Chapter 37: The Peptide Evidence Table — Every Rating in One Place

"It is the mark of an educated person to look for precision in each class of things just so far as the nature of the subject admits." — Aristotle, Nicomachean Ethics, Book I

Overview

This is the chapter you were promised in Chapter 5, and it is the only chapter in the book that issues no new judgments at all.

Every evidence rating this book has made is collected here in one place: 154 of them, spread across 140 molecule-and-indication claims and 14 claim forms, drawn from Chapters 2 through 44. Nothing has been added. Nothing has been softened. Where a chapter split a rating across two versions of the same claim, the split is preserved rather than rounded to whichever half reads better. Where the rating system declined to rate a claim at all, the refusal is recorded as a refusal.

There is a specific reason to gather them. Across thirty-odd chapters the ratings arrive one at a time, embedded in the argument that produced them, and one at a time is exactly how they cannot be compared. You cannot see, reading Chapter 17, that BPC-157's ❌ and Chapter 28's nesiritide ❌ describe two completely different epistemic situations. You cannot see, reading Chapter 30, that the ⚠️ on a cosmetic peptide and the ⚠️ on a myostatin inhibitor in muscular dystrophy sit at opposite ends of what "preliminary" can mean. Assembled, those contrasts become visible in a way that no individual chapter could make them, and some of them are the most useful things in the book.

But a warning belongs at the front, not the back. This table is a snapshot, dated 2026, and it will age. Some entries will move. A few will move in directions that would surprise the person who wrote them. That is not a defect in the table; it is what an honest, date-stamped, falsifiable rating looks like as time passes. A rating that could never change was never a rating — it was a position.

Which leads to the thing this chapter most wants you to understand about itself. Chapter 5 matters more than Chapter 37. Chapter 5 taught you to rate a claim: to find the population, find the endpoint, ask what evidence exists, ask what would change your mind. This chapter is a list of answers that method produced on one particular day. A reader who memorizes the list and cannot rate a new claim has taken the wrong thing away, and will be helpless the first time somebody mentions a compound that is not in it — which will happen, because new compounds arrive faster than reference tables do.

Use this chapter the way you would use a map: to orient, to check yourself, and to notice terrain you had not seen. Do not use it as a substitute for being able to read the ground.

In this chapter, you will learn to:

• State what a rating attaches to, and explain why "is this peptide good?" is a malformed question
• Tell an ❌ that means nobody ran the trial from an ❌ that means the trial ran and answered no
• Find any compound or claim form in the book and report its rating, its population, and its chapter
• Explain the four NOT RATED entries and why adding a fifth tier would be the wrong repair
• Read a split rating — ⚠️/❌, ✅/⚠️, ⚠️/🔬 — as a precise finding rather than a hedge
• Explain what the distribution of 154 ratings does and does not show about peptides
• Predict which ratings are most fragile and which are most durable, and say why
• Audit your own evidence dossier against the table and name the direction of your errors

Learning Paths

All five paths use this chapter, but they use it differently. It is a reference; nobody is expected to read 154 rows in sequence.

💊 GLP-1 — §37.5's Metabolic and weight block is your section, and §37.7 is the single most important page: semaglutide alone carries eight distinct claims across four tiers, which is why the question "does it work?" keeps producing arguments. 🏋️ Performance — go to Growth, repair, and performance. It is the densest ❌ block in the book, and §37.3 explains why almost all of those ❌s are the weaker kind rather than the stronger. 🔬 Science — read §37.2, §37.3, and §37.8 in full; the distribution analysis is the chapter's real argument, and it is an argument about research funding, not about chemistry. 💄 Cosmetic — Cosmetic and topical contains three of the book's ten split ratings, which is not an accident. §37.2's discussion of splits explains what the marketing version of a claim costs. 🏥 Clinical — The approved pharmacopeia, Oncology, and Cardiovascular are review; the useful new material is §37.6, because claim forms are what patients actually bring into the room.

37.1 What this chapter is, and is not

Let me be exact about the object you are holding.

What it is. A reference table reproducing every evidence rating issued in Chapters 2 through 44, organized by therapeutic area and by claim form, with the source chapter given for each so you can go back and read the reasoning. It is complete: if a rating exists in this book, it is here. It is unaltered: the ratings are reproduced as their chapters issued them, including the ones I would write differently today and the ones that resist compression into a single glyph.

And it issues nothing of its own. Not one of the 154 ratings below was decided here. Every one was decided in the chapter that owned the claim, argued there against the evidence available there, and reproduced here unchanged. This chapter and Chapter 40 are the only two in the book that contribute no ratings to the count, and that is a constraint rather than a coincidence: a summary that quietly revised what it summarized would be worse than useless, because you would have no way of knowing which rows had been touched. It is also the same discipline §37.10 is about to ask of you. When you compare your dossier against this table, the comparison is only worth making if neither column was written while looking at the other.

