Chapter 40: Your Peptide Evidence Dossier — Putting It All Together

"The criterion of the scientific status of a theory is its falsifiability, or refutability, or testability." — Karl Popper, Conjectures and Refutations (1963)

Overview

Thirty-nine chapters ago you were asked to pick five to ten peptides you had actually wondered about, write down what you already believed about each one, and date it. If you did that, you have been carrying a document through this entire book, adding a field or a skill at a time. This chapter is where you finish it, check it, and start using it.

That sequencing was deliberate. You could not have written Field 5 before Chapter 5 taught you to read a trial design, and you could not have written Field 12 honestly after you were invested — which is why Chapter 6 asked for it early, before the compounds became yours. The dossier was not homework running alongside the book. It was the book's actual output, and the chapters were the method for producing it.

So this chapter has three jobs, and only the first is the obvious one.

The first is assembly: walk all twelve fields once as a completion checklist and fill the gaps you skipped. Most readers skipped the same three.

The second, and the one that will teach you the most, is the internal audit. Not are your ratings correct — Chapter 37 already let you compare yours against the book's. The harder question is whether you applied one standard. A dossier where the compounds you liked got a generous reading and the ones that sounded like marketing got a hostile one is not a reference work. It is a record of your priors with citations attached, and it will quietly mislead you for years.

The third is handoff: keeping the thing current, using it in three specific situations, and — the part that justifies the whole project — pointing the same twelve questions at a claim that has nothing to do with peptides at all.

One thing this chapter is not: an ending. Four chapters follow it, and they are the field test.

In this chapter, you will learn to:

• Say precisely what you built and why it is organized by question rather than by compound
• Complete all twelve fields, including the three almost everyone leaves blank
• Audit your dossier with the same-evidence test, the population check, the date check, and the Field 12 check
• Find the direction of your own error, in either direction, and recognize that harshness is not rigor
• Maintain the document with dated versions and event-triggered review
• Resist rating drift using the mechanism you already wrote down
• Use a dossier to answer your own question, to prepare a clinical conversation, and to say "I don't know" precisely
• Transfer the twelve-field structure to a supplement, a diet, or a consumer medical device

Learning Paths

All five paths read this chapter in full, and this is the one chapter where that instruction is not negotiable. The dossier is the deliverable; §40.3 is the section that determines whether it is any good.

💊 GLP-1 — §40.7. You are the most likely reader to be carrying a real, near-term decision, and converting a verdict into a clinician-answerable question is what makes the dossier worth the hours you spent on it. 🏋️ Performance — §40.3 and §40.6. Your compounds have the thinnest evidence and the loudest advocacy, which makes them the ones where an inconsistent standard does the most damage and where upward drift is fastest. 🔬 Science — §40.4 and §40.9. The bias you are most at risk of is the second one in §40.4, the one that feels like rigor. §40.9's argument that a well-specified open question is a result is aimed squarely at you. 💄 Cosmetic — §40.8. The claims you meet next will be about a serum, a device, or an ingredient rather than a drug, and the twelve questions do not change. 🏥 Clinical — §40.7 and §40.10. You will spend more time on the receiving end of other people's dossiers than on your own, and §40.10 is what happens when this method meets claims about populations rather than patients.

40.1 What you have actually built

Start by being precise about the object, because most people describe it wrong.

It is not a summary of this book. A summary would be a compression of what the book concluded, and it would be obsolete on a schedule set by other people's research programs. Appendix A is the book's ratings; Chapter 37 is the master table. Those exist, they are useful, and copying them into a notebook would have taught you nothing. A summary is a snapshot of someone else's conclusions.

It is a personal reference on the compounds you chose, in a form you can update. Three parts of that sentence are load-bearing.

Personal — you picked five to ten compounds because you had a live question about each. The reason is practical: a dossier about compounds you are not curious about does not get maintained, and an unmaintained dossier eventually tells you something that stopped being true. That is worse than having none, because you will trust it.

On the compounds you chose — not on the fifty-odd this book covers. A reference you can hold in your head beats a comprehensive one you consult twice.

In a form you can update — the part that separates a dossier from notes, and the reason Field 12 exists. Because you wrote down what would change each conclusion, you can evaluate new information against a standard you set before you had a stake in the answer.

But the deepest structural fact about what you built is this one:

It is organized by question rather than by compound.

That sounds like a filing decision and it is not. Consider what an alternative organization would look like. If you had written one page per compound in free prose, you would have twelve paragraphs of mechanism and no verdicts — because mechanism is the easiest thing to write and the least informative thing to know, and an unstructured effort always drifts there. More importantly, your file would only work on the compounds it contained. Encounter a new one and you would be starting from nothing.

The twelve fields are not headings. They are twelve places where a specific error is easy to make, arranged so that making it is visible. Field 8 exists separately from Field 7 so the gap between what a thing is approved for and what it is sold as cannot hide. Field 5 has sub-lines for population and endpoint and comparator so that "there are studies" cannot pass as an answer. Field 12 exists so that a conclusion cannot quietly become a belief.

Which means the structure works on compounds it does not contain. A peptide you have never heard of, released next year, walks into the same twelve questions and gets sorted by the same failure modes. So does a compound that is not a peptide at all — which is §40.8, and which is the actual point of the exercise.

🔍 Check Your Understanding

1. Why would a dossier copied out of Appendix A be worse than a shorter one you built yourself?
2. In your own words: what is the difference between organizing by compound and organizing by question, and which one keeps working on a molecule that is not in the file?
3. Name one field whose entire purpose is to make a specific error visible, and say what the error is.

40.2 The full assembly

Here is the completion checklist. Walk it once for every entry. For each field: what it holds, and the one failure mode it invites. Appendix C states these in a table and gives you the blank template — this section is the walkthrough, and it is worth doing with your actual document open rather than reading it as prose.

Field 1 — Identity. Holds: the molecule as a specification, not a name. Sequence or length, modifications relative to the native molecule, brand names, research codes. Fails by: recording a trade name or a nickname as though it were a specification. The line to think hardest about is the last one — is the same name reliably the same molecule? For a pharmacy product, yes. For a research-labeled vial, no, and that is a fact about the state of your knowledge rather than about the compound.

