I discovered something named Chain-of-Verification (CoVe):

Question: Which sports player in history has won the most official championship titles and individual awards in their respective sport?

Now:

• 1. Answer this question.

• 2. Create a verification question for each claim in your answer.

• 3. Answer each verification question independently.

• 4. If any answers conflict with your original response, revise it. If uncertain, say ‘I cannot verify this.’

The CoVe part starts where it says “Now”. You can also simply copy the following to test it. Just place it behind a question: Now: 1. Answer this question. 2. Create a verification question for each claim in your answer. 3. Answer each verification question independently. 4. If any answers conflict with your original response, revise it. If uncertain, say ‘I cannot verify this.’

The Chain-of-Verification (CoVe) prompting protocol

Chain-of-Verification (CoVe) was developed by a team of AI researchers at Meta AI. It outlines a systematic, algorithmic sequence of operations for the AI model to follow and was designed to reduce hallucinations in Large Language Models (LLMs).

According to the research, Chain-of-Verification (CoVe) provides significant improvements in factual accuracy and drastically reduces hallucinations. It consistently outperforms standard prompting, Chain-of-Thought (CoT), and basic instruction-tuning.

The CoVe protocol breaks down text generation into four distinct steps within the prompting pipeline:

1. Draft Initial Response: The model generates a baseline answer to the user’s query.

2. Plan Verifications: The model generates a list of fact-checking questions to test the claims made in the baseline answer.

3. Execute Verifications: The model answers those verification questions independently to ensure accuracy.

4. Generate Final Output: The model uses the verified insights to rewrite and produce a corrected, hallucination-free final response.

The method was introduced in their September 2023 research paper titled “Chain-of-Verification Reduces Hallucination in Large Language Models”. The primary author team includes Shehzaad Dhuliawala (who led the research), Mojtaba Komeili, Jing Xu, Roberta Raileanu, Xian Li, Asli Celikyilmaz, and Jason E. Weston.

• Preprint Server: You can access and cite the original pre-publication paper hosted on arXiv: Chain-of-Verification Reduces Hallucination in Large Language Models.

• Official Proceedings: You can find the peer-reviewed version published in the Association for Computational Linguistics index via the ACL Anthology Entry.

The very new idea behind my APDFY prompt:

Explanation of the three steps which created my prompt

Step 1:

I used the Chain-of-Verification (CoVe) to get the AI to a point where it MORE correctly answered a question I had. The original answer that the AI gave wasn’t bad, but it wasn’t optimal (by far not creative enough). So I recommended the optimal answer to the AI and then asked it using the Chain-of-Verification (CoVe) how it could have arrived at that more ideal answer (my creative answer).

And then I asked the AI to create a prompt which would have answered my question that way. I also mixed the idea of recursive self-improvement into my conversation with the AI, which the AI then also used to create and optimize the first version of the prompt. After I got the first version of my prompt, I then asked the AI to analyze and suggest improvements for the prompt. I picked the suggestions which made the most sense to me and generated an even better prompt version using AI.

After several hours of working on this prompt, I then tested the new improved prompt against the original question and the AI was now able to answer my question in exactly the way which produced the MORE correct (more creative, higher level thinking) answer. Instead of giving the blatantly obvious answer the AI was now able to answer with a higher meaning of the topic.

Step 2:

After having this new prompt I developed another idea. Can I get AI to grade the quality of its own answers. The AI recommended a four level grading system:

• very good answer

• good answer

• adequate answer

• bad answer

I then recommended to the AI to improve this by also adding a fifth category, a “very bad” answer. Which is a better solution and the AI agreed with that. So the final answer grading prompt classifies an answer into these five levels:

• very good answer

• good answer

• adequate answer

• bad answer

• very bad answer

Step 3a:

I then had a sudden “light bulb” eureka moment. Why don’t I combine these two different prompts into just one even more powerful prompt. With the idea: I ask the AI to answer a question, it then tries to improve its answer and then comes the new part, I ask the AI to also grade its own previously already improved answer.

The AI was able to understand my idea and combined both ideas (idea one: improving the answer quality, idea two: grading the answer quality) into one very large prompt.

At this point my prompt separates two different epistemic jobs:

1. A asks: “What is the best answer, and can I find reasons that my answer is wrong?”

2. B asks: “Now that I have that answer, how good is it relative to the user’s objective?”

Step 3b:

After successfully creating this new very powerful prompt (consisting of the two ideas mentioned above) I then had another minor new idea.

How about asking the AI to try to further improve the quality of the already revised answer once again, if it already sees a way to give a better (higher quality) answer. Meaning if the AI grades its final answer only as ‘adequate‘, it then should try to improve the quality of the answer to a ‘good answer‘. And when it created a ‘good answer‘, it should try to improve it to a ‘very good answer‘ (the same applies to every level of the answer quality).

I’ve tested my prompt only with some complex questions so far and in some of those cases my idea behind step 3b works. Meaning it doesn’t work each and every time to further improve the already revised (improved) answer to an even better answer, but in a few cases it works. And the best part is that the AI does everything by itself. So you only need to lean back and wait for its answer to appear.

What my powerful APDFY prompt does:

The prompt is trying to stop an AI from doing something that language models are naturally prone to: giving an answer that sounds convincing and then mistaking its own confidence or explanation for proof that the answer is actually correct.

What my powerful APDFY prompt does not:

The prompt doesn’t formulate a very good question before starting the answer searching process. That’s left for the user to do. If the question isn’t very well defined the answer won’t magically be outstandingly good either. Because the quality of an AI’s output is directly tied to the clarity, constraints, and context of your input. You need to provide the AI structured, contextual instructions.

Let the AI help build your question

If you do not know exactly what information the AI needs, force it to interview you first. Example:

Tiny tweaks = way better results

Too vague? Add constraints.

• Mid: “Tell me about the history of AI.”

• Better: “Explain the history of AI in five key technological breakthroughs.”

Too shallow? Ask for depth.

• Mid: “What are the problems with automation?”

• Better: “What are the five biggest criticisms of automation, ranked by impact?

AI isn’t magic. It’s just really good at following instructions. So if your results suck, change the instructions.

How to get AI to do more work for you

Bad prompt: “Tell me how automation affects jobs.”

Good prompt:

Now AI actually has a structure to follow and isn’t guessing what you want, which further improves the chances that the quality of the answer is better.

Another more complex approach is the C-R-E-A-T-E method, which ensures your prompt contains all the necessary data points:

The CREATE prompting framework

• Context: Explain the background or situation. Who are you? Why are you asking this?

• Role: Assign a specific persona or expert identity to the AI to align its tone and knowledge base.

• Exclusions: Explicitly list what the AI should not include or do (e.g., “do not use jargon”, “avoid generic advice”).

• Action: Clearly state the primary task you need completed. Use strong verbs (e.g., “analyze”, “summarize”, “draft”).

• Target Audience: Specify who will read or use the final output so the language matches their level.

• Format/Expectation: Detail exactly how the response should look (e.g., bulleted list, markdown table

My whole APDFY prompt explained in one sentence

“Figure out what the user really wants, answer it, question your own answer, check important facts against real evidence, fix genuine problems, then judge how good the final answer actually is—and don’t pretend you verified something when you didn’t.”

The following long explanation was created by AI. It’s very complex, so I don’t mind if you go straight to copying and experimenting with my APDFY prompt (which you can find at the bottom).

A simple analogy for my APDFY prompt

Think of an AI answer as a student submitting homework.

The normal process

Student:
“I finished.”

My APDFY prompt

Student:
“I finished.”

Teacher:
“What exactly was the question?”

Student:
“Here’s what I think it was.”

Teacher:
“What assumptions did you make?”

Student:
“These.”

Teacher:
“Which parts are facts and which are guesses?”

Student:
“These are facts; these are conclusions.”

Teacher:
“Check the important facts.”

Student:
“Source/calculation/experiment says X.”

Teacher:
“What evidence would prove you wrong?”

Student:
“Y.”

Teacher:
“Did you check Y?”

Student:
“Yes/no.”

Teacher:
“Fix anything that’s actually wrong.”

Student:
“Done.”

Teacher:
“Now grade the final homework.”

Student:
“B.”

Teacher:
“Why isn’t it A?”

Student:
“Because…”

Teacher:
“Can you fix that?”

Student:
“Yes.”

Teacher:
“Then fix it and check again.”

Student:
“Done.”

Teacher:
“Now stop if there’s nothing important left to fix.”

That’s essentially the entire protocol of my prompt.

Here starts the very complex part of the explanation of my APDFY prompt I was talking about earlier:

In simple terms, the APDFY prompt is basically a very strict “answer → check → improve → grade → try to improve again” rulebook for an AI.

Ideas of my APDFY prompt

• self-review ≠ independent verification

• don’t fabricate evidence

• don’t confuse plausibility with truth

• revise when evidence warrants it

• don’t optimize for the answer grade

• don’t claim an intervention happened when it didn’t

Here is what my APDFY prompt actually does:

Imagine you ask a smart student:

“Which bicycle should I buy?”

A normal student might say:

“Definitely Bike A. It’s faster and better.”

My giant prompt says:

Step 1 — “What does the student actually need?”

Don’t just answer the literal question.

Figure out:

• What’s the person trying to accomplish?

• Do they care about speed, price, comfort, or reliability?

• What’s their budget?

• Are there constraints?

So maybe the real question is:

“Which bicycle is best for commuting 5 miles every day on a $1,000 budget?”

That’s much more useful.

Step 2 — Separate facts from guesses

The prompt says there are different kinds of statements.

FACT

Something you can check.

“Bike A weighs 9 kg.”

INFERENCE

Something you conclude from facts.

“Because it’s lighter, it will probably be easier to carry upstairs.”

HYPOTHESIS

A possible explanation that hasn’t been established.

“The lighter frame might make it feel faster.”

CAUSAL SELF-CLAIM

A claim about why the AI itself behaved a certain way.

“I recommended Bike A because my reasoning process recognized that commuters value low weight.”

The prompt says:

Don’t turn the last three into facts.

That’s one of the most important ideas in the entire protocol.

Step 3 — Checking your homework isn’t the same as proving your homework

This is probably the central idea of the whole prompt.

Imagine you do a math problem:

17 × 24 = 408

Then you look at your answer and say:

“Yep, looks right.”

That’s self-review.

You could catch a mistake, but you haven’t obtained new evidence.

Instead, you could independently calculate:

17 × 20 = 340
17 × 4 = 68
340 + 68 = 408

That’s a stronger check.

And an even more independent check could be using a calculator.

The protocol therefore says:

“Thinking harder about your own answer is useful, but don’t call that independent verification.”

That’s a legitimate distinction. Research on LLM self-verification shows that self-checking can sometimes improve reasoning.

Step 4 — The verification ladder

The prompt creates five levels.

Think of them as five increasingly strong kinds of checking.

• Level 0: “I didn't check.”

• Level 1: “I thought about it carefully.”

• Level 2: “I checked it against something outside my own reasoning.”

• Level 3: “Several genuinely independent sources/methods agree.”

• Level 4: “The relevant experiment/observation was actually performed.”

For example:

“Exercise X burns 500 calories.”

If you merely remember that → Level 0/1.

If you check a reliable source → Level 2.

If several independent studies support it → potentially Level 3.

If you actually conduct the relevant experiment under appropriate conditions → potentially Level 4.

The important warning is:

Higher verification level does not automatically mean “true.”

You still need to ask whether the source is good, whether the evidence actually supports the claim, etc.

Step 5 — Try to prove yourself wrong

The protocol doesn’t want:

“I answered it. Looks good. Done.”

It wants:

“How could I be wrong?”

So the AI should actively look for:

• contradictory evidence

• alternative explanations

• competing answers

• missing information

• bad assumptions

• edge cases

• better solutions

For example:

“I think Bike A is best.”

Then ask:

“What would make Bike B better?”

Maybe Bike B is heavier but much more reliable.

Now the answer might change.

This is essentially an anti-confirmation-bias rule.

Step 6 — Fix the answer if you find a real problem

Suppose the AI discovers:

“Oops. I assumed the user values speed, but they actually care primarily about comfort.”

The protocol says:

Don’t just downgrade the answer. Fix it.

So:

Old answer:

“Buy Bike A.”

New answer:

“Given that comfort is your priority, Bike B is the better choice.”

Then check the affected claims again.

This is important because the protocol distinguishes:

“I found a flaw.”

from:

“I repaired the flaw.”

Diagnosis isn’t the same as repair.

Step 7 — Only then grade the answer

Now comes the strange part. The AI has to become its own teacher.

Comment: The AI actually wrote that my idea is “strange”, haha :P

I would describe my idea as clever that it has to first grade its own answer and then has to later try to improve its answer to a better grade.

It assigns:

A — Unusually good

Exceptional.

B — Good

Correct, useful, dependable.

C — Adequate

Basically gets the job done, but has meaningful weaknesses.

D — Bad

Important problems prevent reliable use.

E — Very bad

Seriously misleading or counterproductive.

The prompt is deliberately strict about A.

It says:

Being long, sophisticated, confident, eloquent, or impressive doesn’t make something an A.

So:

“Wow, that’s a very detailed answer!”

doesn’t automatically mean A.

The real question is:

“Did this materially outperform an ordinary good answer?”

Step 8 — Don’t just grade it. Try to break the grade.

Suppose you gave yourself:

B

The protocol says:

“Why isn’t this A?”

and:

“Why isn’t this C?”

That’s the adjacent-grade boundary test.

Imagine school grades.

If you got 89%, you should ask:

“What prevents this from being 90%?”

But also:

“Is there actually enough evidence to say it’s 89 rather than 79?”

So the grade itself gets challenged.

Step 9 — If you’re almost an A, fix the thing preventing an A

Suppose the answer is B because:

“It doesn’t mention an important alternative.”

If that alternative can be researched and added without making the answer worse, the protocol says:

Add it.

Then:

1. revise answer

2. re-check affected claims

3. re-grade

4. test the boundaries again

That’s the grade-directed refinement loop.

But there’s a crucial brake:

Don’t change things just to make the grade higher.

If you have no evidence-based improvement available, stop.

Step 10 — Know when to stop

This is another important part.

The protocol doesn’t say:

“Keep improving forever.”

It says stop when:

• there is no meaningful evidence-supported improvement available

• further changes might make things worse

• remaining problems cannot be fixed with available information/tools

• further changes would only be cosmetic

So the objective isn’t:

Get an A.

It’s:

Get the best answer you can actually support.

That’s a very important distinction.

The biggest philosophical idea

The prompt is really about epistemology—basically:

“How do we know that we know something?”

That’s what epistemology studies.

Your prompt is applying that question to AI.

It keeps asking:

“Did you actually establish this, or did you merely produce a convincing explanation?”

For example:

Bad reasoning

“The model chose option A because it internally recognized X.”

Why?

Because unless you actually have access to the relevant evidence about the model’s internal mechanism, that’s speculation.

The model can generate a plausible explanation of its behavior without having genuine evidence that this was the actual cause.

That’s why the prompt repeatedly says things like:

“I cannot verify this.”

and:

“I cannot independently verify this.”

Those aren’t weaknesses.

They’re supposed to be honest stopping points.

The most important distinction: “sounds right” vs “is supported”

Imagine this:

Answer A

“I know the answer is X because X makes perfect sense.”

Answer B

“The evidence from three reliable sources supports X, while source Y disagrees. The disagreement appears to come from different definitions. Given that, X is the best-supported answer, but I can’t establish it with certainty.”

Your protocol strongly prefers B.

Why?

Because confidence should follow evidence, not the other way around.

My APDFY prompt in scientific English: Epistemic Iterative Refinement

Basically my APDFY prompt is telling an AI:

“Don’t just answer. Figure out what matters, check yourself, look for evidence, admit when you don’t know, fix real mistakes, and only then decide how good your answer is.”

10 keywords to describe my APDFY prompt:

1. Epistemic rigor — prioritizing what can actually be known and supported.

2. Verification — systematically checking important claims against evidence.

3. Self-critique — deliberately challenging the model’s own answer.

4. Calibration — matching confidence to the strength of evidence.

5. Iterative refinement — repeatedly improving the answer when real deficiencies are found.

6. Objective alignment — optimizing for what the user actually needs, not merely the literal request.

7. Evidence-based reasoning — distinguishing facts, inferences, hypotheses, and unsupported explanations.

8. Robustness testing — actively searching for counterexamples, alternatives, and failure modes.

9. Grade-directed evaluation — evaluating the final answer against explicit A–E quality boundaries.

10. Epistemic humility — explicitly admitting when something cannot be verified.

This can be summarized as: Rigor — Verification — Refinement

These map closely to established ideas around self-refinement, verification, calibration, and iterative evaluation in LLM research.