What it is not. It is not a buying guide, a decision aid, or a substitute for a clinician. It contains no doses, no protocols, no sources, and no recommendations, and that is a design constraint rather than an omission — the same constraint that governs the other forty-three chapters. It is also not an argument. The arguments are in the chapters. This is the ledger.

And three things about the ledger deserve saying plainly at the outset.

It is dated. Everything here is assessed as of 2026. Ratings in this book are date-stamped precisely so that they can go stale visibly rather than invisibly. A reader in 2031 holding this page should treat every row as a claim about what was known in 2026, not as a claim about what is known now. Some rows will have held perfectly. Others will read like a weather forecast from last week.

It will age unevenly. This is the part people miss. The table will not decay at a uniform rate; different kinds of entries have different half-lives, and §37.9 works out which is which. Broadly: a ✅ resting on hard clinical outcomes in a large population is close to permanent, a ⚠️ resting on one phase 2 result is close to disposable, and an ❌ that exists because nobody ran a trial can flip in either direction the moment somebody does.

It is less important than Chapter 5. I want to say this as directly as it can be said, because it is the difference between a reader this book helped and a reader it merely entertained.

The value of Chapter 5 is a transferable skill. The value of Chapter 37 is a list. Lists expire; skills do not. If you can rate a claim — identify the population, identify the endpoint, ask what kind of evidence would settle it, ask what evidence exists, ask what would change your mind — then you can handle a compound that was invented after this book went to press, and there will be many. If you cannot, then the first unfamiliar name you encounter puts you exactly where you were before you started reading: dependent on whoever is talking.

So here is the honest use case for this chapter. Read your own dossier first, form your own rating, and only then look here. Used that way, the table is a calibration instrument, and §37.10 turns it into one. Used the other way around, it is a crutch that gets shorter every year.

📋 Your Evidence Dossier — how to approach this chapter

Before you read the master table, do one thing: open your dossier and finish rating your own peptides. All of them. Write the tier and one sentence of reasoning for each claim you have been tracking since Chapter 1.

If you look at the table first, you will not be able to un-see it, and you will lose the single most informative measurement available to you in this entire book — the gap between your judgment and the book's, and which direction that gap runs. That measurement is worth more than any individual row here. §37.10 shows you how to take it.

This is the last time the dossier asks you for something uncomfortable. It is also the time it matters most.

37.2 The rating system, restated

Chapter 5 built the system. Here it is in compressed form, so that this chapter stands alone as a reference.

Four tiers, and two additional states that the table needs and that Chapter 5 introduced more briefly: split ratings and NOT RATED. Both are discussed below, and neither is a fifth tier.

Six rules, compressed

1. A rating attaches to a claim, never to a molecule. A claim names a population and an endpoint. "Semaglutide" is not a claim. "Semaglutide reduces major adverse cardiovascular events in adults with established cardiovascular disease and overweight or obesity, without diabetes" is a claim, and it is the only kind of thing that can carry a rating. This is why the master table's second column is titled Claim (population and endpoint) and why it is the widest column on the page. Anyone who gives you a molecule's overall rating has compressed away the information you needed.

2. ❌ describes the evidence, not the molecule. It is a statement about what has been shown, not a prophecy about what will be. Several ❌ compounds in this table may be ⚠️ or ✅ in a decade. Several will be shown not to work. Both futures are live for most of them, and §37.3 is entirely about telling those two situations apart before the fact.

3. Never upgrade a rating with mechanism. A plausible story about how something would work is not evidence that it does. This is the single most common failure in the peptide literature, in peptide marketing, and — in my experience — in scientists talking about their own compounds. A beautiful mechanism earns you a hypothesis and a grant application. It does not earn you a tier.

4. Never downgrade a rating with distaste. The reverse error is rarer and just as corrosive. Compounds sold badly, marketed with nonsense, or popular with people you find irritating do not thereby acquire worse evidence. Botulinum toxin arrives wrapped in more marketing than almost anything in this book and carries two ✅s, because the trials are there. MK-677 is sold by exactly the kind of vendor this book spends Chapter 19 warning you about, and it carries a ⚠️ rather than an ❌, because real randomized human data exists. Both facts are inconvenient for a tidy narrative, and both are in the table.

5. Every rating is date-stamped and falsifiable. Each source callout names what would change it. That is not a rhetorical flourish; it is the property that makes a rating a scientific statement rather than an opinion. When you disagree with a row in this table, the productive form of the disagreement is not "that's too harsh" but "here is the evidence that moves it, and here is the tier it moves to."