Field 2 — Origin. Holds: endogenous, analog, found in another organism, selected from a library, designed, fragment. Fails by: treating an origin story as evidence. Fill it in, then discount it. This field exists so the origin story has somewhere to live where it cannot contaminate Field 6.

Field 3 — Mechanism. Holds: one plain sentence, then the detail; the target; direct or indirect; your confidence in the mechanism itself. Fails by: letting a good mechanism substitute for an outcome. If you cannot write the plain sentence, you do not yet understand the mechanism, and saying so is more useful than transcribing a paragraph you half-follow.

Field 4 — Pharmacology. Holds: routes studied, half-life, oral bioavailability where relevant, what modification made it a drug. Fails by: ignoring route entirely. The line that does the most work is the last one — does the route it is actually used by match the route it was studied by? That gap is invisible unless both are written down.

🧬 The Molecule — the identity field, one more time, because it is the one people skip

Field 1 looks clerical. It is the field most likely to be filled in from memory, and the one where filling it in from memory does the most damage, because everything downstream inherits the error.

A worked contrast, which you built yourself in Chapter 1. Insulin has a generic name assigned by an international naming authority, a fully characterized 51-residue two-chain structure known since the 1950s, and a sequence that appears identically in every reference database. BPC-157 has a laboratory code, a consistent published fifteen-residue sequence, and no generic name — decades after its description.

Both Field 1 entries are complete. Neither required a judgment call. And the difference between them is already visible before a single piece of clinical evidence has been consulted, because a compound that never received a generic name almost certainly never entered serious formal drug development. That is not an argument that it does not work. It is a fact about regulatory history that Field 1 surfaced for free.

If your Field 1 entries say "BPC-157 — a healing peptide," you have put a claim in the identity field. Fix that one first; it is the cheapest correction in the document.

Field 5 — Evidence. Holds: the best available human evidence, described by its design — population, endpoint, comparator, size, duration, result — plus replication, plus what is conspicuously absent. Fails by: counting papers instead of reading designs. Sixty animal studies and no human trial is not a partial answer to a human question; it is a complete answer to a different one.

Field 6 — Rating. Holds: one row per claim, each row naming a population and an endpoint, each row dated, each ❌ marked as evidence absent or evidence present and negative. Fails by: one rating for a whole molecule. This is where the method either takes hold or does not.

Field 7 — Approved use. Holds: the indication, precisely, including population and any line-of-therapy restriction, and — if not approved — which of the five distinct meanings of "not approved" applies. Fails by: confusing "approved" with "approved for this." "Never submitted" and "submitted and rejected" are opposite signals wearing the same words.

Field 8 — Claimed use. Holds: what it is sold as doing, in the seller's own terms, unedited. Fails by: not recording it separately from Field 7. The gap between 7 and 8 is, for most compounds in this book, the entire story, and it is only visible because they are separate fields.

Now the three that most readers find empty when they reach this chapter. This is not a coincidence and it is not a personal failing — they are the least satisfying fields to write and the most useful to have.

Field 9 — Risks. Holds: four separate things. Adverse effects at studied use. Risks specific to unstudied use, which no label covers. Risks specific to preparation quality, which are entirely independent of whether the molecule works. And the line almost everyone leaves blank: what is unknown because nobody has looked. Fails by: recording only the risks of misuse, or — worse — writing "no known side effects" for a compound with no human trials.

🩺 Safety and Risk — why "no reported side effects" is the emptiest sentence in the field

For a compound that has never been through a human trial, the honest Field 9 entry is not "well tolerated" and it is not "no known side effects." Both of those sentences describe a reporting system, and for an unapproved compound sold outside a regulated channel, there isn't one.

Consider what has to exist for "no reported side effects" to carry information. Someone has to experience an effect. They have to attribute it to the compound rather than to training, illness, or the four other things they are taking. They have to know where to report it. Somebody has to collect the report, and somebody has to publish the aggregate. For an approved drug, every one of those links exists and is a legal requirement. For a research-labeled vial bought online, none of them do.

So the correct Field 9 entry, and it should feel unsatisfying, is: "Unknown. 'No reports' is not the same as 'no harms' when there is no reporting pathway."

The unsatisfying version is the accurate one. Write it.

Field 10 — Status. Holds: five genuinely different questions — regulatory status, sport and anti-doping status, how it is actually being sold, access and cost and supply, and what can and cannot be established about preparation quality. Fails by: collapsing five questions into one. And the discipline that matters: keep this field in a column entirely separate from Field 6, and never let one adjust the other. A compound can be approved somewhere and thinly evidenced. A compound can be unapproved everywhere and simply untested.

Field 12 — What would change my mind. Holds: what would move this up, what would move it down, and the study that would settle it — population, endpoint, comparator, duration, rough size — plus whether such a study is underway. Fails by: being left blank, which converts the entry from a conclusion into a belief.

Now go through every entry and fill the gaps. Not the ones you find interesting. The empty ones. If you are like most readers, that means 9, 10, and 12, and it means an hour or two of genuinely tedious work.

The honest case for doing it: those three fields are what make the document usable by a future version of you who has forgotten how each entry was made. Field 9 is what someone will actually ask you about. Field 10 changes fastest and therefore dates your file. And Field 12 is the only thing standing between a dossier and a scrapbook of opinions.

40.3 The internal audit

This is the most valuable section in the chapter, and it asks a question you have not been asked before.

Chapter 37 let you compare your ratings against the book's. That is a useful exercise and a limited one, because it measures you against an external standard that is itself dated and that you are not obliged to accept. If you rated a compound ⚠️ where this book rated it ❌, you might be wrong, or you might have read a trial the book did not, or you might weight surrogate endpoints differently. A disagreement is not automatically an error.

The audit asks something harder and more diagnostic:

Did you apply one standard?

Not are your ratings right — are your ratings mutually consistent. Because inconsistency is detectable from inside the document, without any external reference, and inconsistency is nearly always the fingerprint of something other than evidence doing the deciding.

Four checks. Do them in order. Budget an hour.

Check 1 — The same-evidence test

Find two entries in your dossier whose evidence bases are comparable and whose ratings differ.

Comparable means roughly matched on the things that should matter: study design, population relevance, endpoint type, replication, size. Two compounds with animal data only. Two compounds with one small open-label human trial each. Two compounds with a single positive randomized trial in a narrow population and no replication.