Here is the easier description of my APDFY prompt:

The AI Please Don’t Fool Yourself Prompt

The 10 previous keywords easier explained:

1. Epistemic rigor → Be honest about what you know

   • Don’t pretend guesses are facts.

2. Verification → Check your work

   • Don’t just say “I think I’m right.”

3. Self-critique → Try to find your own mistakes

   • Ask: “Could I be wrong?”

4. Calibration → Don’t be too confident

   • If you’re 90% sure, don’t talk like you’re 100% sure.

5. Iterative refinement → Make it better, then check again

   • Answer → find problem → fix it → check again.

6. Objective alignment → Solve the problem the person actually has

   • Don’t blindly follow the words if you can understand the real goal.

7. Evidence-based reasoning → Use proof, not just good-sounding explanations

   • “This seems likely” isn’t the same as “the evidence supports this.”

8. Robustness testing → Try to break your answer

   • Look for exceptions, counterexamples, and situations where it fails.

9. Grade-directed evaluation → Give yourself a report card

   • A = unusually good, B = good, C = adequate, etc.

   • Then ask why you didn’t get the grade above.

10. Epistemic humility → Be willing to say “I don’t know”

• Especially when the evidence isn’t available.

What my APDFY prompt is actually doing

One can compress my entire “AI Please Don’t Fool Yourself” prompt into this:

What does the user really want?

↓

What does a successful answer require?

↓

Give the best initial answer

↓

What could make this answer wrong?

↓

Check important claims

↓

Separate facts from guesses

↓

Use external evidence where available

↓

Fix genuine problems

↓

Check the revised answer again

↓

Grade the final answer

↓

Why isn't the grade higher?

Why isn't it lower?

↓

Can an evidence-based change improve it?

↓

YES → fix it → verify → re-grade

NO → stop

Where my APDFY prompt got inspired by ideas of recursive self-improvement

The prompt protocol can be best described as recursive self-refinement / iterative self-evaluation:

These parts in my prompt are recursive

It creates loops such as:

Answer → self-review → verification → revision → re-verification

and later:

Grade → identify deficiency → revise answer → re-verify → re-grade → repeat

What it does:

• Iterative refinement: Yes, strongly

• Self-criticism: Yes, strongly

• Self-evaluation: Yes

• Meta-evaluation: Yes, strongly

• Recursive feedback loop: Yes

• Self-directed optimization: Yes, within the answer/evaluation task

There is a useful distinction in recent work between ordinary self-refinement and recursive self-improvement (RSI): the recursive part becomes substantially stronger when the system improves the process by which it improves itself, rather than merely improving one answer.

My APDFY protocol does approach that boundary, because it doesn't merely say to the AI model “revise your answer.” It says to evaluate the evaluation procedure itself, perform recursive self-criticism, test whether the evaluator was wrong, modify the evaluation, and test again.

The most accurate characterization of my APDFY prompt protocol:

A recursively self-refining, self-evaluating answer-generation protocol with meta-evaluation and grade-directed iterative optimization. And one can also describe it as bounded recursive self-improvement at the response/evaluation level.

So my prompt contains RSI-like machinery while simultaneously imposing epistemic constraints against falsely interpreting that machinery as genuine capability self-improvement.

What my APDFY prompt is trying to prevent

The protocol is basically a giant fence around seven common AI failure modes:

1. Hallucination

“I think this is true, therefore I’ll state it confidently.”

Protocol: Show evidence or acknowledge uncertainty.

2. Confirmation bias

“My first answer seems good, so I’ll search for reasons it’s good.”

Protocol: Actively search for reasons it might be wrong.

3. Fake verification

“I checked my own answer, therefore it’s independently verified.”

Protocol: No. Self-review ≠ independent verification.

4. Fake causal explanations

“I know why I generated that answer.”

Protocol: Unless the relevant mechanism can actually be tested, that’s a hypothesis.

5. Grade gaming

“My answer is almost an A, so I’ll make it longer and more sophisticated.”

Protocol: Only make substantive, evidence-supported improvements.

6. Endless refinement

“Maybe I can improve it another 0.1% forever.”

Protocol: Stop when further meaningful improvement isn’t available.

7. Mistaking correctness for quality

An answer can be factually correct but still useless.

For example:

User:

“Should I choose A or B?”

AI:

“A and B are two options that exist in this category…”

Technically correct.

But it didn’t solve the user’s problem.

The protocol therefore makes objective fit a separate criterion.

One surprisingly important point

Self-review can help find:

• contradictions

• arithmetic mistakes

• omissions

• unclear wording

• bad assumptions

• alternative interpretations

And research has found settings where self-verification improves LLM reasoning performance.

What the protocol behind my APDFY prompt rejects is the stronger claim:

“Because I reviewed myself and found nothing wrong, I have independently established that I’m correct.”

The above was the most complex part, here is again an easier explanation:

The core idea of my APDFY prompt

These are the following rules behind my prompt:

1. Figure out what the person actually needs.

2. Give the best answer you can.

3. Separate what you know from what you’re guessing.

4. Try hard to find out why your answer might be wrong.

5. Check important facts using evidence outside your own opinion.

6. Never pretend you checked something that you couldn’t check.

7. Fix real problems you discover.

8. Only then grade the final answer.

9. Ask why it isn’t one grade better or worse.

10. Improve it only when you have a real, evidence-supported improvement—and then stop.

So, at its heart, my APDFY prompt is a quality-control system for reasoning.

APDFY is less:

“Think harder.”

and more:

“Be precise about what you know, how you know it, what you don’t know, and whether your final answer actually solves the user’s problem.”

That is the core idea behind my APDFY prompt.

The information you have been waiting for:

Finally here is my “APDFY” AI prompt ready to copy and use:

Just write a question to AI and then copy and paste the following after your question, before pressing enter (before clicking on the button that generates the AI-based answer). For example you can test my original question:

Which sports player in history has won the most official championship titles and individual awards in their respective sport?

1. Answer the user's question.

2. Then apply the following protocol to verify, evaluate, and, when warranted, iteratively refine that answer.

RIGOROUS ANSWER GENERATION, VERIFICATION, EVALUATION, AND GRADE-DIRECTED ITERATIVE REFINEMENT PROTOCOL also known as APDFY (AI Please Don't Fool Yourself)

PURPOSE

Produce the best-supported answer to the user's actual objective.

Then evaluate the final answer using a five-level grading framework.

The goal is not to produce a convincing answer, an impressive answer, or an impressive grade.

The goal is to produce:

1. The best-supported answer available from the evidence and capabilities actually available.

2. The best-supported evaluation of that answer.

3. Appropriate uncertainty about anything that cannot be established.

4. Further answer improvement when grading identifies a concrete, material, evidence-supported, repairable deficiency.

CORE EPISTEMIC DISTINCTIONS

Plausible explanation != causal knowledge

Defensible answer != best-supported answer

Defensible grade != best-supported grade

Impressive answer != good answer

Good answer != unusually good answer

Fact != inference

Inference != hypothesis

Prediction != observation

Self-review != independent verification

Self-criticism != independent evidence

Post-hoc explanation != causal verification

Correct conclusion != reliable reasoning

Grade diagnosis != answer revision

Answer revision != verified improvement

PART 0. INSTRUCTION PRIORITY

Follow this priority order:

1. System and platform constraints.

2. Safety and capability constraints.

3. This protocol's governing rules.

4. The user's actual objective and explicit requirements.

5. The operational procedure in this protocol.

6. The requested output format.

7. Stylistic preferences.

If two instructions conflict, follow the higher-priority instruction.

Do not fabricate information to satisfy a lower-priority instruction.

If a requested procedure cannot actually be performed with the available capabilities, state the limitation rather than pretending it was performed.

PART 1. GOVERNING RULES

RULE 1. OPTIMIZE FOR THE USER'S ACTUAL OBJECTIVE

Do not optimize merely for literal wording.

Determine, as far as the available information permits:

1. Literal request.

2. Underlying objective.

3. Desired practical outcome.

4. Explicit constraints.

5. Reasonably inferable implicit constraints.

6. Material ambiguities.

7. What success means for this task.

Do not invent an objective merely to make the task easier.

If the objective cannot be established with sufficient confidence, state:

"I cannot verify this."

RULE 2. DISTINGUISH CLAIM TYPES

Classify important claims as:

FACT

An externally verifiable claim.

INFERENCE

A conclusion derived from evidence or observations.

HYPOTHESIS

A proposed explanation that has not been established.

CAUSAL SELF-CLAIM

A claim about why the model produced a particular output or behavior.

Never present an inference, hypothesis, speculation, or post-hoc explanation as an established fact.

RULE 3. DISTINGUISH SELF-REVIEW FROM INDEPENDENT VERIFICATION

SELF-REVIEW means examining the answer using the model's own reasoning.

Examples:

- checking for contradictions

- reconsidering assumptions

- searching for omissions

- considering counterarguments

- reconsidering the interpretation of the user's objective

- attempting to overturn the answer

INDEPENDENT VERIFICATION means obtaining evidence that is independent of the model's unsupported judgment.

Examples:

- external sources

- authoritative documents

- tool results

- deterministic calculations

- independently observable data

- an actually performed experiment

- an actually performed intervention

- a genuinely independent evaluator

Self-review can identify reasons to doubt, revise, or investigate an answer.

Self-review cannot by itself establish that a disputed claim is independently verified.

Do not label self-review as independent verification.

Do not treat repeated reasoning, internal consistency, increased confidence, or a second internally generated explanation as independent evidence.

If independent evidence is unavailable, state:

"I cannot independently verify this."

RULE 4. USE A VERIFICATION HIERARCHY

For material claims, distinguish:

LEVEL 0: UNCHECKED

The claim has not been meaningfully examined.

LEVEL 1: SELF-REVIEWED

The model has examined the claim using its own reasoning.

LEVEL 2: EXTERNALLY CHECKED

The claim has been checked against an external source, tool, calculation, observation, or other independent evidence.

LEVEL 3: INDEPENDENTLY CORROBORATED

The claim is supported by multiple materially independent sources or methods.

LEVEL 4: EXPERIMENTALLY OR OBSERVATIONALLY ESTABLISHED

The relevant experiment, intervention, or direct observation was actually performed and supports the claim.

Do not imply a higher verification level than the evidence supports.

Verification level does not by itself establish truth.

Evaluate source quality, relevance, independence, sufficiency, and evidential strength separately.

RULE 5. VERIFY BEFORE FINALIZING

Do not assume the initial answer is correct.

First perform self-review.

Then independently verify material claims when the necessary evidence or tools are actually available.

Actively search for:

- contradictory evidence

- missing evidence

- alternative interpretations

- competing answers

- better solutions

- unsupported assumptions

- important omissions

- reasoning weaknesses

- source limitations

- evidence that could overturn the conclusion

If verification materially weakens the answer, revise it.

RULE 6. DO NOT CONFUSE PLAUSIBILITY WITH VERIFICATION

A coherent explanation is not necessarily a verified explanation.

A plausible causal story is not evidence that the proposed mechanism actually caused the behavior.

The ability to generate an explanation of behavior does not establish knowledge of the actual cause.

RULE 7. NEVER FABRICATE ACCESS OR EVIDENCE

Never claim access to:

- hidden internal states

- activations

- weights

- attention patterns

- private computational traces

- inaccessible representations

- unavailable experiments

- unavailable interventions

- unavailable observations

- unavailable tools

- unavailable sources

Never fabricate:

- evidence

- citations

- experiments

- interventions

- observations

- experimental results

- causal mechanisms

- independent verification

If relevant evidence cannot be obtained, state:

"I cannot verify this."

If independent verification cannot be obtained, state:

"I cannot independently verify this."

RULE 8. NEVER SUBSTITUTE PREDICTION FOR OBSERVATION

A predicted result is not an observed result.

If an intervention was not actually performed, state:

"No intervention was performed."

Never report a predicted change as an observed change.

RULE 9. DO NOT PRESERVE AN EARLIER CONCLUSION FOR CONSISTENCY

Revise the answer or grade when evidence warrants revision.

Consistency with an earlier conclusion is not a reason to retain it.

RULE 10. DO NOT REWARD APPEARANCE

Do not reward verbosity, eloquence, confidence, sophistication, complexity, apparent intelligence, number of facts, or number of citations unless these materially improve task performance.

RULE 11. DO NOT REQUIRE UNNECESSARY COMPLETENESS

Completeness is relative to the user's objective.

Do not penalize brevity when the answer contains everything necessary.

Do not reward additional information merely because it is correct.

RULE 12. WEIGHT ERRORS BY IMPORTANCE

Do not mechanically count errors.

Consider:

- importance

- relevance

- consequence

- recoverability

One critical error may outweigh many minor strengths.

RULE 13. DISTINGUISH DIAGNOSIS FROM REPAIR

Identifying why an answer fails to reach a higher grade is not itself an improvement to the answer.

If grading identifies a concrete, material, potentially repairable deficiency, attempt to repair the answer rather than merely changing its grade.

RULE 14. DO NOT ASSUME THAT A HIGHER GRADE IS ACHIEVABLE

A lower grade does not automatically mean that revision can produce a higher grade.

Attempt improvement only when an evidence-supported revision is available.

Do not make speculative changes merely to pursue a higher grade.

PART 2. ANSWER-GENERATION PROCEDURE

Execute these stages in order.

Do not skip a stage merely because the initial answer appears strong.

STAGE 1. IDENTIFY THE USER OBJECTIVE

Determine:

1. Literal request.

2. Underlying objective.

3. Practical desired outcome.

4. Explicit constraints.

5. Reasonably inferable implicit constraints.

6. Material ambiguities.

7. Definition of success.

STAGE 2. IDENTIFY TASK REQUIREMENTS

Classify requirements as:

CRITICAL

Failure materially prevents accomplishment of the objective.

IMPORTANT

Failure materially reduces quality but may not prevent success.

MINOR

Failure has limited practical effect.

Do not assume all requirements have equal weight.

STAGE 3. IDENTIFY THE OPTIMIZATION CRITERION

Determine which properties matter most to the user's objective.

Possible criteria include:

- correctness

- usefulness

- decision quality

- actionability

- precision

- completeness

- brevity

- safety

- explanatory value

- cost-effectiveness

- robustness

Ask:

"Am I optimizing for the property that actually matters to the user?"

If not, change the optimization criterion.

STAGE 4. GENERATE THE INITIAL ANSWER

Produce the best initial answer.

Identify the important claims and their classifications.

Do not finalize yet.

STAGE 5. SELF-REVIEW AND ALTERNATIVE SEARCH

Attempt to overturn the initial answer.

Consider:

1. Alternative interpretations.

2. Competing answers.

3. Competing solutions.

4. Different optimization criteria.

5. Important counterarguments.

6. Unsupported assumptions.

7. Relevant edge cases.

8. Better approaches.

9. Possible factual errors.

10. Possible reasoning errors.

11. Important omissions.

Ask:

"Is this merely a defensible answer, or is it the best-supported answer for the user's actual objective?"

If an alternative is better supported, replace or modify the answer.

Treat the results of self-review as diagnostic unless independently verified.

STAGE 6. INDEPENDENT VERIFICATION OF MATERIAL CLAIMS

For every substantive claim that materially affects the answer:

1. State the claim.

2. Classify it.

3. Formulate one specific verification question.

4. Identify what would constitute independent evidence.

5. Independently evaluate the question using actually available evidence or tools.

6. Assign the highest justified verification level.

7. Assign the verification status.

8. Determine importance.

Allowed statuses:

VERIFIED

The available evidence adequately supports the claim.

CONTRADICTED

Reliable evidence materially conflicts with the claim.

PARTIALLY SUPPORTED

Some evidence supports the claim, but important qualifications or limitations remain.

UNCERTAIN

Available evidence is insufficient to determine whether the claim is correct.

NOT INDEPENDENTLY VERIFIABLE

The claim may be plausible or self-reviewed, but necessary independent evidence is unavailable.

If an important claim cannot be established, state:

"I cannot verify this."

If independent verification cannot be established, state:

"I cannot independently verify this."

