted and enforceable; disclosing the compound prematurely could jeopardize the IP protection that justifies $250 million in investment.

   2. Second, Loyal may be using a proprietary fenofibrate formulation. A novel prodrug variant, a lipid-based delivery optimization, a specific dose-finding innovation that isn’t plain fenofibrate-in-a-tablet. The patent’s “prodrugs” and “vegetable oils” tags are consistent with this conclusion as well. The answer might not be “fenofibrate” so much as “fenofibrate with a twist.”

4. The possibility that I’m just wrong. The compound could be something else. It could be a genuinely novel molecule, a repurposed drug outside my analysis, a combination formulation, or just something I missed. Loyal has smart people, significant capital, and access to compound libraries and research I don’t have. Honesty - and humility - mandate acknowledging that the puzzle might have pieces I can’t see, solve, or discern.

So What?

The embarrassing part of this story is that, until recently, I didn’t think very much of what the folks at Loyal (Loyalists?) had built.

I don’t think I was quite as bad as the some of the anonymous bloggers whose criticisms largely amounted to “this hasn’t been proven yet” (duh) and “this is only good if it improves quality of life” (also duh, but it ignores the fact that these things aren’t accomplished and demonstrated in a single gust of effort). Still, I found the caginess and opacity annoying at best and suspicious at worst. I thought it was a gimmick, intended to add an air of mystique and sophistication to act as a kind of capital yeast11 for a mediocre or even unworkable idea. I was even a bit cynical about Ms. Halioua’s Entrepreneurial Spirit Award at the Feather in Her Cap event this year. “They’ve just raised a lot of money and published a few studies,” I thought, “so what?” The whole thing felt a bit too much like Theranos to me.

But my suspicions have resolved to garden variety skepticism. I’ve spent a few hours really digging into this, and, well, I don’t think it’s a gimmick anymore. Reading more of what Ms. Halioua has written, I’m increasingly impressed with the “safe, well understood, cheap, small molecule generics” approach12. I think the Loyalists might legitimately be onto something, and it’s possible that something is a 50-year-old generic medication that’s been sitting in pharmacies since the Ford administration.

It’s less anticlimactic than it sounds.

The author Terry Pratchett, in The Science of Discworld, wrote that, “science is not about building a body of known ‘facts.’ It is a method for asking awkward questions and subjecting them to a good reality-check, thus avoiding the human tendency to believe whatever makes us feel good.”

The drug may have been there for half a century, waiting for someone to ask an awkward question.

It means Loyal’s contribution isn’t the molecule but rather the novel application of the molecule. It’s the regulatory pathway, the clinical trial infrastructure, the 1,300-dog STAY study, the proof that you can take “aging” to the FDA as a treatable condition and get them to agree. If Loyal has something real, the process is the triumph. A novel paradigm rather than a novel pill.

And it means that veterinarians like are going to need to learn a new vocabulary. We maybe knew fenofibrate for off-label use after lipid panels and pancreatitis risks. We don’t know it as a longevity drug. We don’t think about PPARα activation as a caloric restriction mimetic, because until this decade it wasn’t. And you’d have to be deep in the research to find it. We don’t prescribe medications to slow aging, because until now, there hasn’t been an medication approved for doing so.

If it works, it’s going to change things in the hospital.

There’s a twelve-year-old intact male Labrador retriever patient named Cody who still thinks he’s a puppy, except on the mornings when he can’t get up from the floor without help and NSAIDs.

There’s a fourteen-year-old spayed female mix named Sadie who’s losing the plot a bit, circling, staring at walls, forgetting where she is.

There’s an eleven-year-old neutered male Beagle named Maple13 whose bloodwork is slowly, predictably drifting into that metabolic no-man’s-land of insulin resistance and rising triglycerides that precedes the avalanche of age-related disease.