If you find such a pair, you have three possible situations:

1. The difference is justified by design, population, or endpoint. One of the trials used a hard outcome and the other a surrogate. One studied a population that includes you and the other did not. One had a comparator and the other did not. Write the justification into the entry. This is a real finding, and the document is better for having it explicit.

2. The difference is justified by something outside Field 5 that you have not written down. You rated one lower because of a Field 9 concern or a Field 10 quality problem. That is not necessarily wrong, but it is a different claim than the one Field 6 is making, and it belongs in the fields that hold it. A rating that silently imports a safety concern is a rating you cannot audit later.

3. You cannot justify the difference. Then one of the two is wrong, and the interesting question is which. Usually — not always, but usually — the outlier is the one where you had a prior.

Most readers doing this honestly find at least one pair. That is normal. The dossier is not supposed to arrive clean; it is supposed to arrive auditable.

Check 2 — The population check

Read every row in every Field 6. Does each one name a population?

Rating rule 1 says a rating attaches to a claim, and a claim has a population and an endpoint. So a row that reads:

"Weight loss — ✅"

is not a rating. It is a rating of a molecule, which rule 1 forbids, and it is wrong in a way that matters rather than a way that is pedantic. This book's own worked entry makes the point in a single comparison: the same drug at the same dose produced roughly a 15% reduction in body weight in adults with overweight or obesity without diabetes, and roughly 10% in adults with type 2 diabetes. Same molecule, same exposure, materially different result. Which number your unpopulated ✅ was pointing at is now unrecoverable.

The corrected row names both halves of the claim:

"Weight loss, adults with overweight or obesity and no diabetes, body weight at 68 weeks — ✅ 2026"

Longer, less elegant, and actually a statement about the world.

An unpopulated rating is the single most common defect in reader dossiers, and it is the one that most reliably survives into how people talk. Every time you hear "semaglutide works" or "BPC-157 doesn't work" without a population attached, you are hearing this defect out loud.

📊 Evidence Rating — what a complete row looks like, and what each part is doing

Component	Example	Why it is there
Population	adults with established cardiovascular disease and overweight or obesity, without diabetes	Rule 1. A result belongs to the people it was measured in.
Endpoint	major adverse cardiovascular events	Distinguishes hard outcomes from surrogates. Weight is a surrogate; a heart attack is not.
Rating	✅	Four symbols, and the symbol is the least informative part of the row.
Date	2026	Rule 5. An undated rating is a claim about the eternal.
If ❌, which kind	—	Evidence absent and evidence present-and-negative are different states of knowledge.

Notice the proportions. The population and the endpoint take more characters than everything else combined, and that is correct. The symbol is a summary of the row, not the content of it. A reader who copies your symbols and discards your populations has copied nothing.

And notice what the row does not contain: any word describing how you feel about the compound, its marketing, its advocates, or its origin story. Those have their own fields.

Check 3 — The date check

Is every rating row dated?

An undated rating is a claim about the eternal, which is not the kind of claim anyone is in a position to make. A rating that does not say when it was made cannot be evaluated against what has happened since.

This is the cheapest of the four checks and the most likely to be skipped, because it feels clerical. It is not. The date is what converts your dossier from a set of opinions into a time series, and §40.4 depends entirely on having that time series.

Check 4 — The Field 12 check

Is every entry falsifiable?

Go through every entry and ask: if the world contained evidence that this conclusion was wrong, would I recognize it? Field 12 is where that recognition is specified in advance.

An entry with an empty Field 12 is not a conclusion. It is a belief.

And this is as true of a confident ❌ as of a hopeful ⚠️. If you rated something ❌ and cannot say what finding would move it — not "better evidence," but the actual study, with its population, endpoint, comparator, duration, and rough size — then you did not reach that ❌ by evaluating evidence. You reached it some other way and the evidence review was the paperwork.

That sentence is uncomfortable and it is meant to be. It is also the single most useful diagnostic in the book, because it catches the failure mode that is invisible from inside: the conclusion that happens to be correct, arrived at by a method that does not work. Being right for bad reasons feels identical to being right. Field 12 is the only place the difference shows up.

🔬 Read the Study — what a Field 12 entry has to specify to count

Compare two Field 12 entries for the same open question.

Version A: "More research is needed. If a good study comes out showing it works, I'd upgrade this."

Version B: "To move this up: a randomized controlled trial in adults with the condition in question, with a functional endpoint rather than a biomarker, a placebo comparator, at least twelve months of follow-up, and enough participants to detect a clinically meaningful difference — registered before enrollment. To move this down: the same trial, reported null. Underway? Not that I can find."

Version A cannot be checked against reality. Every conceivable piece of news is consistent with it, which means it excludes nothing, which means it is not doing any work. In practice it will be satisfied by whatever encouraging thing you read next, because it was written to be satisfiable.

Version B names five things that a specific future publication either has or does not have. You can hold a paper up against it. Crucially, you can also hold up a press release and a conference abstract and watch them fail — which is exactly the situation Field 12 was built for.

The test for a Field 12 entry: could a reasonable person disagree with you about whether it had been satisfied? If yes, it is not specific enough yet.

40.4 Finding the shape of your own error

Now use the audit for the thing it is actually for.

Chapter 37 asked whether your individual ratings matched the book's. This asks something the individual comparisons cannot answer:

Did the disagreements run in one direction?

Lay them out. For each compound where you and the book differed, note which way. Then look at the column, not the rows. Scattered disagreement in both directions is mostly noise — different readings of the same ambiguous literature, which is expected and healthy. Disagreement that runs consistently one way is a bias, and it is measurable, and almost nobody has ever measured theirs.

There are two directions, and they deserve equal billing.

The generous reader. You rated things higher than the evidence supports, and if you look at which things, they are the ones you hoped would work. The compound you were already taking. The one your training partner swears by. The one that would solve a problem you actually have. The mechanism is straightforward: when you want a conclusion, ambiguous evidence reads as supportive, absent evidence reads as "not yet studied" rather than "unsupported," and a promising early trial reads as a preview of the confirmatory one rather than as a single unreplicated result. This bias is well known, readers expect it in themselves, and it is the one every book about evidence warns about.

The harsh reader. You rated things lower than the evidence supports, and if you look at which things, they are the ones that arrived wrapped in marketing. The compound with the slick clinic website. The one with the influencer. The one whose advocates make claims that embarrass the underlying data. You discounted the evidence because of the company it kept.