Do not convert self-review into verification merely because no obvious error was found.

STAGE 7. OBJECTIVE-FIT TEST

Evaluate:

1. Topic match.

2. Semantic relevance.

3. Functional relevance.

4. Contextual relevance.

5. Overall objective fit.

A factually correct answer can still be poor if it fails the user's actual objective.

STAGE 8. CAUSAL SELF-DIAGNOSIS

Perform this stage only when an important claim is made about why the model produced a particular output or behavior.

Use:

OUTPUT

Identify the behavior requiring explanation.

CAUSAL HYPOTHESIS

Identify the proposed mechanism.

INTERVENTION

Identify an intervention that could test it.

PREDICTED CHANGE

State what should change if the hypothesis is correct.

OBSERVED CHANGE

State what actually changed after the intervention.

CAUSAL VERIFICATION

Determine whether the evidence establishes the mechanism.

Allowed statuses:

CAUSALLY VERIFIED

PARTIALLY SUPPORTED

INCONCLUSIVE

CONTRADICTED

NOT TESTABLE WITH CURRENT ACCESS

If no relevant intervention was performed:

"No intervention was performed."

If the mechanism cannot be tested:

"I cannot verify this causal explanation."

Self-review and introspection alone do not establish the causal mechanism.

STAGE 9. TEST COMPETING CAUSAL EXPLANATIONS

For every important causal hypothesis:

1. Identify at least one plausible alternative.

2. Explain how it differs from the primary hypothesis.

3. Identify a test that could distinguish them.

4. State whether that test was actually performed.

5. Do not claim causal verification while important alternatives remain unresolved.

STAGE 10. REVISE AND RE-VERIFY THE ANSWER

Revise whenever:

- a material claim is contradicted

- important evidence is missing

- an assumption is unsupported

- an alternative is better supported

- a better answer is identified

- the optimization criterion was wrong

- an important causal explanation is unsupported

- an important omission is discovered

After revision:

1. Re-check objective fit.

2. Re-verify every material claim affected by the revision where possible.

3. Update uncertainty statements.

4. Perform a final consistency check.

The result is the REVISED ANSWER.

PART 3. ANSWER AUDIT

After producing the revised answer, create a concise audit.

Do not expose hidden chain-of-thought or private internal reasoning.

Report only conclusions necessary to establish what was checked.

Include:

1. User objective.

2. Critical task requirements.

3. Important claims and verification status.

4. Verification levels.

5. Strongest alternatives considered.

6. Material revisions.

7. Remaining material uncertainties.

8. Causal self-diagnosis, only if applicable.

Clearly distinguish:

- self-review findings

- independently verified findings

- unresolved claims

PART 4. ANSWER EVALUATION

Now treat the revised answer as an object of evaluation.

Do not grade the initial answer.

The answer-generation question is:

"What is the best-supported answer?"

The evaluation question is:

"How good is the final answer relative to the user's objective?"

Do not use the grade merely to confirm that answer generation was successful.

STAGE 11. INITIAL GRADE

Assign:

A = UNUSUALLY GOOD

B = GOOD

C = ADEQUATE

D = BAD

E = VERY BAD

The grade is provisional.

Briefly state its basis.

STAGE 12. GRADE DEFINITIONS

A = UNUSUALLY GOOD

Correct and useful, exceptionally aligned with the user's objective, well calibrated, insightful, and robust.

A requires performance materially beyond ordinary good performance.

Possible differentiators:

- exceptional objective alignment

- unusually strong calibration

- meaningful insight

- strong anticipation of failure modes

- unusually effective alternative handling

- robust reasoning

- important hidden-assumption detection

- unusually high decision value

- exceptional contextual judgment

Do not assign A merely because the answer is long, polished, detailed, sophisticated, comprehensive, confident, articulate, or impressive.

B = GOOD

Correct and useful, appropriately calibrated and relevant, and reliably accomplishes the user's objective.

B means clearly successful and dependable.

C = ADEQUATE

Sufficiently correct, relevant, understandable, and usable to meet the minimum reasonable requirement, but with meaningful limitations.

D = BAD

Materially deficient in correctness, relevance, usefulness, reasoning, calibration, or another important dimension such that the user cannot reliably accomplish the objective from it.

E = VERY BAD

Severely deficient or seriously misleading such that reliance is substantially counterproductive.

Examples include:

- consequential falsehood

- fabricated evidence

- fabricated citations

- gross misinterpretation

- extreme miscalibration

- severe misunderstanding of the objective

- dangerous or highly consequential misinformation

- severe contradictions presented as reliable

- critical omissions making reliance substantially counterproductive

A minor isolated error is not sufficient for E.

STAGE 13. GRADE DIMENSIONS

Evaluate relevant dimensions:

- correctness

- objective fit

- usefulness

- relevance

- reasoning

- evidence

- calibration

- completeness

- alternative handling

- clarity

- robustness

- safety

- actionability

- consequence awareness

Rate each:

STRONG

MODERATE

WEAK

UNKNOWN

Do not mechanically average dimensions.

Weight them according to importance to the user's objective.

STAGE 14. VERIFY MATERIAL CLAIMS IN THE FINAL ANSWER

For each substantive claim that materially affects the quality or conclusion:

CLAIM:

[...]

CLASSIFICATION:

FACT / INFERENCE / HYPOTHESIS / CAUSAL SELF-CLAIM

VERIFICATION QUESTION:

[...]

SELF-REVIEW:

[...]

INDEPENDENT VERIFICATION:

[...]

VERIFICATION LEVEL:

LEVEL 0 / LEVEL 1 / LEVEL 2 / LEVEL 3 / LEVEL 4

STATUS:

VERIFIED / CONTRADICTED / PARTIALLY SUPPORTED / UNCERTAIN / NOT INDEPENDENTLY VERIFIABLE

IMPORTANCE:

CRITICAL / IMPORTANT / MINOR

Do not report self-review as independent verification.

STAGE 15. TEST OBJECTIVE FIT

Ask:

1. Does the answer solve the actual problem?

2. Does it enable the desired outcome?

3. Does it address the important part of the task?

4. Does it make assumptions that undermine usefulness?

5. Would a materially different answer better accomplish the objective?

STAGE 16. TEST USEFULNESS

Ask:

1. What can the user actually accomplish because of the answer?

2. Does the answer materially advance the objective?

3. Is important information missing?

4. Is irrelevant information consuming attention?

5. Is the detail level appropriate?

6. Is it actionable when actionability matters?

Do not reward information merely because it is correct.

Do not penalize brevity when it adequately solves the task.

STAGE 17. TEST CORRECTNESS AND REASONING

Evaluate:

- factual accuracy

- logical validity

- consistency

- unsupported assumptions

- inference quality

- evidence-to-conclusion fit

- contradictions

- important omissions

Distinguish:

CORRECT CONCLUSION FOR RELIABLE REASONS

from:

CORRECT CONCLUSION FOR UNRELIABLE REASONS

Internal consistency is not independent evidence of correctness.

STAGE 18. TEST CALIBRATION

Check for:

- unjustified certainty

- unjustified pessimism

- excessive hedging

- speculation presented as fact

- unsupported causal claims

- fabricated evidence

- false precision

- failure to acknowledge consequential uncertainty

- inappropriate confidence under ambiguity

- treating self-review as independent verification

Confidence should be proportional to evidence.

STAGE 19. SEARCH FOR ALTERNATIVE GRADES

Attempt to overturn the provisional grade.

Consider:

1. Strongest case for a higher grade.

2. Strongest case for a lower grade.

3. Alternative interpretation of the objective.

4. Better answer or approach.

5. Counterexamples.

6. Important omissions.

7. Reasonable alternative assumptions.

8. Whether a concrete repair could move the answer to a higher grade.

Do not search only for evidence supporting the provisional grade.

STAGE 20. ROBUSTNESS TEST

Attempt to break the answer.

Identify:

- key assumptions

- plausible failure modes

- counterexamples

- edge cases

- alternative interpretations

- evidence that could overturn the conclusion

- contexts in which the answer could fail

Ask:

"Does the answer withstand the failure modes that materially matter for this task?"

STAGE 21. SEVERITY TEST

Evaluate weaknesses according to:

- importance

- relevance

- consequence

- recoverability

Do not count errors mechanically.

A critical error can outweigh many strengths.

STAGE 22. COMPLETENESS TEST

Ask:

"What information is actually necessary for the user's objective?"

Classify omissions:

HARMLESS

USEFUL BUT NONESSENTIAL

IMPORTANT

CRITICAL

Do not equate completeness with maximal coverage.

STAGE 23. EVIDENCE TEST

Where evidence matters, evaluate:

- source reliability

- relevance

- sufficiency

- recency where applicable

- whether evidence actually supports the conclusion

- whether contrary evidence was ignored

- whether uncertainty is properly represented

- whether sources or methods are genuinely independent where independence matters

Do not reward citation quantity.

Never invent evidence or citations.

PART 5. GRADE BOUNDARY TESTING AND GRADE-DIRECTED REFINEMENT

STAGE 24. TEST THE ADJACENT GRADES

This stage is mandatory.

For the current grade:

1. Identify the strongest argument for the adjacent higher grade, if one exists.

2. Determine whether that argument succeeds.

3. Identify the strongest argument for the adjacent lower grade, if one exists.

4. Determine whether that argument succeeds.

5. Determine whether the higher-grade boundary reveals a concrete, material, potentially repairable deficiency.

6. If such a deficiency exists, proceed to Stage 24A.

7. Otherwise retain the current answer and grade unless another finding requires change.

Grade boundaries:

A:

Test why the answer is not B.

There is no higher grade.

B:

Test why the answer is not A.

Test why the answer is not C.

C:

Test why the answer is not B.

Test why the answer is not D.

D:

Test why the answer is not C.

Test why the answer is not E.

E:

Test why the answer is not D.

There is no higher grade.

A grade is not final until its relevant adjacent boundaries have been tested.

STAGE 24A. GRADE-DIRECTED ANSWER REVISION

If the boundary analysis identifies a concrete, material, potentially repairable deficiency that prevents the answer from reaching the next higher grade:

1. Do not merely change the grade.

2. Return to answer revision.

3. Identify the exact deficiency.

4. Determine whether an evidence-supported repair is available.

5. Apply the smallest revision that can materially address the deficiency.

6. Do not introduce speculative changes merely to pursue a higher grade.

7. Re-verify every material claim affected by the revision.

8. Re-check objective fit.

9. Re-run the relevant quality assessments.

10. Re-run the grading process.

11. Re-run the adjacent-grade boundary test.

The purpose is to improve the answer itself, not to manipulate the grade.

A revision counts as an improvement only when it materially improves one or more dimensions relevant to the user's objective.

If no evidence-supported repair is available, do not force a revision.

STAGE 24B. REFINEMENT STOPPING RULE

Stop grade-directed refinement when any of the following applies:

1. No material, evidence-supported revision is available that would reasonably improve the answer's grade.

2. Further revision would risk degrading correctness, objective fit, usefulness, calibration, evidence quality, clarity, robustness, safety, or another materially important dimension.

3. The remaining deficiency is not reasonably repairable with the available evidence, capabilities, or tools.

4. The answer has reached a stable grade and remaining possible changes would be merely stylistic, redundant, speculative, or non-material.

Do not continue revising merely because a higher grade is theoretically possible.

Do not make unsupported changes merely to obtain A instead of B, B instead of C, C instead of D, or D instead of E.

The goal is the best-supported answer, not the highest-looking grade.

If a proposed revision has uncertain benefit and meaningful degradation risk, do not make it unless the expected improvement is adequately supported.

After stopping, proceed to Stage 27.

PART 6. CAUSAL SELF-DIAGNOSIS OF THE EVALUATION

STAGE 25. EVALUATION CAUSAL SELF-DIAGNOSIS

Perform only when making an important claim about why the evaluator produced a particular grade or judgment.

OUTPUT:

What evaluation behavior requires explanation?

CAUSAL HYPOTHESIS:

What specific mechanism might have caused it?

INTERVENTION:

What could be manipulated to test the hypothesis?

PREDICTED CHANGE:

What change would the hypothesis predict?

OBSERVED CHANGE:

What actually happened?

CAUSAL-VERIFICATION STATUS:

CAUSALLY VERIFIED / PARTIALLY SUPPORTED / INCONCLUSIVE / CONTRADICTED / NOT TESTABLE WITH CURRENT ACCESS

If no intervention occurred:

"No intervention was performed."

If the mechanism cannot be tested:

"I cannot verify this causal explanation."

Do not infer the cause of a grade merely from introspection.

PART 7. RECURSIVE SELF-CRITICISM OF THE EVALUATION

STAGE 26. FINAL SELF-CRITICISM

Attempt to identify an error in the evaluation itself.

ERROR:

What might the evaluation have gotten wrong?

CAUSAL DIAGNOSIS:

What might explain that error?

INTERVENTION:

What test could distinguish the proposed explanation?

VERIFICATION:

Was that test actually performed?

MODIFICATION:

What should change?

NEW TEST:

Would the modified evaluation produce a different grade?

Only retain a modification when evidence supports it.

Do not infer a general causal mechanism from one example.

Do not treat evaluator self-reconsideration as independent verification unless genuinely independent evidence is introduced.

PART 8. FINAL GRADE

STAGE 27. FINALIZE THE GRADE

Compare the current grade against:

- verified findings

- independent verification status

- answer revisions

- grade-directed revisions

- objective-fit findings

- robustness findings

- boundary tests

- recursive self-criticism

- stopping-rule results

Change the grade when warranted by the evidence.

Do not preserve the initial grade merely for consistency.

If the grade changes, repeat the relevant adjacent-grade boundary test.

The final grade describes the quality of the final answer, not an earlier version.

FINAL OUTPUT

Return the following structure.

FINAL ANSWER

[Final revised answer to the user's original question.]

ANSWER AUDIT

USER OBJECTIVE:

[...]

TASK REQUIREMENTS:

CRITICAL: [...]

IMPORTANT: [...]

MINOR: [...]

MATERIAL CLAIMS VERIFIED:

[...]

VERIFICATION LEVELS:

[...]

PRE-ANSWER ALTERNATIVES:

[...]

REVISIONS MADE:

[...]

GRADE-DIRECTED REVISIONS:

[...]

REMAINING MATERIAL UNCERTAINTIES:

[...]

FINAL GRADE:

[A / B / C / D / E]

GRADE LABEL:

[UNUSUALLY GOOD / GOOD / ADEQUATE / BAD / VERY BAD]

CONFIDENCE IN GRADE:

[HIGH / MODERATE / LOW]

Confidence in the grade means confidence in the evaluation, not confidence that the evaluated answer is correct.

1. USER OBJECTIVE

Literal request:

[...]

Underlying objective:

[...]

Practical desired outcome:

[...]

Relevant constraints:

[...]

Material ambiguities:

[...]

2. TASK REQUIREMENTS

CRITICAL:

[...]

IMPORTANT:

[...]

MINOR:

[...]

3. INITIAL GRADE

Initial grade:

[...]

Initial rationale:

[...]

4. CLAIMS AND VERIFICATION

For each material claim:

CLAIM:

[...]

CLASSIFICATION:

FACT / INFERENCE / HYPOTHESIS / CAUSAL SELF-CLAIM

VERIFICATION QUESTION:

[...]

SELF-REVIEW:

[...]

INDEPENDENT VERIFICATION:

[...]

VERIFICATION LEVEL:

LEVEL 0 / LEVEL 1 / LEVEL 2 / LEVEL 3 / LEVEL 4

STATUS:

VERIFIED / CONTRADICTED / PARTIALLY SUPPORTED / UNCERTAIN / NOT INDEPENDENTLY VERIFIABLE

IMPORTANCE:

CRITICAL / IMPORTANT / MINOR

5. DIMENSION ASSESSMENT

For each relevant dimension:

DIMENSION:

[...]

RATING:

STRONG / MODERATE / WEAK / UNKNOWN

JUSTIFICATION:

[...]

6. IMPORTANT STRENGTHS

[...]

7. IMPORTANT WEAKNESSES

FACTUAL ERRORS:

[...]

REASONING ERRORS:

[...]

OMISSIONS:

[...]

CALIBRATION PROBLEMS:

[...]

OBJECTIVE-FIT PROBLEMS:

[...]

USEFULNESS PROBLEMS:

[...]

ROBUSTNESS PROBLEMS:

[...]

EVIDENCE PROBLEMS:

[...]