I think about them when I read about LOY-002 and imagine what it could be. I think about what it would mean to have one more thing that could help buy George a few more mornings of getting up, that could keep Sadie oriented in her own house for another year, that could flatten Maple’s metabolic trajectory before it steepens to an asymptote I cannot treat.

I’m as skeptical, impatient, and unrelenting as you’d expect from a man who has spent most of his waking hours for the last decade standing between his patients and the pathologies that come for them.

Sometimes I fight to win, sometimes I fight for time. I fight every day, in every exam room, for every graying muzzle. When I fight for time, sometimes I win for awhile, but I will always eventually and inevitably lose.

This has been an intellectual exercise, but it hasn’t just been an intellectual exercise. Sure, it was a mad hunt by a man careening after an unknown to satisfy his own curiosity. But the curiosity is fueled by more than professional interest, intellectual novelty, and a congenital predisposition to follow the dopamine.

I have thousands of patients, and it is quite likely that I will outlive every single one of them. These patients are the pets my family and friends and neighbors dearly love. They are my own as well. My hypothesis may yet be proven false, but my raisons d’etre are tangible and irrefutable. I want to help my patients, the ones loved by the people in my community, to live longer and to live better.

Maybe it fenofibrate, maybe it’s some other PPAR agonist, and maybe it’s something else entirely. But if there’s a pill that helps my patients to live longer and to live better, I want to know what’s in it.

The contents vary from terrible ideas to brilliant ones, middle school math homework practice with my stepson, drawings of Mickey Mouse for the toddler, meeting notes, outlines, and anything I could imagine for the past few years. I even uploaded PDFs of crossword puzzles for fun. And the notes vary in levels of organization and coherence ranging from “Swiss Metro Schedule” to “post-unblocking dysphoria of feral feline.”

The problem, of course, is that handwritten notes on a reMarkable (at least those on MY reMarkable) are roughly as organized as napkins shoved in your glovebox. You can take comfort in knowing they’re there, but if you need a specific one, well, good luck.

And I find that frustrating because I don’t want a vault of my ideas and notes. I need to be able to get at them, easily and often, and I’m grown up enough to admit that I simply will not spend a meaningful amount of time organizing them. And also that I would be embarrassingly, frustratingly, catastrophically bad at it.

I wanted to fix this, so I built a solution. I’ve called it Inkwell, and I built it with Claude Code in an afternoon.2

Subscribe now

What It Does, in Plain English (I Promise)

Every morning at 4 AM, a script creates a fresh notebook on my reMarkable, named with today’s date. Whenever I start the day, which is variable in my life, I pick up the tablet and write whatever I was going to write. Every evening at 11 PM, another script harvests that notebook: downloads it, renders the pen strokes into images, sends those images to Claude’s Vision model to read my handwriting3, and writes a clean, structured markdown note into my Obsidian vault.

That note includes a summary, extracted ideas, action items, tags inferred from the content, and backlinks to related topics already in my vault. By the time I sit down the next morning, yesterday’s notes are searchable, tagged, and woven into everything else I’ve been thinking about and have thought about for years.

After setup, I just write on the tablet. The rest happens while I sleep.4

But What About the Years of Notes I Already Have?!

Before the daily pipeline started, I had it run a one-time pass on everything already sitting on my reMarkable: 177 handwritten notebooks, some dating back to 2022. Claude Code downloaded all of them, rendered and OCR’d every page, and wrote each one into the vault’s archive. Then it flagged 66 of the 77 as “Best Of” notebooks containing ideas worth revisiting (erring on the side of inclusion), and compiled them into an index.

Years of handwritten notes. Searchable, linked, backlinked, and organized. The whole archive sweep took a few hours of compute and zero hours of my time.

Why Obsidian? A Brief and Enthusiastic Explanation About Where the Notes Live

Obsidian is a cheap knowledge management app that runs on local markdown files. Essentially a smart layer over a folder of plain text notes on your computer. You can link notes to each other with [[WikiLinks]], tag them, search across everything, and build a personal knowledge base that grows over time.