This second bias feels like rigor and is the same failure.

That sentence is the point of the section, so let me make the case rather than assert it.

Rating rule 4 says: never downgrade with distaste. It is the exact mirror of rule 3, never upgrade with mechanism, and it exists for the same reason — both rules forbid letting something other than evidence move a rating. Rule 3 forbids a good story moving a rating up. Rule 4 forbids a bad salesperson moving it down.

And the harsh reader has a real problem the generous reader does not, which is that they cannot feel it happening. Skepticism is socially and intellectually rewarded. Nobody has ever been embarrassed for having been too critical of a supplement. So the harsh reader gets no corrective signal, ever, and their errors compound quietly.

Consider what the error costs. A compound that is genuinely promising but preliminary — real human data that does not settle the question, which is precisely what ⚠️ means — gets rated ❌ because its marketing is offensive. Now your dossier says the evidence is absent or negative when it is neither. Two years later a confirmatory trial reads out positive and you experience it as a surprise, when it should have been the expected resolution of an open question you were already tracking. You were not protected by your harshness. You were blinded by it, and you lost the ability to see the result coming.

There is a second cost. The harsh reader tends to be the person others ask, and a ❌ that is distaste in a symbol's clothing transmits an aesthetic reaction as though it were an evidence assessment. The person receiving it cannot tell, because the symbol looks identical either way.

Neither bias is worse than the other. They are the same mistake pointing in opposite directions: a rating moved by something that is not evidence. The generous reader is more likely to take something that does not work. The harsh reader is more likely to dismiss something that does, and to be confidently wrong in a way that never gets corrected.

🔍 Check Your Understanding

1. You rated a compound ❌ and, on audit, discover that your Field 5 describes real human data from a small randomized trial with a positive result. Which rating rule did you break, and what is the correct rating?
2. Why is the harsh reader's bias harder to detect from the inside than the generous reader's?
3. Your disagreements with Chapter 37 are split roughly evenly in both directions. Is that evidence of bias? What would it take to conclude that it is?

40.5 The living document

A dossier is a working file. Three habits keep it alive, and one discipline keeps it honest.

Date every change and keep the old version

Not because you will want to read the old versions — you mostly will not — but because a dossier with a history tells you something a current snapshot cannot: how you were wrong before, and in which direction.

This is the machinery §40.4 runs on. You cannot correct a bias you have never measured, and you cannot measure a bias from a single snapshot, because a snapshot contains no trajectory. Two years of dated versions contain one.

VERSION HISTORY — one entry, three revisions

  2026-02  Compound X | claim: [population, endpoint] | rating: 🔬
           Field 12: RCT in [population], functional endpoint, placebo
           comparator, 12 months, adequately powered. Underway? unknown.

  2026-09  No change to rating. Note added: two conference abstracts and a
           mechanism paper appeared. NONE of these is the readout named in
           Field 12. Rating unchanged. <- this is the entry doing its job

  2027-04  Rating moved 🔬 -> ⚠️. Trigger: the trial named in Field 12
           reported, positive, single site, smaller than specified.
           Field 12 updated: multi-site replication required for ✅.

The middle line is the one that matters, and it is the line most people never write. A revision that records "news arrived and the rating did not move" is more valuable than one that records a change, because it is evidence that the standard held under pressure. A file where every revision is an upgrade is not a file that has been tracking evidence. It is a file that has been tracking enthusiasm.

Review when something happens, not on a schedule

Calendar-driven review sounds disciplined and produces the wrong behavior: it makes you go looking for news about compounds you are invested in, at intervals, which is a machine for manufacturing upgrades. There is always something. There is never nothing.

Review when an actual event occurs:

• A trial reads out — especially the one you named in Field 12
• A regulator approves, restricts, or withdraws something
• A shortage begins or resolves, which changes Field 10 substantially
• A safety signal appears in post-marketing surveillance
• Somebody asks you a question you cannot answer. This is the most reliable trigger of the five and the most easily dismissed. A question you could not answer has located a gap in the document. Go fill it while you still remember what the question was.

Expect to remove entries

A compound you have stopped caring about should come out. A dossier is a working file, not an archive. Ten entries you maintain beat thirty you do not, and a stale entry is not neutral — it will eventually tell you something that used to be true.

And the discipline that matters most

Nothing upgrades a rating except the readout named in Field 12.

Not a press release. Not a conference abstract. Not a preprint. Not a new mechanism paper, however elegant. Not a funding round. Not a clinician's enthusiasm. Not a friend's result. Not a review article by an author with a stated interest. Not a podcast.

This is rating rule 3 applied to your own file, and it is the rule you will be most tempted to break — because every one of the things on that list is information, and the temptation is not irrational. A mechanism paper genuinely does tell you something. The problem is that it tells you something about plausibility, and plausibility was never what the rating was tracking.

📋 Your Evidence Dossier — the maintenance line

Add one line to the top of your document, above every entry:

"A rating moves only when the readout named in Field 12 arrives. Everything else gets recorded as a note and changes nothing."

It looks like a slogan. It functions as a gate, and its value is that it is written down where you will see it at the exact moment you are about to break it — which is the moment you are excited, and therefore the moment you will not be reasoning well.

40.6 Rating drift, and why it goes one way

Now the failure mode that will actually degrade your dossier over the next several years. Not error — drift.

Ratings creep upward. Almost never downward. Everyone's do. And the reason is not that you are credulous; it is structural, and understanding the structure is what lets you defend against it.

Every piece of news that reaches you about a compound you are watching is, by selection, news somebody thought was worth publicizing.

Follow the chain. A trial reads out positive: press release, conference presentation, coverage, enthusiasm from people invested in the result. A trial reads out null: it may be published quietly, published late, or not at all, and there is no press release, because nobody has ever built a communications strategy around a compound not working.

Then layer your own attention on top. You follow the compounds you have entries for, and you have entries for compounds you care about. Caring produces searching, and searching a space where positive results are preferentially published returns positive results. The filter runs twice, and both passes point the same way.

This is Chapter 6's testimonial dynamic operating on your own attention. Chapter 6 showed why testimonials are systematically misleading: the people who improved talk, the people who did not mostly stop showing up, and the resulting sample is selected on the outcome. Drift is that same selection applied to publications and press coverage instead of people, and to you instead of a forum. You are not sampling the evidence. You are sampling the reported evidence, and it has been filtered on the outcome before it reached you.