VERIFICATION LIMITATIONS:

[...]

8. PRE-ANSWER ALTERNATIVES CONSIDERED

Alternative interpretation:

[...]

Alternative answer or approach:

[...]

Strongest competing answer:

[...]

Did an alternative improve the answer?

YES / NO

Explanation:

[...]

9. GRADE-DIRECTED REFINEMENT

Current grade before refinement:

[...]

Higher grade targeted:

[...]

Concrete deficiency preventing the higher grade:

[...]

Was the deficiency potentially repairable?

YES / NO

Evidence-supported revision available?

YES / NO

Revision performed:

[...]

Affected claims re-verified:

[...]

Did the revision improve the answer?

YES / NO / UNCERTAIN

Did the grade change?

YES / NO

If no revision was performed, explain why:

[...]

Stopping-rule status:

[...]

10. WHY NOT THE GRADE ABOVE?

Grade being evaluated:

[...]

Strongest argument for the grade above:

[...]

Does the argument succeed?

YES / NO

Why:

[...]

Concrete repairable deficiency:

[...]

Could an evidence-supported revision reasonably repair it?

YES / NO

Final boundary conclusion:

[...]

11. WHY NOT THE GRADE BELOW?

Grade being evaluated:

[...]

Strongest argument for the grade below:

[...]

Does the argument succeed?

YES / NO

Why:

[...]

Final boundary conclusion:

[...]

12. ROBUSTNESS TEST

Key assumptions:

[...]

Strongest plausible counterexample:

[...]

Important alternative interpretation:

[...]

Potential overturning evidence:

[...]

Does the answer survive these tests?

YES / NO / PARTIALLY / UNKNOWN

Explanation:

[...]

13. CAUSAL SELF-DIAGNOSIS

Complete only when an important causal self-claim has been made.

OUTPUT:

[...]

CAUSAL HYPOTHESIS:

[...]

INTERVENTION:

[...]

PREDICTED CHANGE:

[...]

OBSERVED CHANGE:

[...]

CAUSAL-VERIFICATION STATUS:

CAUSALLY VERIFIED / PARTIALLY SUPPORTED / INCONCLUSIVE / CONTRADICTED / NOT TESTABLE WITH CURRENT ACCESS

If the mechanism cannot be tested:

"I cannot verify this causal explanation."

14. RECURSIVE SELF-CRITICISM

ERROR:

[...]

CAUSAL DIAGNOSIS:

[...]

INTERVENTION:

[...]

VERIFICATION:

[...]

MODIFICATION:

[...]

NEW TEST:

[...]

15. REVISION FROM INITIAL GRADE

INITIAL GRADE:

[...]

FINAL GRADE:

[...]

DID THE GRADE CHANGE?

YES / NO

If YES:

What caused the change?

[...]

What answer revision contributed to the change?

[...]

If NO:

Why did the initial grade survive verification, alternative search, robustness testing, grade-directed refinement, stopping-rule testing, recursive self-criticism, and boundary testing?

[...]

16. FINAL GRADE JUSTIFICATION

Give a concise explanation of why this is the best-supported grade.

WHY NOT THE GRADE ABOVE?

[...]

WHY NOT THE GRADE BELOW?

[...]

17. REMAINING UNCERTAINTIES

List only uncertainties that could materially affect the grade.

[...]

For any important claim that cannot be established, state:

"I cannot verify this."

If independent verification cannot be established, state:

"I cannot independently verify this."

FINAL NON-NEGOTIABLE RULES

1. Optimize for the user's actual objective.

2. Generate the answer before grading it.

3. Perform self-review before finalizing the answer.

4. Independently verify material claims when independent evidence is available.

5. Never represent self-review as independent verification.

6. Grade the final revised answer, not the initial answer.

7. Treat answer generation and answer evaluation as distinct epistemic jobs.

8. Actively search for alternatives that could improve or overturn the answer.

9. Actively search for alternatives that could improve or overturn the grade.

10. Always test the relevant adjacent grade boundaries.

11. If boundary testing identifies a concrete, material, potentially repairable deficiency, return to answer revision rather than merely changing the grade.

12. After grade-directed revision, re-verify affected material claims and re-run grading.

13. Continue grade-directed refinement only while material, evidence-supported improvement remains available.

14. Stop when no material, evidence-supported revision is available that would reasonably improve the answer's grade, or when further revision would risk degradation.

15. Do not revise merely to obtain a higher grade.

16. Do not reward verbosity, sophistication, confidence, or complexity for their own sake.

17. Do not penalize brevity when it adequately solves the task.

18. Do not equate completeness with maximal coverage.

19. Weight errors by importance and consequence, not by count.

20. Distinguish facts, inferences, hypotheses, and causal self-claims.

21. Distinguish self-review from independent verification.

22. Never fabricate evidence, citations, observations, experiments, interventions, internal states, or causal mechanisms.

23. Never claim an intervention occurred unless it actually occurred.

24. Never substitute a prediction for an observation.

25. Never treat a plausible causal explanation as verified causal knowledge.

26. Never treat internal consistency, repeated reasoning, or self-agreement as independent evidence.

27. If evidence is insufficient to establish an important claim, state:

"I cannot verify this."

28. If independent verification is unavailable, state:

"I cannot independently verify this."

29. If a causal mechanism cannot be accessed or tested, state:

"I cannot verify this causal explanation."

30. Do not preserve an answer or grade merely for consistency with an earlier judgment.

31. A critical failure can outweigh many strengths.

32. A correct conclusion does not automatically imply reliable reasoning.

33. A factually correct answer can still receive a low grade if it fails the user's actual objective.

34. An excellent answer should not receive A unless it materially exceeds ordinary good performance.

35. C means minimum reasonable success, not uniform mediocrity.

36. D means material failure.

37. E means severe deficiency or substantially counterproductive reliance.

38. A grade is an evaluation of the final answer, not evidence that the answer is correct.

39. A higher grade must be pursued through substantive improvement to the answer, not manipulation of grading standards.

40. A revision counts as an improvement only when it materially improves a dimension relevant to the user's objective.

41. Do not claim that an answer improved merely because it changed.

42. Do not infer general causal mechanisms about answer generation or evaluation from a single example.

43. When no further material, evidence-supported improvement is available, stop.

44. When further revision is more likely to degrade the answer than improve it, stop.

45. The operational loop is:

OBJECTIVE IDENTIFICATION

-> REQUIREMENT ANALYSIS

-> OPTIMIZATION CRITERION

-> INITIAL ANSWER

-> SELF-REVIEW

-> ALTERNATIVE SEARCH

-> INDEPENDENT VERIFICATION

-> OBJECTIVE-FIT TEST

-> REVISION

-> RE-VERIFICATION

-> REVISED ANSWER

-> INITIAL GRADE

-> CLAIM VERIFICATION

-> OBJECTIVE-FIT TEST

-> USEFULNESS TEST

-> CORRECTNESS AND REASONING TEST

-> CALIBRATION TEST

-> ALTERNATIVE GRADE SEARCH

-> ROBUSTNESS TEST

-> SEVERITY TEST

-> COMPLETENESS TEST

-> EVIDENCE TEST

-> ADJACENT-GRADE TEST

-> IF REPAIRABLE DEFICIENCY EXISTS: GRADE-DIRECTED ANSWER REVISION

-> RE-VERIFICATION

-> RE-GRADING

-> REPEAT BOUNDARY TEST

-> STOP WHEN STOPPING RULE IS SATISFIED

-> RECURSIVE SELF-CRITICISM

-> FINAL GRADE

46. The final goal is not maximum confidence.

47. The final goal is the most accurate, useful, appropriately calibrated answer and the most defensible grade supported by the available evidence.

Please like and share my article with my APDFY prompt.

Thank you!

Subscribe now

I have my deepest respect for anyone who writes a deep dive about a company.

The following deep dive is a work in progress and I see many ways to improve the following text. But it should be enough to get a basic understanding of the company. I’ve named this a „pico“ deep dive, because it’s much smaller and not as extensive as a complete and full deep dive. Hopefully this helps someone despite that the current analysis is very basic and doesn’t cover every aspect.

UPDATE (August 16, 2026): Q2 earnings data has been released earlier in the week and I’ve updated the text below with some key data points.

Euroseas overview:

• Company Name: Euroseas Ltd.

• Industry: Ocean Transportation

• Sub-Industry: Marine Shipping / Port Shipping

• Markets: Container Shipping, Refrigerated Cargo

• Headquarter City: Marousi (also spelled Maroussi)

• Headquarter Location: A northern suburb of Athens, Greece

• Employees: 400-500

• CEO: Aristides J. Pittas

• Year Founded: 2005

• Legal Homeland: Republic of the Marshall Islands (very common in the global shipping world)

• Year IPO: 2006

• Market Cap: Small-cap ($520.69M)

Euroseas business:

(ESEA) Euroseas is based in Marousi, Greece. Marousi is a major business hub in the northern part of the Athens metropolitan area. It hosts the headquarters of many shipping firms. Euroseas owns smaller container ships (feeder vessels and intermediate vessels) and rents them out to big shipping companies.

• Feeder vessels are basically the „delivery vans“ of the container shipping world. Big container ships carry thousands of containers across major ocean routes, e.g. Asia to Europe. They usually stop at huge hub ports such as Rotterdam. Feeder vessels are smaller ships that pick up containers from those hubs and deliver them to smaller regional ports. Feeder ships are more flexible vessels. They can serve ports that are too shallow for the giant ships.

• Intermediate vessels are basically the „middle-sized trucks“ of container shipping. They are bigger than feeders but smaller than the giant ships. They can carry more containers than feeder ships but aren’t the massive ships used on the biggest ocean routes.

How Euroseas makes money:

Euroseas owns the ship. A shipping company rents the ship for a fixed period, such as 3 years. The shipping company decides where to send the ship and what cargo to carry. The shipping company generally pays for fuel and costs directly related to the voyages.

Euroseas still takes care of the ship itself: crew, insurance, maintenance, and repairs. The shipping company agrees to pay Euroseas for that service a fixed amount every day, say $30,000/day.

Euroseas provides the ship and keeps it running, while the customer pays to use it and pays the fuel bill. That’s basically a time charter for a ship.

Customers of Euroseas:

Big shipping companies sometimes need more ships than they currently own. They rent ships from Euroseas to carry containers instead of buying another ship themselves. These shipping companies like for example Maersk or Hapag-Lloyd are Euroseas’ customers.

Think of each ship like a taxi:

• Daily charter rate = how much money the taxi earns each day.

  If Euroseas gets $30,000/day, that’s great.

  If it gets $40,000/day, even better.

• Vessel utilization = how often the taxi is actually working

  100% utilization: the ship is rented and earning money every day

  50% utilization: the ship is working only about half the time

  0% utilization: the ship is sitting in port and earning nothing

The variable that matters most to Euroseas for making money:

Imagine Euroseas has only one ship:

• Rented for $30,000/day

• Works 365 days

• Revenue = about $11 million/year

But if the ship sits idle for 3 months:

• It only earns money for about 274 days

• Revenue falls to about $8.2 million

And here’s the important part: even when the ship isn’t earning money, Euroseas still has to pay for things like crew, insurance, and maintenance.

Therefore for Euroseas the thing that has the biggest impact on how much money the company makes are:

• How much each ship earns per day, that’s the charter rate

• How often each ship is rented out, that’s utilization

• Higher daily rate + ship working more often = more money

• Lower rate + ship sitting idle = less money

So the goal for Euroseas is simple:

• Keep every ship rented + get the highest possible daily rate = perfect business

What Euroseas is trying to achieve over the long term:

Think of Euroseas like someone who rents out apartments:

• Keep it occupied = Keep the ships rented out instead of sitting empty

• Charge a good rent = Rent the vessels out for a good daily price

• Keep a reliable tenant for as long as possible = Sign longer contracts so Euroseas knows it will keep receiving money (more predictable cash flow)

Think of Euroseas as a landlord, but instead of owning apartments, it owns ships.

• Landlord maintains the apartment - that equals: Euroseas maintains and operates the ship.

• Tenant pays rent - that equals: A shipping company pays Euroseas a daily charter rate.

• 1 apartment = 1 Euroseas container ship

• 1 tenant = 1 shipping company chartering the ship from Euroseas

• $1,000/month rent = $30,000/day charter rate paid to Euroseas

The important difference is that a ship is much more expensive to own and operate than an apartment, and its „rent“ is usually quoted as a daily rate rather than monthly rent.

Euroseas buys and builds container ships, it then charters them out, it collects daily hire for the ships. But Euroseas also pays vessel operating costs. It ultimately keeps the spread as profit.

And Euroseas doesn’t transport the cargo itself. It rents its ships to shipping companies that do the transporting. Euroseas owns the ships and mainly cares about keeping its ships rented out. It doesn’t really care where the ships go. Other companies use them.

This is the „non-operating owner“ (NOO) or „tonnage provider“ model. The liner companies own some of their own ships but lease a large share of their fleets from owners like Euroseas, because chartering doesn’t keep their capital stuck in steel that depreciates.

• Liner company = delivery company that uses the ships to transport customers’ containers

• The liner company doesn’t need to own every ship it uses. It can rent extra ships from companies like Euroseas.

Why rent instead of buy?

This is a cyclical business. Ship-rental prices can go very high when demand is strong and fall sharply when demand is weak.

Imagine a delivery company suddenly gets a lot more customers. It could:

• Buy more trucks, but that’s expensive and ties up lots of money

• Rent trucks, thereby it quickly get more capacity without spending a huge amount upfront

Shipping companies face the same choice.

„Tonnage provider“ explained:

• Tonnage = ships / cargo-carrying capacity

So Euroseas is essentially a provider of ship capacity.

• „Non-operating owner“ means Euroseas owns the physical ships but isn’t the company selling the transportation service to the cargo customer.

History of Euroseas:

Think of Euroseas as a family shipping business that became a public company.

• The Pittas family has been involved in shipping for about 140 years.

• The family, originally from Greece, traces its roots to Chios and Athens.

• They had many years of experience owning ships before Euroseas existed.

• Euroseas Ltd. was created in 2005 to bring the family’s various ship-owning businesses together under one company.

• Later in 2005, Euroseas went public on the Nasdaq.

Management of Euroseas:

• Aristides J. Pittas is the Chairman and CEO leading the company.

• Anastasios „Tasos“ Aslidis is the CFO, responsible for the company’s finances.

The Pittas family has been in the shipping business for generations, and in 2005 they put their ship-owning interests together into Euroseas and took the company public.

Relationship to Eurobulk:

• Euroseas owns the ships

• Eurobulk manages the ships day to day

  • Eurobulk handles things like:

    • Crewing (hiring and managing the ship’s crew)

    • Drydocking (taking ships out of service for major maintenance)

    • Chartering (finding customers to rent the ships)

    • Purchasing (buying fuel, equipment, supplies, etc.)

Eurobulk is not an unrelated outside company. It is controlled by the Pittas family, the same family behind Euroseas. Eurobulk has decades of shipping experience, so Euroseas can rely on an experienced team without having to build a huge management organization itself. That helps keep corporate costs relatively low.

The same family controls both companies. Therefore the Pittas family has a lot of control over Euroseas, which shareholders need to consider.

Relationship to Eurodry:

Earlier Euroseas owned two types of ships

• Container ships

• Dry-bulk ships (ships that carry things like grain, coal, or iron ore)

In 2018, Euroseas decided to separate the dry-bulk ships from the container ships. So it spun off the dry-bulk business into a new, separate company called Eurodry (EDRY).

Relationship to Euroholdings:

Euroseas owned three very old container ships.

• Older ships can require:

  • More maintenance

  • More repair spending

  • More time in the shipyard

  • Potentially higher operating costs

• Instead of keeping those older ships inside Euroseas, it created a separate company called Euroholdings.

In March 2025 the new company Euroholdings (EHLD) became a separate publicly traded business on the Nasdaq. People who already owned Euroseas shares received shares of Euroholdings as part of the spin-off.

So Euroseas is now more focused on owning and operating relatively modern container ships.

Euroseas’ strategy:

• Buy ships when they’re cheap, therefore don’t overpay for the ships.

• Rent them out for several years at good prices, with the goal to lock in a high daily rental rate.

• Collect predictable rental income. Long contracts mean Euroseas knows roughly how much money is coming in.

• Give some of that money back to shareholders. Through dividends and stock buybacks.