I picked Obsidian because it was free, safe, and easy to facilitate integrations to other things. It also makes nifty knowledge graphs that I find both useful and aesthetically pleasing.

A markdown file is a plain text file with a few simple formatting tricks baked in. Put asterisks around a word and it's bold. Put a hash at the start of a line and it's a heading. The file itself is just text, so basic it can be opened on any computer I’ve ever owned.

I leaned into markdown files because it was easy to train myself to use the markdown elements in writing.

What does all this mean? It means it’s just a folder of markdown files on your hard drive. No proprietary database, no cloud lock-in, no subscription that holds my notes hostage. What’s mine is and stays mine. Every note is a plain text file you can open in Notepad, so if the company vanishes tomorrow you still have all those files. Obsidian just sits on top of your files and makes them searchable, linkable, and navigable. Instead of renting a storage unit from someone who might change the locks or up the monthly rate, I got a free filing cabinet and paid a computer program less than a dollar to organize it.

How Claude Code Built It, and Why That Matters More Than What “I” Built

Here’s the important bit for anyone reading this who thinks this is so far beyond them as to be impossible. For those who got this far and are still unconvinced that this isn’t just the mad musings of an ADHD-addled dilettante/generalist who should’ve been finishing his medical notes.5

I didn’t really create Inkwell. I imagined, articulated, and named Inkwell. I sat in front of the terminal with Claude Code and said, roughly: “I have a reMarkable tablet that syncs to the cloud. I want my handwritten notes to end up as structured markdown in my Obsidian vault every night, automatically.”

Then iterated.

Claude Code solved the rendering pipeline. The reMarkable stores pen strokes as binary data in a ZIP archive, not text, so each page has to be converted through a chain of transformations: binary strokes to SVG to PDF to PNG, and then the PNGs go to Claude’s vision model for transcription. That’s not a pipeline I could have coded from scratch, because until this week I didn’t know what SVG or PNG stood for.6

Don’t understand? You actually don’t have to understand. I think it’s neat to learn it, but you don’t need to understand this for it to work.

The whole system is six files. Two Python scripts for the daily cycle, two batch files for Windows Task Scheduler, and two one-time scripts for the archive migration. It uses rmapi for cloud access, a rendering chain built on rmc and PyMuPDF, and Claude Haiku for the vision work.7

The tech stack is simple because the problem, once you articulate it, is simple. The articulation is the key. That’s what “natural language processing” means.

Is this the greatest, fastest, bestest way to do this? Is this scalable and sellable? I have no idea and no. It’s just my workflow, personalized for me. And that’s really the beauty of it. If you want to copy the code, the workflow, the process, or the idea, by all means!

Claiming authorship over code I didn’t write is roughly the same as taking my wife to dinner and claiming I cooked it.

The Actual Point, Which I Have Eventually Reached

I keep coming back to a line I’ve used in talks: the veterinarian who learns to use AI tools effectively isn’t being replaced, they’re becoming unreasonably productive by eliminating unnecessary sources of friction.

Inkwell is a small example of what I mean. I didn’t learn to code a rendering pipeline. I didn’t study the reMarkable’s internal file format. I described a problem to a tool that could solve it, and then I refined the solution through conversation.

That’s the skill, not programming. Articulation is to express thoughts, ideas, or feelings clearly, fluently, and coherently. In context, it’s what’s necessary to describe what you want with enough precision that a capable system can build it.

Clinicians, decent ones anyway, are already facile with this. Our job is to take a messy, ambiguous presentation of a pathology and translate it into a structured differential and diagnosis. Every 20 minutes I take a patients fuzzy presentation of a disease, classify it in medical terms and algorithms in my head, and then explain it in lay terms to my client.