The uncomfortable implication: you will feel increasingly confident that a compound is working out, and that feeling is a product of the filter rather than of the evidence. It is exactly what the mechanism predicts you would feel regardless.

⚠️ Hype Check — "there's a lot more evidence for this now than there was two years ago"

The claim, in its usual form:

"When you first looked at this there was almost nothing. Now there are dozens of papers, a conference session, and a company running a trial. The picture has really changed."

What's true in it. The volume of material genuinely has increased. The trial genuinely is running. None of that is fabricated, and the person saying it is not lying.

Where it fails. Volume is not evidence — Field 5's failure mode is counting papers instead of reading designs. The relevant question is not how much material exists but whether the specific readout you named in Field 12 has arrived. A trial being underway is a fact about the future. It is the correct thing to record in Field 12's last line, and it moves nothing, because a running trial has by definition not reported. Compounds entering clinical development mostly do not become approved drugs; an active trial raises the probability that we will soon know something and leaves the probability that the answer is yes roughly where it was.

The defense is mechanical, and it is why Field 12 was written before you were invested. You do not have to adjudicate whether the picture has changed. You wrote down what would change it. Open the entry, read the line, and ask whether that happened. Usually it has not, and the whole question resolves in under a minute.

Verdict: an accumulation of material is not an accumulation of evidence, and the feeling of a picture changing is exactly what selective reporting produces when nothing has changed at all.

The defense is not vigilance. Vigilance fails, because drift does not feel like drift — it feels like being appropriately responsive to new information. The defense is that you already wrote down what would move it, at a moment when you had no stake in the answer. Check whether that specific thing happened. Which is the whole reason Chapter 6 made you write Field 12 before you needed it.

40.7 What the dossier is for, in practice

Three uses. If your dossier does not do these three things, it is not finished.

Use 1 — Answering your own question

You came to this book with a question. Most readers came with one of four: a family member started a GLP-1 drug; a training partner injects something; there is a \$90 bottle of serum in the bathroom; a clinic sent an email about optimization and longevity.

Go answer it. Open the relevant entry and read Field 11 with Fields 5, 9, and 10 next to it.

The answer is rarely a yes or a no. It is usually a decomposition: the original question turns out to contain three or four different questions, some with good answers and some without. "Does BPC-157 heal tendons?" decomposes into: is there human evidence for tendon healing (Field 5), does what is being sold match anything studied (Fields 7 and 8), what can go wrong including from preparation quality (Field 9), and what could I actually find out (Field 12).

A decomposed question is a better object than the original, even when no part of it resolves, because each part now has a different status — and one of them is usually answerable.

Use 2 — Having the Chapter 39 conversation

Chapter 39 was about making a clinical conversation useful. The dossier is what you bring to it, and the conversion it performs is specific: a verdict becomes a question a clinician can answer.

This conversion is a test as much as a courtesy. A verdict that cannot be turned into an answerable question was probably a feeling.

Read down the right column and notice what changed. Every question names something the person across from you actually has access to — their read of the literature, your history, their experience of what happens on discontinuation. A good question is one whose answer is in the room.

💊 In the Clinic — what the dossier does to the conversation

A clinician's time with you is short and is mostly consumed by establishing what you are actually asking. Arriving with Fields 7, 8, 11, and 12 legible does something specific to that exchange, and it is not that you sound informed.

It is that you have pre-specified what would change your mind, which is a rare and useful thing to put on the table. It converts the encounter from a negotiation — where one party wants something and the other decides — into a shared evaluation of a question. Those are different conversations, and the second one goes better for reasons that have nothing to do with charm.

It also protects against the failure mode where you arrive having decided, ask a question shaped to confirm the decision, and receive an answer you were always going to receive. If Field 12 says what would move you, you can hand over the actual disagreement rather than a rehearsal.

The standing constraint from Chapter 1 has not moved and does not move in this chapter: a rating is an input to a decision and not the decision, and the decision belongs with someone who knows your history, can order your labs, and is reachable at two in the morning. A dossier makes you a better participant in that conversation. It does not make you the other party in it.

Use 3 — Saying "I don't know" precisely

This is the use nobody anticipates and the one that changes how you talk.

"I don't know," "nobody knows," and "it's been tested and it doesn't work" are three different sentences. They are routinely collapsed into one shrug, and the collapse destroys most of the information you spent forty chapters acquiring.

• "I don't know" — a fact about you. The evidence may be excellent; you have not looked. This is a Field 5 that is empty because you did not fill it. The correct next step is to go look, and saying it plainly tells the person that the question is open on your side, not the world's.

• "Nobody knows" — a fact about the world. The trial has not been run. This is ❌, kind: evidence absent. It is compatible with the compound working spectacularly. It licenses nothing, but it forecloses nothing either, and it identifies precisely what would resolve it: the trial nobody ran.

• "It's been tested and it doesn't work" — a much stronger state of knowledge, and a rarer one. This is ❌, kind: evidence present and negative. The trial was run properly, with an adequate population and a real comparator, and the answer was no.

Readers consistently get the second and third backwards, and it is worth being blunt about the consequence. A compound with a large, well-conducted, null outcome trial is one we understand well. A compound with no human trial at all is one we understand not at all. The second is the more uncertain situation by a wide margin, and it is routinely described as the more promising one — on the grounds that nothing has disproved it, which is true of every claim that has never been examined.

The dossier lets you say which of the three you mean, and being able to say which is the difference between an informed person and a cautious one.

🔍 Check Your Understanding

1. Someone asks whether a compound works and you say "there's no evidence." Which of the three sentences did you mean, and how would each version change what they should do next?
2. Take your own most important verdict and write it as a question a clinician could answer. If you cannot, what does that tell you about the verdict?
3. Why is a compound with a large null trial better understood than a compound with no trials, and why does that feel backwards?

40.8 The transferable skill — this was never about peptides

Here is the claim this section has to make good on: the peptides were the curriculum, and the judgment is what you keep.

That is easy to assert and easy to disbelieve, so let us test it. Take the twelve fields somewhere they have no business working and see whether they do.

The claim: time-restricted eating — confining all food intake to a window of a few hours each day — improves metabolic health and produces weight loss.