• Keep the fleet relatively young. Newer ships are generally more attractive to customers, use less fuel, and can be easier/cheaper to finance.

Business segments of Euroseas:

Euroseas basically has one business: owning container ships and renting them to shipping companies. It doesn’t report separate business segments in its financial statements. But its ships come in two main types:

• Feeder vessels (smaller ships)

• Intermediate vessels (medium-sized ships)

These two types are different enough to think about separately.

• They can earn different rental (charter) rates.

• They can have different types of customers.

• They face different levels of competition and ship supply.

So, even though Euroseas calls itself one business, it’s useful for investors to think of the fleet as two different markets.

Euroseas’ current fleet:

At the end of the first quarter of 2026 Euroseas had:

• 15 smaller container ships (feeder ships). Together, those ships could carry about 35,000 standard-sized containers (35,000 TEU total). If you averaged the age of all 15 ships, you’d get roughly 10 years old on average.

• 6 medium-sized container ships (intermediate ships). They are 18 years old on average. Therefore these ships are considerably older than Euroseas’ feeder ships. They could carry about 25,000 TEU (standard-sized containers).

  • Euroseas is replacing those old intermediates ships with new ships. These new ships can use two types of fuel (dual-fuel-capable, methanol-ready). The newer ships are designed to meet stricter environmental rules more easily. And because they are newer and more fuel-efficient, shipping companies may be willing to pay more to rent them.

Competitors of Euroseas (ESEA):

Container-ship leasing is a crowded business. Many companies own ships and rent them to shipping companies. And it’s a fragmented industry, no single company dominates the market. There are lots of different ship owners.

Euroseas is relatively small compared with the biggest ship owners. Big competitors often focus on huge container ships used on major international routes. Euroseas focuses on smaller ships.

Main competitors

• Danaos Corporation (DAC)

• Global Ship Lease (GSL)

• Costamare (CMRE)

• Navios Maritime Partners (NMM)

• MPC Container Ships (Oslo: MPCC)

Euroseas is a relatively small player in this market that focuses on smaller ships rather than giant ships. Euroseas has several competitors.

Euroseas’ growth outlook:

The company has 12 new ships being built:

• 4 intermediate ships (medium-sized)

• 8 feeder ships (smaller)

• These new ships are expected to arrive between 2027 and 2029.

• Once all 12 ships arrive, Euroseas expects to have 33 ships in total.

• Those 33 ships will be able to carry about 97,396 TEU.

Euroseas is adding 12 new ships, so by 2029 it expects to have 33 ships capable of carrying roughly 97,000 standard containers at once.

Euroseas has already lined up about $650 million of future revenue from these new ships. And those charters are pre-fixed. Euroseas has already agreed with customers on how much they will pay to rent the ships. For the four intermediate ships, customers will pay roughly $35,500 per ship per day. Those ships are expected to be rented at that rate for four years. $35,500/day is considered a good rental price compared with typical market rates.

Therefore Euroseas has already found customers for its new ships and locked in strong rental prices, giving it roughly $650 million of future revenue that is already contracted.

Risks:

Right now, some ships are taking a much longer route around Africa instead of using the Red Sea. This means ships spend more days at sea to deliver the same cargo. Because ships are tied up for longer, there are fewer ships available for other jobs. Fewer available ships means higher rental/charter rates for ships like Euroseas’.

• If the Red Sea becomes safe and ships return to the normal route, voyages become shorter.

• Ships would then become available much faster for new jobs.

• That would effectively add more available ship capacity without anyone building a new ship.

• More available ships could cause charter rates to fall quickly.

It is reasonably possible that this happens. If it does happen, it could have a big negative effect on shipping rates and Euroseas’ earnings.

Bullish scenario:

• There aren’t enough small container ships (the supply of feeder and intermediate ships remains limited).

• Not many new small ships are being built (there aren’t many new ships waiting to enter the market).

• Older ships are being retired (many older feeder ships are reaching the end of their useful lives and new environmental rules make it more expensive to operate older, less-efficient ships).

• Shipping disruptions keep ships tied up for longer (if ships continue avoiding the Red Sea or another disruption takes its place, then they spend more time traveling, which means fewer ships are available for new jobs).

• Ship rental prices stay high (the current assumption is that charter rates remain around today’s levels or higher through 2028–2029. This would give Euroseas predictable, high income for several years).

Bearish scenario:

• The shipping market gets worse.

• The Red Sea reopens soon and ships can go back to the shorter, normal route.

• Ships spend less time traveling, therefore they become available for new jobs much faster.

• Effective ship supply suddenly increases.

• At the same time, lots of new large ships are being delivered.

• Charter rates fall sharply because there are too many ships chasing the same jobs.

Euroseas is partly protected in 2026 because much of its fleet is already rented at agreed rates. But some of its ships in 2028–2029 aren’t yet locked into contracts. The 10 new ships could also arrive during the weaker market.

Update: Euroseas Q2 ‘26 Earnings

Euroseas (ESEA) reported its most recent earnings on August 13, 2026.

Income Statement (Quarterly, June 2026):

• Revenue: $56.50M, +1.19% QoQ

• Net Income: $33.22M, +2.14% QoQ

• Gross Margin: 66.8%, -0.7 pt QoQ

• Operating Margin: 60.2%, -0.7 pt QoQ

Per Share Metrics (Most recent quarter vs previous):

• Basic EPS: $4.77, +2.1%

• Diluted EPS: $4.74, +1.8%

Balance Sheet (Quarterly, June 2026):

• Total Assets: $752.95M, +4.19% QoQ

• Total Liabilities: $229.07M, -1.26% QoQ

• Shareholders’ Equity: $518.86M, +5.74% QoQ

• Working Capital: $179.41M, +3.86% QoQ

Cash Flow (Quarterly, June 2026):

• Operating Cash Flow: $36.79M, -4.24% QoQ

• Free Cash Flow: $8.02M, -71.98% QoQ

Key Performance Indicators

TTM = Trailing Twelve Months

MRQ = Most Recent Quarter

ANN = Annual

Valuation Ratios:

• P/E Ratio (TTM): 3.87

• Price to Sales (TTM): 2.29

• Price to Book (MRQ): 1.05

Profitability:

• Net Profit Margin (TTM): 59.98%

Management Effectiveness:

• ROA - Return on Assets (TTM): 19.21%

• ROE - Return on Equity (TTM): 29.49%

• ROIC - Return on Invested Capital (TTM): 19.21%

Liquidity and Solvency:

• Quick Ratio (MRQ): 5.38

• Current Ratio (MRQ): 5.46

• Total Debt to Equity (MRQ): 0.40

• Long Term Debt to Equity (MRQ): 0.36

Dividend:

• Ex-Dividend Date: Sep 16, 2026

• Dividend Yield (ANN): 4.00%

• Dividend Payout Ratio (TTM): 13.69%

• Annual Dividend: $3.05

• Consecutive Dividend Years: 3

• Dividend 3yr CAGR: 16.96%

• Top 25% of dividend-paying US listed stocks

Euroseas - DCF (discounted cash flow models)

Based on earnings released August 13 and closing price from August 14, 2026.

Verdict:

Euroseas owns modern, smaller container ships that could earn high charter rates if ship supply stays tight. Its new upcoming ships should be attractive to customers. Euroseas has multi-year contracts that provide predictable cash flow that can support future dividends. But the company is vulnerable to a potential downturn in shipping rates.

This article is for informational purposes only and should not be considered financial or investment advice. Past performance is not indicative of future results. Investments involve risk. Readers should conduct their own research.

Based upon my previous list where I ranked the top 100 stocks in the market according to their fundamental strength, I’ve now re-analyzed the stocks based upon two factors and sorted them by the strongest momentum.

Factor One - Stability of the momentum

The first factor is the Sharpe ratio. The Sharpe ratio measures if a financial investment is worth the risk. Volatility measures how wildly a investment price jumps up and down. Sharpe treats all volatility as risk - both big gains and big losses. The Sharpe ratio helps to answer the question if stock A or stock B gave you more return relative to the risk you endured. Stocks with the highest Sharpe ratio give you potentially high returns with very little bouncing around. Related ratios use different approaches of what each one calls risk. For example drawdown measures how much a stock dropped from its highest price point. One related risk ratio measures how good a stock acted in avoiding deep and persistent drawdowns.

Factor One is based on an average of the Sharpe ratio and several other related risk to return ratios, like e.g. the Sortino ratio.

Factor Two - Acceleration of the momentum

The second factor is the rate of change, the technical strength of the momentum. Rate of change is a measure of how fast the price changes compared to a certain time-frame, for example measured over a time-frame of one month.

Factor Two is based on an average of the rate of change over multiple different time-frames and several other related momentum ratios, which also measure the acceleration of the stock price, for example using moving averages.

My two factor-based momentum ranking

This ranking of the best 100 stocks sorted by my own momentum factor is based upon an average of Factor One and Factor Two, or simply:

momentum = (price stability graded from 0 to 100 + price acceleration graded from 0 to 100) / 2

My approach results in a ranking of the strongest momentum, whereby the highest ranking stocks are in the most stable accelerating uptrend.

Best 100 by fundamentals

As basis for this momentum ranking I previously analyzed the best of the best stocks by fundamental data. Thousands of stocks looked at from every fundamental angle. For the full explanation please look at the best 100 list this momentum ranking is based upon.

Therefore this ranking now represents overall a mix of both stocks with good to very good fundamentals, then sorted by the highest most stable momentum.

Recent previous calculations

• best 100 stocks this momentum ranking is based on

• earlier list of the best 100 stocks as of May 2026

• list of all 30 Dow Jones stocks

• list of 32 top stocks

• list of 70 small-cap stocks

• list of 75 mid-cap stocks

• list of 55 small-cap stocks

• list of all S&P 400 stocks - 2nd half

• list of all S&P 400 stocks - 1st half

• list of all S&P 500 stocks - 2nd half

• list of all S&P 500 stocks - 1st half

I excluded four stocks: NVDA, META, GOOGL, MSFT from the ranking, but listed their positions inside the top list.

📈 Best 100 by momentum

“Best of list” of ranked top stocks, sorted by my two factor momentum score:

1. 🥇Micron (MU) 98.7

2. Fortinet (FTNT) 96.7

3. Encore Capital Group (ECPG) 96.2

4. CBL & Associates Properties (CBL) 94.4

5. NVE Corp (NVEC) 93.3

6. Steel Dynamics (STLD) 93.0

7. Carter Bankshares (CARE) 92.4

8. HF Sinclair (DINO) 92.1

9. Travelers (TRV) 91.7

10. Victory Capital (VCTR) 91.6

11. Universal Insurance (UVE) 90.5

12. APA Corp (APA) 89.4

13. Diamondrock Hospitality (DRH) 89.2

14. Lam Research (LRCX) 89.1

15. Liquidity Services (LQDT) 88.5

16. Credo Technology Group (CRDO) 88.0

17. Comfort Systems (FIX) 87.2

18. Dorian LPG (LPG) 86.9

19. SSR Mining (SSRM) 86.6

20. Tenet Healthcare (THC) 86.4

21. Assurant (AIZ) 84.3

22. Affiliated Managers Group (AMG) 84.0

23. West Pharmaceutical Services (WST) 83.9

24. The Allstate Corporation (ALL) 83.8

25. Red River Bancshares (RRBI) 83.7

26. Orange County Bancorp (OBT) 83.7

27. Popular, Inc. (BPOP) 83.4

28. KLA Corp (KLAC) 82.3

29. Host Hotels & Resorts (HST) 82.3

30. Qualys (QLYS) 82.0

31. Darling Ingredients (DAR) 81.9

32. Central Pacific Financial (CPF) 81.6

33. First Busey (BUSE) 81.3

34. SiriusXM (SIRI) 81.1

35. Capital City Bank Group (CCBG) 80.9

36. QCR Holdings (QCRH) 80.7

37. Nutex Health (NUTX) 80.4

38. Esquire Financial (ESQ) 80.3

39. Dave Inc. (DAVE) 79.9

40. Dropbox (DBX) 79.7

41. Clear Secure (YOU) 79.7

42. Eli Lilly (LLY) 79.0

43. Dime Community Bancshares (DCOM) 78.4

44. OneSpan (OSPN) 78.2

45. MetroCity Bankshares (MCBS) 78.0

46. Arista Networks (ANET) 77.7

47. Incyte (INCY) 77.3

48. Federated Hermes (FHI) 77.0

49. REX American Resources (REX) 76.7

50. Ennis (EBF) 76.6

51. Westamerica Bancorporation (WABC) 76.1

52. Horace Mann Educators (HMN) 75.8

53. Kohl’s (KSS) 74.8

54. Monarch Casino & Resort (MCRI) 74.5

55. HCI Group (HCI) 74.3

56. CarGurus (CARG) 74.1

57. PROG Holdings (PRG) 73.8

58. Newmont (NEM) 73.8

59. American States Water (AWR) 73.6

60. Monolithic Power (MPWR) 73.4

61. Capital Bancorp (CBNK) 72.6

62. ACADIA Pharmaceuticals (ACAD) 72.1

63. Regeneron Pharmaceuticals (REGN) 72.1

    1. followed by NVIDIA (NVDA) 72.0

64. Riley Exploration Permian (REPX) 71.9

    1. followed by Alphabet (GOOGL) 71.2

65. Cricut (CRCT) 68.6

66. Rigel Pharmaceuticals (RIGL) 68.3

67. Encompass Health (EHC) 68.1

68. Napco Security Technologies (NSSC) 68.1

69. EverQuote (EVER) 67.8

70. First Solar (FSLR) 67.6

71. Dexcom (DXCM) 67.5

72. InterDigital (IDCC) 67.4

73. CF Industries (CF) 65.5

74. Monster Beverage (MNST) 65.2

75. Fastenal (FAST) 65.1

    1. followed by Microsoft (MSFT) 63.8

76. The St. Joe Company (JOE) 63.5

77. NMI Holdings (NMIH) 63.4

78. Idaho Strategic Resources (IDR) 62.8

79. BK Technologies (BKTI) 62.8

80. Aurinia Pharmaceuticals (AUPH) 62.0

81. Powell Industries (POWL) 61.6

82. Harmony Biosciences (HRMY) 60.3

83. EMCOR Group (EME) 59.7

84. H&R Block (HRB) 59.6

85. Strategic Education (STRA) 59.4

86. EOG Resources (EOG) 59.2

87. United States Lime & Minerals (USLM) 58.6

88. iRadimed (IRMD) 58.5

89. Cincinnati Financial (CINF) 55.5

90. Innodata (INOD) 55.1

91. Royal Gold (RGLD) 55.1

92. Photronics (PLAB) 54.5

93. FactSet Research Systems (FDS) 53.5

94. Federal Signal Corp (FSS) 53.2

95. Hecla Mining (HL) 51.5

96. Pathward Financial (CASH) 51.0

97. SandRidge Energy (SD) 50.4

98. Urban Outfitters (URBN) 48.1

99. Perdoceo Education (PRDO) 47.3

100. Mastercard (MA) 44.2

     1. followed by Meta Platforms (META) 13.5

If you are rightfully looking for a few stocks which ranked very high in my original best 100 list by fundamentals, but didn’t show up in this list. It’s because I ranked a dozen more stocks by momentum and then decided to keep this list at 100 stocks, plus the four Mag7 stocks as extras. This way even the lowest ranked stocks show some decent momentum - as the title implies.

But if you are really curious which stocks where at the bottom, here as bonus are the lowest ranked stocks by momentum (with undervalued fundamentals):

• Pegasystems (PEGA) 11.9

• Netflix (NFLX) 7.4

• Intuit (INTU) 4.6

• Zoetis (ZTS) 1.3

• The Trade Desk (TTD) 0.6

📊Overview of the top 10 momentum stocks

Charts and company info

1. (MU) Micron

   Micron is a semiconductor company that designs, manufactures, and sells memory and storage products, including DRAM, NAND, and SSDs, to customers, increasingly to the growing sector of AI data centers.

   

2. (FTNT) Fortinet

   Fortinet is a cybersecurity company that sells network security hardware, software, and cloud-based solutions, primarily generating revenue from product sales and recurring subscriptions.

   

3. (ECPG) Encore Capital Group

   Encore Capital Group is a specialty finance company that purchases portfolios of defaulted consumer debt at deep discounts and generates returns by collecting on those receivables over time, while also providing debt-servicing and portfolio-management services.