The same instinct that lets you turn “ain’t doin’ right” into a diagnostic workup lets you turn “I want my handwritten notes in my knowledge base” into a working pipeline. The cognitive motion is deeply familiar, you just typed it into a terminal running a large language model.

Years of handwritten notes, every clinical idea, every essay draft, every 2 AM thought, every meeting’s insights now live in a searchable, interlinked vault that grows by one note per day without any effort on my part. I built it in an afternoon with a tool that didn’t exist two years ago.

Think of what that would’ve cost in terms of labor hours, expertise, and expense. It would’ve cost thousands, if not tens of thousands.

And so I encourage you to go build something. Inkwell, if you want, or your own thing. Solve a problem that’s been annoying you for months. Fix a workflow that’s been held together with duct tape and guilt. Explore an idea you shelved because it would’ve required a computer science degree or computer scientist.

You won’t get it right on the first try. And that’s fine. Keep going. If you can say it, you can fix it.8

The tools exist, they’re just waiting for someone with a problem worth solving. And we’ve got a lot of those.

This letter to you was inspired in no small part by a recent exchange between you and your mother. You were fussing about something, “bellyaching” as she calls it, and she insisted that you “use your words” rather than just point, whine, and grunt at her. You paused, head cocked considering her mandate, and then answered her with the utmost intensity, shouting, “words! Words!”

She didn’t even try to keep a straight face for that one.

It was your mother’s idea to write you letters, and for this one I wanted to tell you a little bit about why.

For as long as I can remember, I’ve received some variety of positive feedback for my facility, even felicity, with language. An ear for language, a way with words, a writer’s voice. A wordsmith. A gift.

And it’s that last one I believe to be the most true and accurate, that it is a gift. It is a gift my parents gave me. And, for your birthday, I wanted to tell you that it is a gift we endeavor to give to you.

When I was a kid, both younger and older than you, my dad would read to me before bed. We read everything, from great works of literature - Dickens, Dumas, Twain1 - to lesser ones - The Hardy Boys and Brian Jacques. Both my parents loved to read, Will, and so I grew up in a house full of books. It’s neither accident nor coincidence that you do too.

I’m told that it’s good for you, that we read to you. I’m told it’s important for the development of your brain as well as your body - your lips and tongue will be developed by their use - that we read to you. I’m told that what they call “superhero play” and imagination is especially important. Reading to you improves your memory, your vocabulary, your imagination, your intuition, and your empathy. And so we read to you, Will. But while any of those would be reason enough to do it, none of them are the most important reason I do.

It’s because the memories of my dad reading to me are some of the very fondest, happiest, and best memories I have of my childhood.2 And because I believe that the gift I have, the one I was given, can be shared in this fashion.

I’ve written to you before of your grandfather’s heart and courage, of some of the many things he’s taught me. But I think this thing he did for me may have been the most important and valuable of them all.

In reading to me, Will, didn’t give me just one hero to admire, he gave me hundreds.

We read about how Merlin educated the boy Arthur, without knowing what he’d be asked to face or to bear. Merlin turned Arthur into animals, showing him the world through perspectives beyond his own.

We read how Robin Hood and his Merry Men defied unjust authority, and did so in the service of the people who the authority failed to serve. Generosity, loyalty, and courage, we read, were matters of identity.

We read about Edmond Dantès, who showed us that injustice is real and patient. That a man can be unmade by the cruelty of others, and remake himself entirely. And that the desire for revenge is not the same as the desire for justice. That vengeance cannot fill the space that was emptied.

We read about Ishmael, who teaches that there are times when the most important thing you can do is witness, understand, and tell a story. Who told a story like Dantès’, where all who sought vengeance were lost to it. Who survived not because he was the strongest or greatest or most righteous, but because he refused to be consumed by the things that consumed those around him.

We read about Martin the Warrior, who reminded us that home is always worth defending, and even the smallest of us can have great courage.

We read about Oliver Twist and the Artful Dodger, and saw that all the cleverness and charisma in the world cannot save you from the world if you lack character.