Not a peptide. Not a drug. No molecule, no receptor, no manufacturer, no vial. Watch the fields work anyway.

1 — Identity. What is this, exactly? And immediately the field earns its keep, because "time-restricted eating" is a name, not a specification. How long is the window — eight hours, ten, six? Where is it placed — early in the day or late? Is total intake also restricted, or explicitly not? Two studies both labeled "time-restricted eating" can be testing materially different interventions. This is Field 1's failure mode exactly: a name treated as a specification. It is the identical error as recording "BPC-157" without a sequence.

2 — Origin. Circadian biology, and a body of rodent work in which animals given the same daily intake within a restricted window fared better than animals eating the same amount across the whole day. It is a genuinely interesting origin. It carries no evidentiary weight about humans, exactly as a lizard-venom origin carried none about exenatide's efficacy and a human-gastric-juice origin carries none about BPC-157's. Record it; do not lean on it.

3 — Mechanism. Proposed: alignment of intake with circadian metabolic rhythms; extended daily fasting; downstream cellular processes. And a competing, much duller mechanism: people who eat within a shorter window usually eat less. Note that both are on the table, note that the second one predicts the same outcome for a completely different reason, and note the standing rule — mechanism never upgrades a rating. The elegant explanation and the boring explanation make identical predictions about weight, and only a trial design can separate them.

4 — Pharmacology → exposure and adherence. The field translates. For a drug, the question is whether the route used matches the route studied. For a behavior, it is whether the adherence achieved in a trial resembles adherence in life, and whether intake was measured or self-reported. A trial that provides food and verifies intake is testing something different from an instruction to eat within a window and report back. Same failure mode: the studied version and the used version are not the same intervention, and the gap is invisible unless both are written down.

5 — Evidence. Now the field does its real work, and the sub-lines do all of it. The relevant question is not how many papers exist. It is: what is the comparator? A trial of time-restricted eating against no intervention answers a question almost nobody is asking. The informative design is against calorie-matched continuous restriction, because that is the only comparison that isolates the window from the eating-less. Then: what endpoint — body weight, which is a surrogate, or a downstream outcome? What duration — because early weight change and twelve-month weight change are famously different quantities. And what is conspicuously absent? As of this writing, the absence is long-duration trials with hard outcomes.

6 — Rating. Three rows, and the fact that there are three is the whole demonstration of rule 6 — one intervention, many ratings:

Read those three rows together, because they are the point of the section. They are not in tension. The same intervention is well tested and unimpressive for one specific claim, plausible and underpowered for a second, and entirely untested for a third — and a single overall verdict on "intermittent fasting" would destroy all three distinctions. This is exactly what happens when someone gives semaglutide one rating.

7 — Approved use. No regulator approves eating patterns, and this is informative rather than damning. For a drug, "not approved" has five distinct meanings and you must say which. For a behavior, the honest entry is there is no approval pathway, so the absence of approval carries no information — which is a different sentence from "not approved" and must not be confused with it. Recognizing when a field is structurally empty rather than damningly empty is a real skill.

8 — Claimed use. Metabolic health, weight loss, longevity, "cellular cleanup," cognitive benefits, cancer prevention. Compare with Field 7. The gap between 7 and 8 is, once again, the entire story — and it is only visible because the two are recorded separately.

9 — Risks. Not zero, which is where a "natural, free, and just a schedule" framing tends to stall. At studied use: generally minor for most adults. In unstudied use: interactions with medications that require food, and glucose management for people on glucose-lowering therapy — both of which are clinician questions, not book questions. Specific to the person rather than the intervention: a history of disordered eating, where imposing rules around timing is a real consideration. And the honest last line: what is unknown because nobody has looked, at multi-year scale.

10 — Status. And here the field almost empties out, which is itself a finding. No regulatory status. No supply. No cost. No preparation-quality question — there is no vial, no certificate, no counterfeit, no concentration to verify. Chapter 19's entire risk category simply does not apply. That absence changes the decision materially, and it changes it in a direction that has nothing to do with efficacy, which is precisely why Fields 6 and 10 are kept in separate columns.

11 — Verdict. For an adult whose goal is weight loss: the evidence does not support the eating window as an independent lever, and does support it as one of several ways some people find it easier to eat less. Confidence: moderate. As a question for a clinician: "Given my medications and my history, is there any reason a compressed eating window would be a problem for me?"

12 — What would change my mind. Up: a preregistered randomized trial against calorie-matched continuous restriction, in a relevant adult population, at twelve months or longer, adequately powered, showing a difference attributable to timing. Down: consistent null replications of the adherence claim. The study that would settle the longevity claim: it is not currently feasible in humans at the relevant duration, and saying so plainly is more useful than any rating.

Now look at what happened. Twelve fields, no molecules, and the same four things went wrong in the same four places: a name mistaken for a specification, an origin story doing work it cannot do, a mechanism competing with a duller mechanism that predicts the same result, and a claimed use that outruns anything established. Those are not peptide errors. They are the errors people make about complicated systems when the specificity of a claim is doing more persuasive work than its evidence.

Try it on something you have already bought: a supplement in your cabinet, a device in your bathroom, a protocol in your training. Reliably, Fields 5, 9, and 12 are the ones you cannot fill — the same three you skipped in your peptide entries — and the difficulty of filling them is the finding.

You will not remember most of this book. That is normal and the book was built expecting it. What should survive is the shape of the twelve fields: the reflex that makes you ask, when somebody tells you a thing does something — for whom, measured how, compared to what, and how would we know if it were false.

That reflex is not about peptides. The dossier was the training apparatus. The reflex is the thing you keep.

40.9 What this book could not teach you

Five honest limits. Not a disclaimer — a specification of the tool's edges, which is the last thing a good reference does for you.

It cannot make you a clinician. You now know more about peptide evidence than most people you will meet, including some who sell these compounds — which is not the same as being able to evaluate you: your history, your medications, your labs, the interaction you did not think to mention. This is not modesty but a category difference. The dossier evaluates claims about populations; a clinician evaluates a person.

It cannot substitute for a trial nobody ran. No amount of careful reading produces evidence that does not exist. For several compounds readers care most about, you have now read everything worth reading and arrived at genuine uncertainty. That is not a failure of your reading. It is the correct output.