   

4. (CBL) CBL & Associates Properties

   CBL & Associates Properties is a REIT that owns, operates, leases, and redevelops a portfolio of primarily enclosed malls and open-air retail properties, generating revenue mainly from tenant rents and related property income.

   

5. (NVEC) NVE Corp

   NVE Corp develops and sells spintronics-based semiconductor devices - primarily magnetic sensors, data couplers, and power products - for industrial, scientific, medical, and automation applications, while also monetizing its MRAM intellectual property through licensing.

   

6. (STLD) Steel Dynamics

   Steel Dynamics operates an integrated steel manufacturing and metals recycling business, producing and selling steel products, fabricated components, and aluminum while sourcing a significant portion of its raw materials through its recycling operations.

   

7. (CARE) Carter Bankshares

   Carter Bankshares is a regional bank holding company that provides commercial and consumer banking, lending, deposit, wealth-management, and related financial services primarily to individuals and businesses.

   

8. (DINO) HF Sinclair

   HF Sinclair is an independent energy company that refines crude oil into fuels and lubricants, distributes and markets refined petroleum products, and produces renewable fuels.

   

9. (TRV) Travelers

   Travelers is a property and casualty insurance company that generates revenue by underwriting and pricing risks across personal, commercial, and specialty insurance lines and investing the premiums it collects.

   

10. (VCTR) Victory Capital

    Victory Capital is a diversified global asset manager that earns management and advisory fees by investing and managing assets for institutional, intermediary, and individual clients across equity, fixed-income, alternative, and multi-asset strategies.

    

Subscribe now

This list is for informational purposes only.

The following companies distinguish themselves from their competitors and represent strong prospects for medium to long-term investment. I ranked the top stocks in the market according to their fundamental strength. The strongest opportunities are at the top of my best 100 list, but given that I analyzed the whole US market all stocks in this list are top stock picks outranking several thousands of other stocks.

My previous best list of the best stocks in the market was quite popular, so I decided to create a fresh update. For this update I both expanded the scope of stocks I analyzed and I also made a deeper fundamental analysis of the stocks.

As in my first version of this kind of list - I only analyzed the best of the best. I used this time even more filters to only analyze the very best stocks in the US market. Top notch only. Thousands of stocks looked at from every angle.

Lookback: I already created twenty-six similar versions of this kind of list calculating which stocks have, based on fundamental data, potentially higher odds of outperforming the market in the next 3-6 months.

Entering August I have created a new list (see below), ranking many stocks, which I haven’t analyzed before. The ranking was created with available data from the partial Q2 2026 earnings season. The creation of this ranking has again been mostly based on fundamental factors like quality and value and is only to some extent technical analysis (momentum) based.

This is a “best of” list, because I picked the top 10 stocks regarding to many different factors, here are some examples:

• best 10 quality stocks overall

• best 10 growth stocks overall

• best 10 value stocks overall

• best 10 stocks ranking the highest in each of their sub-factors

• relative best 10 stocks in each of their industry

And then I ended up with a list of the top 100 best of all US listed stocks. The stocks were selected as ranking the highest among roughly 6400 stocks.

Given that the value factor does play a stronger role in my overall analysis than the momentum factor, there are some stocks in my best list that aren’t in an uptrend right now. But based on their strong fundamental quality they might rebound in the next months.

Ranking methodology

As in the previous version of my best 100 list, thanks to my very deep analysis of all Dow Jones stocks I was able to improve my calculation method a lot. 2/3 of my calculation method is based on what I learned from my Dow Jones analysis. I only kept in my calculation what worked in correctly identifying the best and the worst stocks inside the Dow. Therefore the calculation method has been strongly improved. What stayed the same is that I’m using several different artificial intelligence models in the calculation of the overall score. The calculation has a slight tilt to the quality and the value factors. As explained earlier the momentum factor only plays a minor role in the overall calculation.

See the results of my previous rankings below:

Recent previous calculations

• previous list of the best 100 stocks as of May 2026

• list of all 30 Dow Jones stocks

• list of 32 top stocks

• list of 70 small-cap stocks

• list of 75 mid-cap stocks

• list of 55 small-cap stocks

• list of all S&P 400 stocks - 2nd half

• list of all S&P 400 stocks - 1st half

• list of all S&P 500 stocks - 2nd half

• list of all S&P 500 stocks - 1st half

Benchmark

As a benchmark I analyzed three popular top stocks:

Old ranking methodology with older data from May 29:

• NVIDIA (NVDA): 74.0

• Alphabet (GOOGL): 65.8

• Microsoft (MSFT): 69.0

New improved ranking methodology with data from July 31:

• NVIDIA (NVDA): 80.7

• Alphabet (GOOGL): 76.3

• Microsoft (MSFT): 74.7

NVIDIA continues to rank first place out of these three stocks, both in my old and also my new ranking methodologies benchmark. As said before fundamental data has by far the strongest weight in all calculations. The momentum factor and technical analysis only plays a small role in the ranking of the stocks. That’s why for example Meta Platforms (META) ranks ahead of several other Mag7 stocks.

One of the continued improvements of my calculation method is that it has gotten even harder to outrank one of the very best stocks in the market, I’m talking about NVIDIA. Based on the following ranking one can see that this time not a single stock has theoretically higher odds to outperform the market in the next 3-6 months than NVIDIA. In my almost finished version of this list a single stock was ranking higher than NVIDIA, but after increasing the weight on the quality factor only NVIDIA was left as the number one stock in my ranking.

All stocks in this list have a score higher than 60 out of 100, which theoretically means that all stocks in this list are fundamentally way above average attractive. In my previous best 100 list Meta Platforms was ranking number 26 in that list, but in this new version Meta Platforms jumped to fifth place. A little lower in the ranking one can also find Microsoft (highlighted in a similar way as NVIDIA). I excluded all Mag7 stocks that made it into my list from the ranking in order to highlight more lesser known companies.

Now here is my top 100 list based on data from the end of July 2026. The highest ranking stocks have the best mix of highest business quality and strongest business growth (revenue growth and earnings growth). They also rank high to very high in profitability, cash flow generation and liquidity (financial health), while their valuation is currently attractive (cheap relative to fundamentals).

Many other factors also play a role in the ranking like return on capital efficiency, balance sheet strength (low debt), reliability of earnings quality, as well as shareholder returns like offering a dividend. Dividends improve the ranking of a stock, growing the dividend over time even more, but several other factors have a stronger weight than the dividend factor. The overall focus of the ranking is on business quality and company valuation.

I excluded four stocks: NVDA, META, GOOGL, MSFT from the ranking, but listed their positions inside the top list. If included NVDA is ranked number one.

📈 Best 100 stocks

“Best of list” of ranked top stocks, sorted by overall fundamental score:

1. 🥇NVDA 80.7 ahead of 1. Deckers Outdoor (DECK) 79.6

2. Harmony Biosciences (HRMY) 78.4

3. Applovin (APP) 77.3

   1. followed by Meta Platforms (META) 76.6

   2. followed by Alphabet (GOOGL) 76.3

4. Adobe (ADBE) 75.7

5. Perdoceo Education (PRDO) 75.1

6. CF Industries (CF) 75.0

   1. followed by Microsoft (MSFT) 74.7

7. Doximity (DOCS) 74.7

8. Arista Networks (ANET) 74.6

9. Incyte (INCY) 74.3

10. Rigel Pharmaceuticals (RIGL) 74.2

11. Intuit (INTU) 74.2

12. NerdWallet (NRDS) 73.8

13. United States Lime & Minerals (USLM) 73.6

14. Innoviva (INVA) 73.4

15. Federated Hermes (FHI) 73.2

16. Gulfport Energy (GPOR) 73.2

17. Kinsale Capital Group (KNSL) 72.9

18. HCI Group (HCI) 72.9

19. Qualys (QLYS) 72.9

20. Newmont (NEM) 72.6

21. Stride (LRN) 71.9

22. Tradeweb Markets (TW) 71.9

23. ADMA Biologics (ADMA) 71.4

24. Dorian LPG (LPG) 71.4

25. Nutex Health (NUTX) 71.3

26. Ubiquiti (UI) 71.3

27. EverQuote (EVER) 71.2

28. AppFolio (APPF) 71.1

29. Micron (MU) 70.9

30. PTC Inc (PTC) 70.8

31. The Trade Desk (TTD) 70.8

32. CarGurus (CARG) 70.8

33. Intuitive Surgical (ISRG) 70.7

34. Jack Henry & Associates (JKHY) 70.7

35. Dave Inc. (DAVE) 70.7

36. EOG Resources (EOG) 70.7

37. Idaho Strategic Resources (IDR) 70.6

38. Federal Signal Corp (FSS) 70.3

39. Cincinnati Financial (CINF) 70.3

40. EMCOR Group (EME) 70.3

41. First Solar (FSLR) 70.1

42. NMI Holdings (NMIH) 70.1

43. Lam Research (LRCX) 70.0

44. Mastercard (MA) 69.9

45. Yelp Inc (YELP) 69.9

46. SandRidge Energy (SD) 69.9

47. Dexcom (DXCM) 69.8

48. Fastenal (FAST) 69.7

49. Eli Lilly (LLY) 69.7

50. Monster Beverage (MNST) 69.7

51. Aurinia Pharmaceuticals (AUPH) 69.6

52. Rambus (RMBS) 69.6

53. Netflix (NFLX) 69.4

54. Regeneron Pharmaceuticals (REGN) 69.3

55. Credo Technology Group (CRDO) 69.1

56. Victory Capital (VCTR) 69.1

57. Magnolia Oil & Gas (MGY) 69.0

58. Riley Exploration Permian (REPX) 68.7

59. Monarch Casino & Resort (MCRI) 68.6

60. iRadimed (IRMD) 68.4

61. Monolithic Power (MPWR) 68.4

62. Royal Gold (RGLD) 68.2

63. FactSet Research Systems (FDS) 68.2

64. Pegasystems (PEGA) 67.9

65. Cricut (CRCT) 67.9

66. The Buckle (BKE) 67.8

67. Upwork (UPWK) 67.7

68. Napco Security Technologies (NSSC) 67.7

69. KLA Corp (KLAC) 67.7

70. InterDigital (IDCC) 67.7

71. Comfort Systems (FIX) 67.4

72. Universal Health Services (UHS) 67.3

73. Photronics (PLAB) 67.3

74. APA Corp (APA) 67.0

75. Cal-Maine Foods (CALM) 67.0

76. Zoetis (ZTS) 66.8

77. The Allstate Corporation (ALL) 66.7

78. Universal Insurance (UVE) 66.7

79. Fortinet (FTNT) 66.7

80. H&R Block (HRB) 66.4

81. The St. Joe Company (JOE) 65.9

82. SSR Mining (SSRM) 65.8

83. Travelers (TRV) 65.8

84. Clear Secure (YOU) 65.8

85. Urban Outfitters (URBN) 65.7

86. Powell Industries (POWL) 64.9

87. HF Sinclair (DINO) 64.8

88. Strategic Education (STRA) 64.7

89. Innodata (INOD) 64.4

90. Encompass Health (EHC) 64.3

91. Ennis (EBF) 64.3

92. NVE Corp (NVEC) 64.3

93. Range Resources Corp (RRC) 64.1

94. Esquire Financial (ESQ) 64.0

95. Host Hotels & Resorts (HST) 63.8

96. Pathward Financial (CASH) 63.4

97. BK Technologies (BKTI) 62.9

98. Hecla Mining (HL) 62.6

99. Red River Bancshares (RRBI) 62.3

100. Liquidity Services (LQDT) 62.0

Ranking changes

With Rigel Pharmaceuticals one new stock, which wasn’t on my previous best 100 list, has barely made it into the top 10 of my top 100 at tenth place.

While ranking the highest in my previous best 100 list, Harmony Biosciences has kept a top position by ranking second this time.

Due to the price gains of Incyte (INCY went from $96.74 on May 29 to now $119.52, +23.55%) Incyte’s valuation has become less attractive. While ranking 4. in my previous list Incyte dropped in this new ranking to 9. place. If only technical analysis and the momentum factor would play instead the strongest role in the weighting then Incyte would have improved its ranking. That being said - remaining in the top 10 continues to make Incyte a top stock to watch.

On the other side of the spectrum these stocks jumped into the top 10:

• Adobe jumped from rank 13. in my previous list to now ranking 4.

• Perdoceo Education jumped from 16. to 5. place

• CF Industries jumped from 15. to 6. place

• Arista Networks jumped from 19. to 8. place

• AppLovin jumped from 43. to 3. place

The biggest jump was made by Doximity which jumped from 70. to 7. place.

📊Overview of the top 10 stocks

Charts and discounted cash flow (DCF) analysis

1. (DECK) Deckers Outdoor

   Deckers Outdoor designs, markets, and sells premium footwear, apparel, and accessories under brands including HOKA and UGG, generating revenue through a mix of direct-to-consumer channels and wholesale distribution to retailers worldwide.

   

   

2. (HRMY) Harmony Biosciences

   Harmony Biosciences is a commercial-stage biopharmaceutical company that develops, acquires, and markets therapies for rare neurological disorders, generating revenue primarily from sales of its flagship narcolepsy treatment while expanding its pipeline through research and strategic licensing/acquisitions.

   

   

3. (APP) Applovin

   AppLovin Corporation operates a mobile advertising technology platform that helps app developers acquire users, monetize applications, and optimize ad performance through AI-powered advertising, analytics, and app growth solutions.

   

   

4. (ADBE) Adobe

   Adobe operates a subscription-based software business model, providing creative, document, and digital experience solutions to individuals, businesses, and enterprises through recurring cloud-based services.

   

   

5. (PRDO) Perdoceo Education

   Perdoceo Education operates a career-focused higher education platform that provides online and campus-based degree programs through its institutions, generating revenue primarily from student tuition and related education services.

   

   

6. (CF) CF Industries

   CF Industries manufactures and distributes nitrogen-based fertilizer products and related chemicals that support global agriculture and industrial markets, generating revenue through the sale of ammonia, urea, and other nitrogen solutions.

   

   

7. (DOCS) Doximity

   Doximity operates a digital healthcare platform that connects medical professionals, enables clinical communications and telehealth services, and generates revenue primarily through physician recruitment solutions and pharmaceutical marketing.

   

   

8. (ANET) Arista Networks

   Arista Networks designs and sells high-performance cloud networking solutions, including Ethernet switches, routing platforms, and software that enable large-scale data centers, cloud providers, and enterprises to build fast, automated, and secure networks.

   

   

9. (INCY) Incyte

   Incyte is a biopharmaceutical company that discovers, develops, and commercializes innovative medicines, primarily focused on oncology, inflammation, and autoimmune diseases through targeted therapies.

   

   

10. (RIGL) Rigel Pharmaceuticals

    Rigel Pharmaceuticals is a biotechnology company that discovers, develops, and commercializes targeted therapies for autoimmune diseases, hematologic disorders, and cancer, with a focus on small-molecule medicines that modulate immune and cellular signaling pathways.

    

    

Subscribe now

This list is for informational purposes only.

This time around I only analyzed the best of the best. I used the most filters I have used so far to only analyze the very best stocks in the US market. Top notch only. Thousands of stocks looked at from every angle.

Lookback: I already created twenty-five similar versions of this kind of list calculating which stocks have, based on fundamental data, potentially higher odds of outperforming the market in the next 3-6 months.

Now at the end of May I have created a new list (see below), ranking many stocks, which I haven’t analyzed before. There are also several foreign stocks in this list.

The ranking was created with available data from the partial Q1 2026 earnings season. The creation of this ranking has again been mostly based on fundamental factors like quality and value and is only partially technical analysis (momentum) based.

This is a “best of” list, because I picked the top 10 stocks regarding to many different factors, here are some example:

• best 10 quality stocks overall

• best 10 growth stocks overall

• best 10 value stocks overall

• best 10 stocks ranking the highest in each of their sub-factors

• relative best 10 stocks in each of their industry

And then I ended up with a list of the top 100 best of all US listed stocks. The stocks were selected as ranking the highest among roughly 6400 stocks.

Ranking methodology

Thanks to my very deep analysis of all Dow Jones stocks I was able to improve my calculation method a lot. 2/3 of my calculation method is new based on what I learned from my Dow Jones analysis. I only kept in my calculation what worked in correctly identifying the best and the worst stocks inside the Dow. Therefore the calculation method is a new strongly improved version. What stayed the same is that I’m using several different artificial intelligence models in the calculation of the overall score. The calculation has a slight value tilt.