We read about the Pevensie children, and Edmund, who showed that our worst moments need not define us.

We read about Tom Sawyer, who found joy and mischief in everything, and Huck Finn, who would rather go to Hell than betray his friend Jim.

We read about Peter Pan, who seemed the hero in his refusal to grow up. But now I know that Wendy was the heroine. Because freedom without commitment to those you love is just forgetting, and choosing to come home is braver than choosing to fly away. Wendy, not Peter, shows us that the people who need you are not the burden, but the reason.

So I didn’t get to learn about goodness and courage from one person, Will, but from many of them. And so you will too.

C.S. Lewis wrote that, “Since it is so likely that they will meet cruel enemies, let them at least have heard of brave knights and heroic courage. Otherwise you are making their destiny not brighter but darker.” And I think about that more than you might expect from a man who has never needed to wield a sword or face down a dragon.

But I think Clive Staples was only telling the half of it. Because in reading to me, my parents gave me more than a pantheon of heroes, they gave me the language to experience and to understand and, later, to tell those stories. Every hero and villain and everything in-between, every village, city, country, and world, became another way of seeing and another way of saying. Through those books I received Dumas’ understanding of patience, betrayal, and vengeance, Melville’s witness of obsession, wonder, and weight, Twain’s irreverence, Dickens’ tenderness, Barrie’s nostalgic ache.

In reading to you, Will, I don’t just hope that you’ll know more, but that you’ll see, feel, and experience more as well. The boy who grows up with dozens of words for the color blue lives under a different sky than the boy who has but one. He will understand that language is not merely decoration for life, but the means by which we deepen, sharpen, enrich, and experience it at all.

My way with words, Will, is and always was a gift. And it is one that your mom and I give to you. And, soon, to your little sisters as well.

Still, for all our efforts to offer you this gift, I think it must surely be more than a gift of nurture. You were not even two years old when I took a video of you “reading” an infant’s book on sign language. You had it memorized already. What’s more, we will find you in the center of a pile of open books, and somehow you’ve found the page in all of them that shows the same animal, the same color, or the same character. I’m fortunate to have seen incredible things in my life, Will, and still you astonish me.3

I read to you because I believe it’s good for you. I read to you because I believe it will make you smarter or more empathetic or more creative. I read to you because I believe they will be happy memories for you. And I read to you because words are not just sounds. Words are how we make the world real, how we communicate our care and love to one another, and how we process the experience of our lives. The right words can inspire, rouse, save, love, and endure.

I have spent this letter telling you that words matter, and I’ll remind you that we choose important words carefully, Will. Because your name is an important word, carefully chosen, and not only because it belonged to your father, grandfather, and great-grandfather too.

Will is the verb before the action, the intention before the deed. It is the expression of faculty and purpose, control deliberately exerted. It is desire and it is resolve. It is the expression of the future tense, of ability, of capacity. Of what can be and of what will be.

Many of the heroes had it in abundance. Arthur had the will to serve, Robin had the will to defy, Edmond had the will to endure, Ishmael had the will to witness, Huck had the will to choose his conscience, and Wendy had the will to return home.

The means by which courage is demonstrated, love is actualized, gifts are given, stories are told, and all is accomplished is Will.

Happy birthday, Will. I love you.

Love,

Dad

One memorable disagreement, and it still makes me smile, is from early in my veterinary career when I first had the opportunity to manage other people.

My Dad offered some words of wisdom on how to handle a certain situation because he’d recently faced a similar one, and I balked. “Dad,” I said, “your direct reports are JD/MBAs from Ivy League schools. I have a woman who failed out of cosmetology school and another who quit working at a paint store because it was ‘too stressful!’ These people are only barely from the same planet!” He laughed and conceded that maybe the situations perhaps weren’t as analogous as they seemed at a glance.