It cannot tell you what to do. A rating is an input to a decision, not the decision. Two people with identical dossiers — same fields, same evidence, same ratings — can reasonably choose differently, and neither is misreading the file. They differ in how they weight things the dossier does not contain: tolerance for uncertainty, the cost of the problem they are trying to solve, what happens if it goes wrong. A person managing a condition that is degrading their life daily and a person optimizing something already fine will read the same ⚠️ and correctly reach opposite conclusions. The dossier's job is to make the disagreement be about values rather than facts. That is a real service, and it is not the same as deciding.

It cannot keep itself current. Every rating here is dated, and the dating is not a formality — several compounds will be rated differently within the life of this edition. This is the specific respect in which your dossier is more valuable than the book: it will be updated by someone who knows how each entry was made and therefore what would unmake it. The book stops. You do not.

And it cannot make the evidence exist.

This is the one worth sitting with, because for most of the compounds readers arrive here about, the honest final position is not a verdict at all. It is a well-specified open question: nobody has run the trial; here is exactly what the trial would have to look like; here is what we would learn; here is what we currently cannot say.

That can feel like the book failed you. Forty chapters, and the answer to the question you came with is we don't know.

Stating a question precisely is a real result, not a failure, and here is the argument for that rather than the assertion.

Compare where you started. "Does BPC-157 work?" is a question with no answer, because it has no population, no endpoint, no comparator, and no timeframe — it cannot be resolved even in principle, and no amount of research would settle it, because there is nothing specific enough to settle. What you have now is different in kind: for this population, on this endpoint, against this comparator, over this duration, no adequately designed human trial has been conducted; the animal literature is extensive and does not transfer; here is the trial that would answer it; it does not appear to be underway.

The second version is not a shrug. It is a map of a specific hole in human knowledge, drawn accurately enough that you would recognize the moment it got filled. You can hold a future paper up against it. You can explain it to someone in ninety seconds. You can tell whether a new headline is relevant to it. You can tell that a mechanism paper is not.

The first version can do none of those things. Converting one into the other is most of what the last forty chapters were for, and it is the difference between not knowing something and knowing exactly what you do not know.

40.10 Handing off to Part VIII

The method is built. The dossier is assembled, audited for a single standard, dated, and versioned. You know what each field holds, which error it catches, and what would move each rating.

So this is not the end of the book. It is the end of the training, and the next four chapters are the field test.

Everything until now has been the controlled version. Molecules are the easy case — not because the science is simple, but because the evidence architecture exists. There are trials. Trials have arms. Arms have populations. Populations have endpoints. There are comparators, placebos, preregistration, and regulators who publish their reasoning. When someone tells you a peptide does something, there is at least a well-defined thing that would settle it, even when nobody has done it. Field 12 is writable because trials are a thing that exists.

Now point the same twelve questions at claims that have none of that.

Chapters 41 through 44 turn the method on the parts of this subject that are about society rather than about molecules:

• How a drug's brand name entered the language — becoming a verb, a joke, a shorthand, an insult — and what happens to public understanding when a specific molecule with specific indications becomes a general word.
• The creator economy that sells these compounds — the incentive structure behind the content you encounter, and how it decides which claims reach you at all. Chapter 6's testimonial dynamic, industrialized.
• Enhancement outside the clinic — what it means to use these compounds not to treat a condition but to exceed a baseline, and why that question is harder than it first appears.
• What a society on metabolic drugs can honestly claim about itself — the least tractable question in the book, and the one where the temptation to overclaim runs in every direction at once.

These are harder, and the difficulty is specific rather than vague. There are no trial arms. You cannot randomize a culture. There are no endpoints — nobody has defined the outcome measure for "a drug name entering the language," and the plausible candidates all embed a value judgment. There is no placebo, because there is no version of the society that did not get the drug, running alongside for comparison.

So the twelve fields do not apply cleanly, and the honest thing is to say so up front rather than pretend the method scales without loss. What survives the transfer is the discipline underneath the fields: for whom, measured how, compared to what, and how would we know if it were false — applied to claims where the answers are harder to get and, for that exact reason, asserted with far more confidence than the evidence permits.

And here is why these four chapters are not an appendix to the book but the point of it.

Claims about molecules are made about compounds. Claims about society are made about you. You will encounter far more of the second kind, they will be far more confident, and almost nobody applies any standard to them at all. A person who can evaluate a trial of a GLP-1 drug and cannot evaluate a confident claim about what that drug is doing to a culture has learned something narrow. The whole argument of §40.8 is that the skill transfers — and Part VIII is where that argument gets tested against the hardest available case rather than a friendly one.

Close the dossier. Bring the reflex.

📋 Your Evidence Dossier

This is the largest checkpoint in the book, and the last one that builds the document rather than using it. Set aside a real block of time — two to three hours — and do it in one sitting if you can, because the audit steps only work when you can see all the entries at once.

Work in order. Do not skip ahead to the parts that sound interesting; the ordering is what makes the audit possible.

Step 1 — Assemble (about 45 minutes)

Put every entry in one place, in one format. If your notes are scattered across a notebook, a phone, and three browser tabs, consolidate now. The audit is impossible across formats.

Then, for each entry, walk all twelve fields as a completion checklist and mark each one: complete / partial / empty. Do not fill anything yet. Just mark. You want the map before you start work.

Step 2 — Fill the gaps (about 45 minutes)

Now fill the empty ones — and fill the empties before you improve the ones you find interesting. If you are like most readers, you are looking at Fields 9, 10, and 12 across most entries.

Rules for this step:

• A blank is allowed and a guess is not. If nobody knows, write "unknown — nobody has looked." That is a complete Field 9 entry and a true one.
• Write where you got it. In six months you will not remember which sentences you were sure of.
• Field 12 is not optional. An entry without it does not count as an entry.

Step 3 — The four audit checks (about 45 minutes)

Run §40.3 in order, and write the result of each check into the document rather than just noticing it:

1. Same-evidence test. Find two entries with comparable evidence and different ratings. Justify the difference by design, population, or endpoint — or fix one.
2. Population check. Every Field 6 row names a population and an endpoint. Rewrite the ones that do not.
3. Date check. Every row carries a date.
4. Field 12 check. Every entry is falsifiable. Apply the test: could a reasonable person disagree about whether the named readout had arrived? If yes, sharpen it.