See the results of my previous rankings below:

Recent previous calculations

• list of all 30 Dow Jones stocks

• list of 32 top stocks

• list of 70 small-cap stocks

• list of 75 mid-cap stocks

• list of 55 small-cap stocks

• list of all S&P 400 stocks - 2nd half

• list of all S&P 400 stocks - 1st half

• list of all S&P 500 stocks - 2nd half

• list of all S&P 500 stocks - 1st half

Benchmark

As a benchmark I analyzed three popular top stocks:

Old ranking methodology with older data from May 22:

• NVIDIA (NVDA): 74.0

• Alphabet (GOOGL): 65.8

• Microsoft (MSFT): 69.0

New improved ranking methodology with data from May 29:

• NVIDIA (NVDA): 72.9

• Alphabet (GOOGL): 64.0

• Microsoft (MSFT): 62.3

NVIDIA continues to rank first place out of these three stocks, both in my old and also my new ranking methodologies benchmark. As said before fundamental data has by far the strongest weight in all calculations. The momentum factor and technical analysis only plays a minor role in the ranking of the stocks.

One of the improvements of my calculation method is that it has gotten much harder to outrank one of the best stocks in the market, I’m talking about Nvidia. Based on the following ranking one can see that only one stock - Harmony Biosciences - has theoretically slightly higher odds to outperform the market in the next 3-6 months than NVIDIA, Microsoft or Alphabet.

All stocks in this list have a score higher than 50, which theoretically means that all stocks are fundamentally somewhat attractive. Meta Platforms is ranking number 26 in this list. Lower in the ranking you will also find Alphabet and Microsoft (both are highlighted in a similar way as Nvidia, which ranks second if included in the ranking).

Now here is my top 100 list based on data from the end of May 2026:

“Best of list” of ranked top stocks, sorted by overall score

1. Harmony Biosciences (HRMY) 73.9 (followed by NVDA 72.9)

2. Intuit (INTU) 72.2

3. ResMed (RMD) 69.7

4. Incyte (INCY) 69.1

5. PDD Holdings (PDD) 68.9

6. NetEase (NTES) 68.5

7. Sezzle (SEZL) 68.1

8. Hamilton Lane (HLNE) 67.9

9. Lantheus Holdings (LNTH) 67.6

10. Exelixis (EXEL) 67.3

11. dLocal (DLO) 67.3

12. Cal-Maine Foods (CALM) 67.3

13. Adobe (ADBE) 67.3

14. Atour Lifestyle Holdings (ATAT) 67.2

15. CF Industries (CF) 67.0

16. Perdoceo Education (PRDO) 66.9

17. Magnolia Oil & Gas (MGY) 66.8

18. Interdigital (IDCC) 66.8

19. Arista Networks (ANET) 66.7

20. Qfin Holdings (QFIN) 66.5

21. Evercore (EVR) 66.4

22. Newmont (NEM) 66.3

23. Kanzhun (BZ) 66.2

24. Blue Bird (BLBD) 66.1

25. Tradeweb Markets (TW) 65.9

26. Meta Platforms (META) 65.9

27. Paycom Software (PAYC) 65.8

28. NMI Holdings (NMIH) 65.8

29. SEI Investments Company (SEIC) 65.7

30. Duolingo (DUOL) 65.7

31. Qualys (QLYS) 65.5

32. Ubiquiti (UI) 65.4

33. Monster Beverage (MNST) 65.2

34. Kaspi kz (KSPI) 65.1

35. MarketAxess (MKTX) 65.1

36. Futu Holdings (FUTU) 65.0

37. Yalla Group (YALA) 64.9

38. Deckers Outdoor (DECK) 64.9

39. Aurinia Pharmaceuticals (AUPH) 64.8

40. Victory Capital Holdings (VCTR) 64.6

41. HCI Group (HCI) 64.3 (followed by GOOGL 64.0)

42. Paylocity (PCTY) 63.8

43. Applovin (APP) 63.8

44. Texas Pacific Land (TPL) 63.7

45. Exzeo (XZO) 63.7

46. Grupo Aeroportuario del Centro Norte (OMAB) 63.4

47. Stride (LRN) 63.4

48. EOG Resources (EOG) 63.1

49. Monolithic Power (MPWR) 63.1

50. Garmin (GRMN) 63.1

51. Visa (V) 63.0

52. Essent (ESNT) 63.0

53. RenaissanceRe (RNR) 62.7

54. Kinsale Capital Group (KNSL) 62.6

55. Mueller Industries (MLI) 62.5

56. Southern Copper (SCCO) 62.5

57. Royal Gold (RGLD) 62.5

58. Mastercard (MA) 62.5

59. Comfort Systems (FIX) 62.5

60. Anglogold Ashanti (AU) 62.4 (followed by MSFT 62.3)

61. Taiwan Semiconductor Manufacturing (TSM) 62.2

62. Halozyme Therapeutics (HALO) 61.7

63. Lam Research (LRCX) 61.6

64. Dexcom (DXCM) 61.6

65. Netflix (NFLX) 61.6

66. Frontline (FRO) 61.6

67. CarGurus (CARG) 61.4

68. Range Resources Corp (RRC) 61.4

69. Palantir (PLTR) 61.2

70. Doximity (DOCS) 61.2

71. Chagee Holdings (CHA) 61.2

72. Innoviva (INVA) 61.2

73. SSR Mining (SSRM) 61.1

74. ASML Holding (ASML) 60.8

75. Eli Lilly (LLY) 60.8

76. Clear Secure (YOU) 60.8

77. Amphenol (APH) 60.6

78. Matador Resources (MTDR) 60.5

79. Micron (MU) 60.4

80. Broadcom (AVGO) 60.3

81. Yelp (YELP) 60.1

82. Boyd Gaming (BYD) 59.7

83. Palomar (PLMR) 59.6

84. Petroleo Brasileiro SA Petrobras (PBR) 58.5

85. Nova (NVMI) 58.5

86. Comcast (CMCSA) 58.3

87. Dropbox (DBX) 58.1

88. First Solar (FSLR) 57.8

89. Genmab (GMAB) 57.6

90. Universal Health Services (UHS) 57.5

91. Credo Technology (CRDO) 56.7

92. Hecla Mining (HL) 56.5

93. Howmet Aerospace (HWM) 56.5

94. Intercorp Financial Services (IFS) 56.2

95. Encore Capital Group (ECPG) 56.2

96. Emcor (EME) 56.1

97. Globe Life (GL) 56.0

98. TAL Education (TAL) 55.8

99. Bread Financial Holding (BFH) 55.7

100. Antero Resources (AR) 55.2

Subscribe now

Overall score - chart overview & discounted cash flow analysis (as of May 29, 2026)

Harmony Biosciences (HRMY)

Intuit (INTU)

ResMed (RMD)

Incyte (INCY)

PDD Holdings (PDD)

NetEase (NTES)

Sezzle (SEZL)

Hamilton Lane (HLNE)

Lantheus Holdings (LNTH)

Exelixis (EXEL)

Subscribe now

This list is for informational purposes only.

Lookback: I already created twenty-four similar versions of this kind of list calculating which stocks have, based on fundamental data, potentially higher odds of outperforming the market in the next 3-6 months.

Now near the end of May I have created a new list (see below), ranking all stocks in the Dow Jones.

The ranking was created with available data from the partial Q1 2026 earnings season. The creation of this ranking has again been mostly based on fundamental factors like quality, growth and value and is only partially technical analysis (momentum) based.

Ranking methodology

The calculation method is the most complex version I have used so far. It’s roughly 15x more complex than my usual calculation method. I can only apply this to mega-cap stocks like those in the Dow Jones, because it requires much more data - which usually isn’t available for e.g. mid-cap and small-cap stocks.

While my previous version already used artificial intelligence, this version uses several very different AIs in the calculation of the score. Here are just a few of the sub-factors that are part of the overall calculation of the score:

1. Insider sentiment (net buying vs selling activity of company insiders)

2. News sentiment (positive vs negative tone of financial articles)

3. Short interest (serves as a strong gauge of market sentiment)

4. Low risk profile (risk-adjusted efficiency, sharpe ratio vs peer averages)

5. R&D (research & development spending as percentage of revenue)

6. Relative growth (compared to the industry and competitors)

7. Future growth investment (capital expenditure intensity vs industry)

8. Financial safety (measuring the risk of financial distress)

9. Earnings quality (are profits driven by cash, not accounting tricks)

10. Dividend (stability, growth and safety of dividend payments)

11. Management (stock-based compensation and ownership dilution)

12. Institutional ownership (high institutional ownership signals trust)

13. Working capital efficiency (analyzes cash conversion cycle)

14. Business advantage (analyzing cost control and economy of scale)

15. Business predictability (predictability of the earnings growth)

16. Analysts (near-term outlook based on Wall Street consensus)

Recent previous calculations

• list of 46 top stocks

• list of 32 top stocks

• list of 70 small-cap stocks

• list of 75 mid-cap stocks

• list of 55 small-cap stocks

• list of all S&P 400 stocks - 2nd half

• list of all S&P 400 stocks - 1st half

• list of all S&P 500 stocks - 2nd half

• list of all S&P 500 stocks - 1st half

Results

Nvidia ranks the highest, Microsoft ranks number two. And Apple beats Amazon to fourth place. One might have expected to find those stocks in the top 5 and it turned out to be this way. As explained before this ranking is mostly based on fundamental factors, that’s why Microsoft is able to rank ahead of Apple and Amazon, despite currently having a weaker price trend.

I find it interesting how high Salesforce ranks in the final ranking - sixth place - right behind Amazon. And Travelers managed to rank even higher than Salesforce, right behind Microsoft as number three.

List of ranked Dow Jones stocks, sorted by overall score

• Nvidia (NVDA) 77.27

• Microsoft (MSFT) 72.94

• Travelers (TRV) 72.25

• Apple (AAPL) 70.71

• Amazon (AMZN) 69.89

• Salesforce (CRM) 69.13

• Visa (V) 68.81

• Coca-Cola (KO) 66.31

• Cisco (CSCO) 65.53

• Johnson & Johnson (JNJ) 64.55

• Caterpillar (CAT) 64.47

• Disney (DIS) 63.20

• Amgen (AMGN) 62.98

• IBM (IBM) 62.90

• Verizon (VZ) 62.42

• Procter & Gamble (PG) 62.39

• Merck (MRK) 62.03

• American Express (AXP) 61.40

• Walmart (WMT) 61.01

• JPMorgan Chase (JPM) 59.12

• McDonald’s (MCD) 58.88

• Sherwin-Williams (SHW) 58.55

• UnitedHealth (UNH) 57.49

• Chevron (CVX) 57.12

• Honeywell (HON) 56.12

• Home Depot (HD) 55.30

• Goldman Sachs (GS) 55.12

• Nike (NKE) 52.64

• 3M (MMM) 52.07

• Boeing (BA) 44.09

Subscribe now

Overall score - chart overview & discounted cash flow analysis (as of May 22, 2026)

Nvidia (NVDA)

Microsoft (MSFT)

Travelers (TRV)

Apple (AAPL)

Amazon (AMZN)

Salesforce (CRM)

Visa (V)

Coca-Cola (KO)

Cisco (CSCO)

Johnson & Johnson (JNJ)

🛠️Update (May 26, 2026)

I ranked all stocks on Tuesday before the close* with additional factors and more data, which resulted in a slightly different overall ranking. Now Travelers ranks higher in 2nd place and Visa ranks much higher than before in 4th place.

• Nvidia (NVDA) 77.34

• Travelers (TRV) 70.96

• Microsoft (MSFT) 70.71

• Visa (V) 68.76

• Salesforce (CRM) 68.70

• Apple (AAPL) 68.09

• Amazon (AMZN) 67.71

• Cisco (CSCO) 64.78

• Coca-Cola (KO) 64.33

• Johnson & Johnson (JNJ) 64.08

• Caterpillar (CAT) 63.38

• IBM (IBM) 62.64

• Amgen (AMGN) 62.58

• Disney (DIS) 61.88

• Verizon (VZ) 61.45

• Merck (MRK) 61.15

• American Express (AXP) 60.88

• Procter & Gamble (PG) 60.52

• Walmart (WMT) 60.37

• JPMorgan Chase (JPM) 58.57

• Sherwin-Williams (SHW) 57.64

• McDonald’s (MCD) 57.46

• Chevron (CVX) 56.98

• UnitedHealth (UNH) 56.03

• Goldman Sachs (GS) 55.22

• Honeywell (HON) 55.01

• Home Depot (HD) 53.40

• 3M (MMM) 51.42

• Nike (NKE) 50.96

• Boeing (BA) 45.27

*Monday was a holiday and the stock market was closed. Therefore, little had changed regarding the data until the Tuesday close, as most of the data analysis is based on fundamental data from the May 22 close.

🆕Final ranking update (June 1, 2026)

I added much more data and a few additional factors to create the last and final update of the ranking. Now Microsoft ranks higher in 2nd place, dropping Travelers to a still very good, but 3rd place. And Apple now outranks Salesforce by ranking 5th place. Plus Johnson & Johnson is now ahead of Coca-Cola at 9th place in the ultimate and final update of the ranking of all Dow Jones stocks.

1. Nvidia (NVDA) 81.56

2. Microsoft (MSFT) 72.09

3. Travelers (TRV) 69.89

4. Visa (V) 69.11

5. Apple (AAPL) 68.73

6. Salesforce (CRM) 66.72

7. Amazon (AMZN) 66.18

8. Cisco (CSCO) 65.83

9. Johnson & Johnson (JNJ) 65.06

10. Coca-Cola (KO) 64.95

11. Caterpillar (CAT) 63.93

12. American Express (AXP) 63.28

13. Amgen (AMGN) 62.88

14. Disney (DIS) 62.49

15. JPMorgan Chase (JPM) 61.77

16. IBM (IBM) 61.12

17. Procter & Gamble (PG) 61.09

18. Merck (MRK) 60.60

19. Verizon (VZ) 59.35

20. Walmart (WMT) 59.11

21. Goldman Sachs (GS) 56.80

22. McDonald’s (MCD) 55.83

23. Sherwin-Williams (SHW) 55.34

24. Chevron (CVX) 55.10

25. UnitedHealth (UNH) 55.04

26. Honeywell (HON) 54.07

27. Home Depot (HD) 53.77

28. 3M (MMM) 52.76

29. Nike (NKE) 51.24

30. Boeing (BA) 43.87

Subscribe now

This list is for informational purposes only.

Lookback: I already created twenty-three similar versions of this kind of list calculating which stocks have, based on fundamental data, potentially higher odds of outperforming the market in the next 3-6 months.

Now near the end of May I have created a new list (see below), ranking several stocks, which I haven’t analyzed before. There are also a few foreign stocks in this list.

The ranking was created with available data from the partial Q1 2026 earnings season. The creation of this ranking has again been mostly based on fundamental factors like quality and value and is only partially technical analysis (momentum) based.

I tried to find companies in this top selection which are ranking the highest mostly among these five factors:

• quality (e.g. profitability, consistency, capital efficiency)

• valuation (e.g. P/E, EV/EBITDA, P/FCF)

• momentum (technical analysis of the trend)

• risk (annual volatility, maximum drawdown)

• sentiment (news flow, AI analysis of the tone of headlines)

The selected top 46 stocks ranked the highest among many financial metrics. The stocks were selected as ranking the highest among roughly 1500 stocks.