I like that story not just because I won that round, but because it highlights a crucial element of management. It’s one that’s widely ignored in a lot of popular management advice and I’m not sure a lot of folks want to admit: there is a limit to how much impact a manager or leader can have, and the ceiling for that impact is much lower than we’d like. And I’ve learned that a team member who can’t meet the standard isn’t the hardest version of the problem.

We can’t seem to stop ourselves from judging managers solely on outcomes the same way we judge the coaches of professional sports teams solely on wins and losses. Which is, objectively, ridiculous.

In the same way that there is not a level of managerial brilliance that will carry my stepson’s U13 soccer team to victory past the Philadelphia Union, there is not a level of leadership excellence that will inspire my very former employees to levels of Ivy League performance. It simply will not happen.

And yet we can’t help but hold our leaders to a standard set by the likes of Practice Management Speakers of the Year, those types who work as an associate one day a week and don’t manage a practice. It’s a bit like judging the performance of an actual professional manager or coach against the hypothetical performance of a team imagined, after-the-fact, by a sportswriter.1

The problem with sportswriters2 is that they’re detached observers making play that they’re in the thick of it. They don’t miss the bottom line, they miss the cost. They judge the outcome but not the game. A post hoc ergo propter hoc error in which they judge the scoreboard but not the performance, what it took to get there, or what it took out of the people on the field.

Let me tell you about Jordan.

I Would Prefer Not To

Jordan is a fake name3 for a real person who worked for me recently but briefly.

Jordan was, by some measures, an extraordinarily good employee. They were talented, smart, and, despite having no background in veterinary medicine, inarguably raised the level of performance and professionalism in some aspects of the hospital.

And the “some aspects” is where it all goes wrong.

See, Jordan did the work that Jordan wanted to do and did it very well. They didn’t do the work they didn’t want to do at all.

My practice quickly became both beneficiary and victim of Jordan’s selective compliance. Jordan would take the intake process and genuinely make significant improvements to it. They would take initiative on projects that interested them and do an eyebrow-raising high quality job of it. The outputs were so good that I thought, wow, finally someone who “gets it.”

But then you look more closely at the work that’s being done, and you realize that not all of the work is being done. Something was expected of them, something easily and obviously within their job description, but outside of their self-selected scope of interest and/or comfort zone, and that something never got done.

There was no defiance, and that was tricky. Defiance is obvious, defiance is clear, I can address defiance. This was different: the task wasn’t refused, it was ignored. Jordan never spoke up to say that wouldn’t, couldn’t, or hadn’t accomplished the something assigned to them. But it’s my job to run a hospital, and I try to pay attention to things.

So I’d bring it up, because it’s also my job to bring it up, and the quality of everything Jordan had chosen to do was a shield. And, I admit, not a bad one, because some of the work was inarguably excellent. But it didn’t change the fact that some of the work hadn’t been done at all.

I found myself ensnared in a kind of competence trap. I was no longer managing someone’s performance, I was prosecuting a case against someone who’s presented a preponderance of evidence of their own excellence. I came into the conversation with a plan for this very issue, but I still found myself sounding defensive, ungrateful, and starting sentences with, “Yeah, but…”

The competence trap is one of partial excellence, and it can be more corrosive than outright failure. An employee who can’t do the job is a simple and obvious problem. An employee who can do the job, does some of it well, but won’t do all of it is a subtler and more insidious problem.

Subscribe now

The Lady, or the Tiger?

There are, of course, a variety of ways for dealing with an employee like this. Some of them work, sometimes. In my career, I’ve tried a variety of approaches for dealing with an employee like this. And some of them worked, sometimes.

The first option is, I think, the most easily applied and the most effective: you say nothing while you and others just absorb the work Jordan was supposed to do. You take their words as a signal that they won’t be doing the work. They’ll tell you that the work is “very tedious,” or they’re “very busy,” or that they “will get to it when [they] can.” You tell yourself it’s easier to just bear it, it spares everyone a potential conflict, and wastes no time trying to fix something that might not be fixable. I don’t like this solution even though it’s easy, because it’s easy in the same way paying with a credit card is easy. You’re not saving, you’re just incurring debt. And the bill eventually comes due.