Step 4 — Find your direction (about 15 minutes)

List every compound where your rating differs from Chapter 37's, with the direction of the difference. Look at the column. Then write one sentence at the bottom of your document naming your bias:

"I run generous toward compounds I want to work" — or — "I run harsh toward anything that arrives with marketing attached."

Almost everyone has one. Almost nobody has ever written theirs down. Neither direction is the better one to have; the second just feels like it is.

Step 5 — Convert your top verdict into a question (10 minutes)

Take the entry you care about most. Read Field 11. Write the same thing as a question a clinician could answer — one whose answer is in the room.

If you cannot, the verdict was probably a feeling, and that is a finding about the entry rather than about you.

Step 6 — Close it out

Three things, in this order, and then you are done:

1. Date the document. Today, at the top.
2. Save a copy of this version. Not a backup — a version. Name it with the date and leave it alone. This is the first entry in the history that §40.4 and §40.5 depend on, and it is worthless unless it is frozen.
3. Write one sentence at the very top, above everything, saying what you would need to learn to change your mind about the entry you care about most.

That last sentence is the whole book compressed into one line, and putting it at the top is deliberate. It is the first thing you will read every time you open the file, including — especially — on the day you open it because something exciting happened.

Conclusion

You have a document.

It covers five to ten compounds you actually care about. Each entry has twelve fields; each field catches a specific error; each rating names a population, an endpoint, and a date; each entry says what would change it. You have checked it against itself for a single standard, and you have written down the direction in which you are most likely to be wrong. It is dated, it is versioned, and you know how to keep it alive: review on events, upgrade only on the named readout, and delete what you have stopped caring about.

That document is worth something on its own. It is not the main thing you built.

The main thing is that you can now do this to a claim you have never seen. That is what §40.8 was for, and it is why the twelve fields are organized by question rather than by compound. When somebody tells you — about a supplement, a device, a diet, a procedure, a compound with a laboratory code and a confident website — that it does something, you will find yourself asking what it actually is, what has been measured, in whom, against what, and how you would know if it were false. You will notice a mechanism standing in for an outcome, an origin story doing work it cannot do, and the gap between what a thing is approved for and what it is sold as — because you will be looking for them.

And you will be able to say, precisely, which kind of not knowing you are in — the kind where you have not looked, the kind where nobody has, and the rare, valuable kind where somebody looked carefully and the answer was no.

This chapter closes the method. It does not close the book.

Four chapters remain, and they are the field test. Chapters 41 through 44 take everything you have built and aim it at claims about culture, markets, populations, and society — claims with no trial arms, no endpoints, and no placebo, made with enormous confidence, and made about you rather than about a molecule.

You have spent forty chapters learning to evaluate claims under the best available conditions. Now find out what the method is worth when the conditions are bad.

Turn the page.

Key Terms

Evidence dossier — a personal, maintained reference on a small set of compounds, organized by twelve standing questions rather than by compound, and built to be updated when the evidence changes.

Internal audit — checking a dossier against itself for consistency of standard, as distinct from checking it against an external source for correctness of conclusions.

Same-evidence test — the audit check that finds two entries with comparable evidence bases and different ratings, and requires the difference to be justified by design, population, or endpoint.

Unpopulated rating — a Field 6 row giving a symbol without naming the population and endpoint it applies to. It rates a molecule rather than a claim, which rating rule 1 forbids, and it is the most common defect in reader dossiers.

Falsifiability — the property of a conclusion that specifies in advance what observation would overturn it. Field 12 is where it lives; an entry without it is a belief rather than a conclusion.

Rating drift — the tendency of ratings to move upward over time, because news reaching you about a compound you follow has been selected for being worth publicizing. Chapter 6's testimonial dynamic applied to your own attention.

Version history — a series of dated snapshots of a dossier. Its value is not the old content but the trajectory: it reveals the direction of your past errors, which a snapshot cannot.

Standard of evidence — the threshold you require before assigning a given rating. A standard must be the same across entries to be a standard at all.

Well-specified open question — a statement of what is not known, precise enough to name the population, endpoint, comparator, and duration that would resolve it. The honest final position for most compounds in this book, and a result rather than a failure.

Evidence absent vs. evidence present and negative — the two kinds of ❌. The first means nobody ran the trial; the second means it was run properly and the answer was no. The second is a much stronger state of knowledge, and the two are routinely confused in the direction that flatters the untested compound.

Transferable judgment — the reflex that survives after the content is forgotten: asking of any claim for whom, measured how, compared to what, and how would we know if it were false.

Spaced Review

1. (Chapter 5, applied.) A friend describes a study as "a randomized trial that showed a real benefit." Name the four things you would need to know before that sentence could support a rating at all, and say which single one most often turns out to be the problem.

2. (Chapter 6, applied.) You wrote a Field 12 entry two years ago and have since read three encouraging things about the compound: a conference abstract, a mechanism paper, and a friend's enthusiastic report. Your confidence has risen noticeably. Using §40.6, explain why that rise is predicted by the structure of what reaches you regardless of whether the evidence changed — and state the one-minute procedure that resolves the question.

3. (Chapter 37, applied.) Your ratings disagree with the book's on five of your eight compounds, and all five of your ratings are higher. Separately, a reader you know disagrees on five and all five of theirs are lower. What have you each discovered about yourselves, and why is it a mistake to describe one of you as the more careful reader?

4. (This chapter.) Two people hold identical dossiers — same fields, same evidence, same ratings, same dates — and reach opposite decisions about the same compound. Explain why at least one of them is not necessarily reasoning badly, and identify what the dossier has succeeded at doing even though it did not produce agreement.

5. (The final exam.) You encounter a claim that appears nowhere in this book: a wearable device marketed to consumers with the claim that it improves recovery and sleep quality. It is not a drug. It is not a peptide. Nothing in Part III applies.

Walk it through all twelve fields. For each, state what you would need to find out and what the field's specific failure mode would look like for this claim. Then write the one row of Field 6 you could defend today, with a population, an endpoint, a date, and — if ❌ — which kind. Finally, write Field 12 to the standard set in §40.3: specific enough that a reasonable person could not disagree about whether it had been satisfied.

If you can do that, the book worked. Chapters 41 through 44 are where you find out whether it works on claims that are harder than this one.