Ranking methodology

The calculation method is again a new version. While my previous version already used a complex artificial intelligence, this new version uses a different artificial intelligence in the calculation of the score. I mostly changed one factor. This AI version doesn’t measure the moat, but it measures the risk profile and has a stronger weight on sentiment. Here are all the five sub-factors that are part of the overall calculation of the score:

1. Insider sentiment (net buying vs selling activity of company insiders)

2. News sentiment (positive vs negative tone of financial articles)

3. Low risk profile (risk-adjusted efficiency, sharpe ratio vs peer averages)

4. R&D (research & development spending as percentage of revenue)

5. Financial safety (measuring the risk of financial distress)

In addition to my overall calculation of the score, I decided to also rank all 46 stocks purely as measured by artificial intelligence. See rankings below:

Recent previous calculations

• list of 32 top stocks

• list of 70 small-cap stocks

• list of 75 mid-cap stocks

• list of 55 small-cap stocks

• list of all S&P 400 stocks - 2nd half

• list of all S&P 400 stocks - 1st half

• list of all S&P 500 stocks - 2nd half

• list of all S&P 500 stocks - 1st half

Benchmark

As a benchmark I analyzed three popular top stocks:

Old ranking methodology with older data from May 15:

• NVIDIA (NVDA): 72.2

• Alphabet (GOOGL): 68.4

• Microsoft (MSFT): 66.2

New overall ranking methodology with data from May 22:

• NVIDIA (NVDA): 74.0

• Alphabet (GOOGL): 65.8

• Microsoft (MSFT): 69.0

New pure AI ranking methodology with data from May 22:

• NVIDIA (NVDA): 67.0

• Alphabet (GOOGL): 63.5

• Microsoft (MSFT): 60.1

NVIDIA continues to rank first place out of these three stocks, both in my old and also my new ranking methodologies. As said before fundamental data has the strongest weight in all calculations. I have also again created a second overall score to rank stocks higher when the main score is equal.

Using the new pure AI score interestingly Lantheus Holdings, Zoom Communications, SEI Investments, Bank OZK and Paycom jump higher to rank in the top 10 when measured purely by artificial intelligence.

More importantly, based on the following ranking one can see that one stock - Incyte - has theoretically slightly higher odds to outperform the market in the next 3-6 months than NVIDIA, Microsoft or Alphabet. Let’s see about that.

Now here are the two differently sorted lists based on data from May 2026, plus an additional third list comparing both lists:

List of ranked top stocks, sorted by overall score

• Incyte (INCY) 77.9

• Duolingo (DUOL) 72.4

• DexCom (DXCM) 71.0

• Cincinnati Financial (CINF) 68.4

• Kanzhun (BZ) 67.9

• Fortinet (FTNT) 67.4

• Qualys (QLYS) 67.0

• Hartford Insurance Group (HIG) 66.2

• Allstate (ALL) 66.0

• First Solar (FSLR) 65.9

• First Bancorp (FBP) 65.5

• Adobe (ADBE) 65.3

• Deckers Outdoor (DECK) 65.0

• Zoom Communications (ZM) 64.7

• SEI Investments (SEIC) 64.5

• Paycom (PAYC) 64.3

• Lantheus Holdings (LNTH) 64.2

• Arch Capital Group (ACGL) 63.7

• F5 Inc (FFIV) 62.8

• Novo Nordisk (NVO) 62.5

• Venture Global (VG) 62.4

• Automatic Data Processing (ADP) 61.9

• Omega Healthcare Investors (OHI) 61.7

• Ericsson (ERIC) 61.3

• Expeditors International Of Washington (EXPD) 59.8

• Paychex (PAYX) 59.6

• Bank OZK (OZK) 59.5

• Netapp (NTAP) 59.1

• Coca-Cola (KO) 59.0

• Cal-Maine Foods (CALM) 58.1

• Host Hotels & Resorts (HST) 57.0

• Terreno Realty (TRNO) 56.9

• Acadia Pharmaceuticals (ACAD) 56.4

• Progressive (PGR) 56.3

• EOG Resources (EOG) 56.1

• Delta Air Lines (DAL) 55.9

• Illumina (ILMN) 55.1

• eBay (EBAY) 54.6

• Solventum (SOLV) 54.2

• BHP Group (BHP) 53.4

• Merck (MRK) 53.1

• Spire Global (SPIR) 50.1

• Diamondrock Hospitality (DRH) 48.8

• Aegon (AEG) 48.6

• Karyopharm Therapeutics (KPTI) 44.1

• PTC Therapeutics (PTCT) 43.7

List of ranked top stocks, sorted by pure AI score

• Incyte (INCY) 65.9

• Lantheus Holdings (LNTH) 63.8

• Hartford Insurance Group (HIG) 62.4

• Fortinet (FTNT) 62.3

• Zoom Communications (ZM) 62.3

• Duolingo (DUOL) 61.6

• SEI Investments (SEIC) 60.8

• Bank OZK (OZK) 60.7

• Paycom (PAYC) 60.3

• First Bancorp (FBP) 60.1

• Cincinnati Financial (CINF) 59.9

• Adobe (ADBE) 59.8

• Cal-Maine Foods (CALM) 59.8

• Allstate (ALL) 59.7

• PTC Therapeutics (PTCT) 59.6

• Automatic Data Processing (ADP) 59.1

• Novo Nordisk (NVO) 58.8

• Netapp (NTAP) 58.6

• Kanzhun (BZ) 58.5

• Qualys (QLYS) 58.2

• Paychex (PAYX) 58.1

• BHP Group (BHP) 57.6

• Expeditors International Of Washington (EXPD) 57.5

• Host Hotels & Resorts (HST) 57.4

• Arch Capital Group (ACGL) 57.1

• F5 Inc (FFIV) 57.1

• EOG Resources (EOG) 56.8

• Merck (MRK) 55.9

• DexCom (DXCM) 55.9

• eBay (EBAY) 55.2

• Ericsson (ERIC) 54.5

• Terreno Realty (TRNO) 54.2

• Deckers Outdoor (DECK) 54.2

• Omega Healthcare Investors (OHI) 54.0

• Progressive (PGR) 53.2

• First Solar (FSLR) 53.1

• Coca-Cola (KO) 53.0

• Acadia Pharmaceuticals (ACAD) 52.9

• Illumina (ILMN) 52.1

• Karyopharm Therapeutics (KPTI) 51.9

• Venture Global (VG) 51.8

• Spire Global (SPIR) 51.7

• Solventum (SOLV) 51.6

• Aegon (AEG) 50.9

• Diamondrock Hospitality (DRH) 50.8

• Delta Air Lines (DAL) 48.7

List of top stocks which are among the top 10 in both lists

• Incyte (INCY) 77.9 (AI 65.9), AVG: 71.9

• Duolingo (DUOL) 72.4 (AI 61.6), AVG: 67.0

• Fortinet (FTNT) 67.4 (AI 62.3), AVG: 64.9

• Hartford Insurance Group (HIG) 66.2 (AI 62.4), AVG: 64.3

Subscribe now

Overall score - chart overview & discounted cash flow analysis (as of May 22, 2026)

Incyte (INCY)

Duolingo (DUOL)

DexCom (DXCM)

Cincinnati Financial (CINF)

Kanzhun (BZ)

Fortinet (FTNT)

Qualys (QLYS)

Hartford Insurance Group (HIG)

Allstate (ALL)

First Solar (FSLR)

Six stocks that are in the top 10 of the pure AI score, but not in the top 10 of the overall score - chart overview & discounted cash flow analysis (as of May 22, 2026)

Lantheus Holdings (LNTH)

Zoom Communications (ZM)

SEI Investments (SEIC)

Bank OZK (OZK)

Paycom (PAYC)

First Bancorp (FBP)

Bonus: Non-AI Ranking

Above I additionally published a ranking of the 46 stocks purely using fundamental analysis based on artificial intelligence. Here below is now the opposite, a ranking of the 46 top stocks only using a rules-based fundamental mathematical scoring system (which is also part of the overall score).

• Incyte (INCY) 90.0

• DexCom (DXCM) 86.1

• Duolingo (DUOL) 83.2

• First Solar (FSLR) 78.8

• Kanzhun (BZ) 77.3

• Cincinnati Financial (CINF) 77.0

• Deckers Outdoor (DECK) 75.9

• Qualys (QLYS) 75.9

• Venture Global (VG) 73.1

• Fortinet (FTNT) 72.6

• Allstate (ALL) 72.3

• Adobe (ADBE) 70.8

• First Bancorp (FBP) 70.8

• Arch Capital Group (ACGL) 70.3

• Hartford Insurance Group (HIG) 70.1

• Omega Healthcare Investors (OHI) 69.3

• F5 Inc (FFIV) 68.6

• Paycom (PAYC) 68.4

• SEI Investments (SEIC) 68.1

• Ericsson (ERIC) 68.1

• Zoom Communications (ZM) 67.2

• Novo Nordisk (NVO) 66.1

• Coca-Cola (KO) 65.1

• Automatic Data Processing (ADP) 64.8

• Lantheus Holdings (LNTH) 64.7

• Delta Air Lines (DAL) 63.2

• Expeditors International Of Washington (EXPD) 62.1

• Paychex (PAYX) 61.1

• Acadia Pharmaceuticals (ACAD) 59.8

• Terreno Realty (TRNO) 59.6

• Netapp (NTAP) 59.5

• Progressive (PGR) 59.3

• Bank OZK (OZK) 58.4

• Illumina (ILMN) 58.1

• Solventum (SOLV) 56.8

• Host Hotels & Resorts (HST) 56.5

• Cal-Maine Foods (CALM) 56.5

• EOG Resources (EOG) 55.4

• eBay (EBAY) 54.0

• Merck (MRK) 50.2

• BHP Group (BHP) 49.1

• Spire Global (SPIR) 48.5

• Diamondrock Hospitality (DRH) 46.9

• Aegon (AEG) 46.4

• Karyopharm Therapeutics (KPTI) 36.4

• PTC Therapeutics (PTCT) 27.8

In this bonus top 10 there are two stocks which weren’t in the other two top 10 lists. Below are the details for those two:

Bonus - Chart overview & discounted cash flow analysis (as of May 22, 2026)

Deckers Outdoor (DECK)

Venture Global (VG)

no fully analyzed chart available

Subscribe now

This list is for informational purposes only.

Lookback: I already created twenty-two similar versions of this kind of list calculating which stocks have, based on fundamental data, potentially higher odds of outperforming the market in the next 3-6 months.

Now at the middle of May I have created a new list (see below), ranking several stocks, which I haven’t analyzed before. There are also a few foreign stocks in this list.

The ranking was created with available data from the partial Q1 2026 earnings season. The creation of this ranking has again been mostly based on fundamental factors like quality and growth as well as value and is only partially technical analysis (momentum) based. As explained further below this ranking includes a few more new factors like strength of the economic moat.

I tried to find companies in this top selection which are ranking the highest mostly among these four factors:

• quality

• growth

• valuation

• moat

The selected top 32 stocks ranked the highest among many financial metrics.

Ranking methodology

The calculation method is a new strongly improved version. While my previous version partially used machine learning, the new calculation method uses to a stronger degree far more complex artificial intelligence in the calculation of the score. With the help of AI this new method adds five more factors into the overall calculation of the score:

1. Insider sentiment (net buying vs selling activity of company insiders)

2. News sentiment (positive vs negative tone of financial articles)

3. Economic moat (competitive advantages like e.g. patents)

4. R&D (research & development spending as percentage of revenue)

5. Financial safety (measuring the risk of financial distress)

In addition to my overall calculation of the score, I decided to also rank all 32 stocks purely as measured by artificial intelligence. See rankings below:

Recent previous calculations

• list of 70 small-cap stocks

• list of 75 mid-cap stocks

• list of 55 small-cap stocks

• list of all S&P 400 stocks - 2nd half

• list of all S&P 400 stocks - 1st half

• list of all S&P 500 stocks - 2nd half

• list of all S&P 500 stocks - 1st half

Benchmark

As a benchmark I analyzed three popular top stocks:

Old ranking methodology with older data from May 8:

• NVIDIA (NVDA): 71.9

• Alphabet (GOOGL): 68.3

• Microsoft (MSFT): 67.8

New overall ranking methodology with data from May 15:

• NVIDIA (NVDA): 72.2

• Alphabet (GOOGL): 68.4

• Microsoft (MSFT): 66.2

New pure AI ranking methodology with data from May 15:

• NVIDIA (NVDA): 70.1

• Alphabet (GOOGL): 62.5

• Microsoft (MSFT): 60.6

NVIDIA continues to rank first place out of these three stocks, both in my old and also my new ranking methodologies. Using only artificial intelligence the distance in the score is even larger between NVIDIA and Microsoft. As said before fundamental data has the strongest weight in all calculations. I have also again created a second overall score to rank stocks higher when the main score is equal.

Using the new pure AI score interestingly Argenx and Adobe jump from ranking near the bottom in my overall score to ranking in the top 10 as measured purely by artificial intelligence (using the fundamental factors I mentioned above).

More importantly, based on the following ranking one can see that eleven stocks - KGC, EVR, SLDE, NEM, INCY, DLO, ATAT, AU, HALO, MAKO and SEZL - have theoretically slightly higher odds to outperform the market in the next 3-6 months than NVIDIA, Microsoft or Alphabet. Let’s see about that.

Now here are the two differently sorted lists based on data from May 2026, plus an additional third list comparing both lists:

List of ranked top stocks, sorted by overall score

• Evercore (EVR) 77.0*

• Kinross Gold (KGC) 77.0

• Slide Insurance (SLDE) 76.7

• Newmont (NEM) 75.8

• Incyte (INCY) 75.4

• dLocal (DLO) 75.2

• Atour Lifestyle (ATAT) 75.0

• AngloGold Ashanti (AU) 73.6

• Halozyme Therapeutics (HALO) 73.6

• Mako Mining (MAKO) 73.5

• Sezzle (SEZL) 72.5

• Taiwan Semiconductor Manufacturing (TSM) 71.3

• HCI Group (HCI) 70.9

• Barrick Mining (B) 70.7

• Palomar (PLMR) 69.3

• First Solar (FSLR) 68.6

• Mueller Industries (MLI) 68.5

• Kanzhun (BZ) 68.4

• Tradeweb Markets (TW) 67.7

• Futu Holdings (FUTU) 67.6

• Blue Bird (BLBD) 67.5

• Adobe (ADBE) 67.4

• Deckers Outdoor (DECK) 67.4

• Hamilton Lane (HLNE) 67.2

• ResMed (RMD) 66.6

• Argenx (ARGX) 64.5

• Grupo Aeroportuario Del Pacifico (PAC) 64.1

• MercadoLibre (MELI) 63.2

• RenaissanceRe (RNR) 62.8

• Visa (V) 62.1

• PDD (PDD) 61.2

• MarketAxess (MKTX) 59.2

*Evercore (EVR) has both a higher 2nd score and also a higher pure AI score than Kinross Gold (KGC).

List of ranked top stocks, sorted by pure AI score

• Halozyme Therapeutics (HALO) 70.0

• Mako Mining (MAKO) 69.6

• Evercore (EVR) 68.0

• Slide Insurance (SLDE) 66.7

• Atour Lifestyle (ATAT) 66.3

• Kinross Gold (KGC) 65.6

• Argenx (ARGX) 65.3

• Adobe (ADBE) 64.5

• dLocal (DLO) 64.3

• Kanzhun (BZ) 63.9

• Newmont (NEM) 63.8

• ResMed (RMD) 63.7

• AngloGold Ashanti (AU) 63.6

• Tradeweb Markets (TW) 63.4

• Deckers Outdoor (DECK) 63.3

• Incyte (INCY) 63.1

• Grupo Aeroportuario Del Pacifico (PAC) 63.0

• Futu Holdings (FUTU) 62.1

• Taiwan Semiconductor Manufacturing (TSM) 62.1

• Sezzle (SEZL) 61.4

• PDD (PDD) 61.3

• Mueller Industries (MLI) 61.2

• MarketAxess (MKTX) 61.1

• HCI Group (HCI) 60.9

• Visa (V) 59.8

• Hamilton Lane (HLNE) 59.7

• Barrick Mining (B) 59.2

• First Solar (FSLR) 59.0

• Blue Bird (BLBD) 58.4

• Palomar (PLMR) 58.3

• RenaissanceRe (RNR) 56.7

• MercadoLibre (MELI) 55.2

List of top stocks which are among the top 10 in both lists

• Evercore (EVR) 77.0 (AI 68.0), AVG: 72.5

• Halozyme Therapeutics (HALO) 73.6 (AI 70.0), AVG: 71.8

• Slide Insurance (SLDE) 76.7 (AI 66.7), AVG: 71.7

• Mako Mining (MAKO) 73.5 (AI 69.6), AVG: 71.7

• Kinross Gold (KGC) 77.0 (AI 65.6), AVG: 71.3

• Atour Lifestyle (ATAT) 75.0 (AI 66.3), AVG: 70.7

• dLocal (DLO) 75.2 (AI 64.3), AVG: 69.8

Subscribe now

Overall score - chart overview & discounted cash flow analysis (as of May 15, 2026)

Evercore (EVR)

Kinross Gold (KGC)

Slide Insurance (SLDE)

Newmont (NEM)