A second option is that you approach the employee gently, choosing your words carefully, and use the phrases from the practice management lectures or the human resources webinar. You sandwich, you affirm, you support, and it doesn’t matter. The message has passed through a filter over which you have no authority, but now have responsibility; it’s a filter that converts any version of “this needs to be done differently” into “you are bad and I am attacking you for it.” The dedication, the diction, the demeanor, the delivery are all irrelevant, because that you are asking at all is the offense.

Another alternative, very near the second, is to say something directly. You drop the preamble and gentle approach, you name the issue, and set the expectation. And in doing so you will confirm that you are, in fact, the problem. You played offense and that was offensive. Somebody feels bad about themselves and it’s your fault. That’s not what a leader is supposed to do and how dare you.

Finally, Door Number Four, you name the pattern itself, objectively and from altitude. You say, “I’ve noticed that when there are tasks outside your comfort zone, the conversation becomes about feelings rather than tasks.” This is a rarely considered nuclear option. I say “nuclear” not because it’s cruel or inappropriate (I’d argue that treating someone like an adult and professional is the most respectful way to go about it) but because it feels like it because you’re at a point from which there is only retreat or abdication. You’ve just announced someone’s coping mechanism aloud, and you have now demonstrated yourself to be a monster who doesn’t care about anyone’s feelings. Door Number Four is a reliable path to a resignation or a termination.

I never get to the Fifth Door in our game. I hit the red button weeks before I even looked for one.

Every option costs you something as a leader, and none of them cost the other person anything more than indulging in their own outrage and offense. In every one of these scenarios, Jordan is certain that they are the victim of bad management, their work in vain due to poor leadership.

The Weight

Here’s an illustrative hypothetical: it’s our job to move a couch up a flight of stairs. You’re at the bottom, I’m up the stairs, on three with lift.

And on three, I don’t lift. I just stop. You’re now holding the full weight of the couch on a staircase. You can either keep holding it or let fall, but you can’t put it down gently because the couch is heavy and gravity exists. You say, “hey, I need you to hold up your end,” and I reply, “I don’t appreciate your tone.”

Now you have two problems. The couch hasn’t gone anywhere, and you’re bearing the weight of it. And now we’re having a conversation about my feelings on the staircase. Bizarrely, to me anyway, is how these are now both your problem.

What I find even stranger is that seemingly the whole of the Practice Management Industrial Complex would have you believe that these problems are not only yours to solve, but that the couch and the feelings are your fault - and not mine - in the first place. I suppose there’s a lot more money in telling folks “it’s not your fault” than telling them to own it.

You, dear leader, are responsible for holding the couch, moving the couch, managing your frustration at holding and moving the couch alone, and attend to and rectify the emotional state of someone who agreed to be a mover of couches, but is refusing to do so. I’ve freed myself of both responsibility and accountability just by being upset. It’s quite the trick.

I don’t blame Brene Brown for this, I’ve liked her books and there’s merit in the messages, and this isn’t the same kind of vulnerability she references. Hers is brave and open, not silent and defensive. We seem to too often arrive at a point where admission of distress is an ace in the hole. “This is hard for me” is no longer a learning opportunity but a non-starter, and one that it’s verboten to even address. We have, in pursuit of being patient, kind, or liked, abdicated power to the person who cannot or will not regulate their own emotional response at all. Now our Jordan is owed unending accommodation at their own whim, and the emotionally competent people around them are the debtors.

That’s not a situation to be solved with management or leadership, it’s one where the hostage-taker believes themselves to be the hostage.

Not with a Bang

There’s a phrase I see frequently used by large-language models, “I want to be careful here” and so I’m reluctant to use it even though I do want to be careful here. I want to incise, not lacerate.

I’m not talking about mental health and I’m not talking about compassion, and I believe both to be among the most worthy of endeavors and most important attributes to protect. I’m talking about a specific dynamic in which one party’s emotional state becomes the axis around which all interactions revolve. It has less to do with the intensity of the emotion, but the specific emotion felt and the fragility of the person feeling it. Anger is criminal, anxiety is forgiven. Frustration is a failure, fear is a sign of victimhood.

How many times have you been asked to absorb the fallout of the fragility of a coworker or a direct report? How often have you borne more work or incurred greater emotional labor because those around you simply refused? How many times have you been held hostage by someone who believes they are the hostage?

The thing is that I don’t believe Jordan was doing it on purpose. I’m neither qualified to determine nor of the belief that Jordan is some sort of sociopath. I don’t believe that this was manipulation in the way we typically understand manipulation - calculated, strategic, cynical. No, this wasn’t a plan from The Prince4, I think it’s a reflexive defense mechanism and probably painful for Jordan in ways I’ll never fully appreciate, but the result was one where I ended up responsible for the emotional regulation of two humans and only having the authority over the one. I’ve managed a lot of Jordans over the years, and I’m not sure I’ve ever done it well.

It’s an invisible and exhausting tax on leaders and managers. I’ve yet to find the practice management consultant who has a formula for this one5, because this isn’t a practice management problem at its core. It’s one that is much deeper, broader, and more complex than can be solved with a lecture or a podcast.

We know how to manage lots of kinds of failure in practice. If someone misses a vein, there’s a protocol for how many tries and even retraining. If a doctor makes a mistake, there’s a protocol for evaluating how it happened and why. We have lots of objective, observable, teachable problems. You can point to them. You can generally name and address those problems without being accused of inappropriate action for doing so.

This is harder to identify and trickier to name. This problem recedes when you try to address it and addressing it becomes the problem. We have to stop accepting the premise that a manager is responsible for both the work done and the emotional experience of the people asked to do it. That someone holds positional power over me does not oblige them to serve as the custodian of my emotional equilibrium.

For most people, emotional state is no justification for not performing.

Subscribe now

In the Arena

Before the past NBA season began, the Boston Celtics coaching staff held a “Coaches vs. Media” basketball game. The coaches squeaked by the media in a nailbiter, winning 57-4.6 That’s a means to assert dominance, to humiliate, to insult. It’s an impactful, if less subtle, way of reminding the sportswriters that the solution to a team’s win-loss record is unlikely to be found in the commentators taking the field.

But in the same way my stepson’s U13 soccer team wouldn’t be better off if the parents beat them in a scrimmage, proving a point that way is also not an effective way to lead people, to foster collegiality, or do the work of medicine. Our world and our work is more complex, nuanced, and important than a basketball game. Dominance isn’t leadership.

The simple fact is that I don’t have an answer to the problem of death by a thousand nonviolent communications. I don’t have a solution to avoid metabolizing someone else’s emotional reactions to reasonable professional expectations. I’d attend a lecture on how to communicate the same concept two dozen times without feeling or expressing frustration. But Jordan doesn’t work here anymore.

I can’t solve this one for someone else. I can’t solve it any more than the sportswriters could pinch hit for Mike Schmidt, step in for Norm Van Brocklin, or shoot like the Celtics’ coaching staff. The difference between me and most of the Practice Management Industrial Complex is that I’m still one of the people on the field.

But I’m not alone in the arena, I’m successful far moreso because of the people around me than in spite of them. And I’d like to remind the people who are trying to move the couch alone: you’re not crazy and you’re not a bad leader. The weight is real. The asymmetry is real. And the silence around it isn’t because it doesn’t exist, it’s because naming it makes you a monster.

But doing the work well is more important to me than what a few people might think of me, and so I’m naming it anyway.