6. One molecule, many ratings. The direct consequence of rule 1. Semaglutide occupies ten rows in this table, resolving to eight distinct claims. BPC-157 carries three. Desmopressin carries three, across two chapters — twice for central diabetes insipidus and once for bedwetting in children, which is one hormone doing one physiological job in two clinical presentations that look nothing alike from the outside. Growth hormone carries two that point in opposite directions. This is not inconsistency. It is precision.

On split ratings

Ten entries in this table carry two glyphs separated by a slash: ⚠️/❌, ✅/⚠️, ⚠️/🔬. Their authors wrote them as "⚠️ → ❌" or "❌ to ⚠️" or "⚠️ for the modest version; ❌ for the strong version," and I have preserved them rather than collapsing them, because in every case the split is the finding.

There are three things a split can mean, and the source chapter always tells you which.

Different versions of the same claim. This is the commonest and the most useful. Topical GHK-Cu is ⚠️ for modest, instrument-measured effects on skin appearance and ❌ for the marketing version — "rebuilds collagen," "reverses skin aging." Same molecule, same population, same route; two claims of very different strength, and the evidence supports exactly one of them. A single glyph would have to choose, and choosing would either flatter the product or libel the science.

Different sub-populations or endpoints. Insulin analogs are ✅ for nocturnal and severe hypoglycemia in type 1 diabetes and ⚠️ for broader claims of superiority in type 2 diabetes. The question "are analogs better?" has two answers because it was two questions.

A trajectory. Calcitonin and intranasal oxytocin for autism were both ⚠️ and became ❌ as larger studies arrived. Recording the arrow rather than the endpoint tells you something the endpoint alone cannot: that this rating has already moved once, under the pressure of better data, in the direction better data usually pushes.

A split is not indecision. It is what happens when you insist that a rating attach to a claim and then discover that two claims were wearing the same name.

📊 Evidence Rating — the format, and why every row here is an abbreviation

Every rating in this book was issued in a four-line callout: Claim, Rating, Why, and What would change it. Only the first two survive into the master table, for the obvious reason that 154 four-line callouts is not a table, it is Appendix A.

That compression costs something real, and you should know what. The Why line is where the quality of the evidence lives — whether a ✅ rests on mortality data or on a surrogate, whether an ❌ rests on failed trials or on no trials. The What would change it line is where the falsifiability lives, and it is the line that tells you how to update the row yourself when news arrives.

So the table's real function is as an index, not as a verdict. Every row carries a chapter number for exactly this reason. If a row matters to you — because it is a compound you use, or one somebody is selling you, or one you disagree with — the row is not the answer. It is the pointer to where the answer is argued.

This is the same discipline the book has asked of you since Chapter 5, now turned on the book itself: do not accept a compressed conclusion when the full reasoning is three pages away.

37.3 The two kinds of ❌ — read this before the table

Fifty entries in this table are ❌, and they are not the same thing. Sorting them is the single most useful analytical move available in this chapter, and it is the one readers most reliably get backwards.

TWO ❌ SYMBOLS, TWO OPPOSITE STATES OF KNOWLEDGE

  ❌  EVIDENCE ABSENT                    ❌  EVIDENCE PRESENT AND NEGATIVE
  ─────────────────────────────         ──────────────────────────────────
  Adequate human trials were            Adequate human trials were run.
  never run.                            They answered no.

  The claim is unsupported              The claim is unsupported
  AND UNTESTED.                         AND TESTED.

  We do not know.                       We know.

  Could move to ⚠️ or ✅ tomorrow        Very unlikely to move. Somebody
  if somebody funds a trial.            already paid to find out.

  Example: BPC-157 for tendon           Examples: NK1 receptor antagonists
  and soft-tissue healing (Ch 5,        for major depressive disorder
  Ch 17). No completed, peer-           (Ch 22). Nesiritide for death or
  reviewed, randomized controlled       rehospitalization in acute
  human trial exists.                   decompensated heart failure (Ch 28).

The diagram states the distinction; now here is why it matters more than it looks like it should.

The second state is a far stronger state of knowledge, and readers routinely treat it as the weaker one. The intuition runs the wrong way. "The trials failed" sounds like a story still in progress — maybe the dose was wrong, maybe the population was wrong, maybe the next one works. "There are no trials" sounds like an empty space where anything might be true, and in the peptide market that emptiness is sold as promise: nobody has proven it doesn't work.

Invert that. A compound with no trials is a compound about which we know essentially nothing in humans. A compound with failed trials is a compound about which we know something specific and expensive: that a serious, funded, adequately powered attempt was made to demonstrate the effect and the effect did not appear.

Look at what Chapter 22 records for the NK1 receptor antagonists. Multiple adequately powered randomized controlled trials, across several structurally distinct compounds and several sponsors, failed to show benefit over placebo in major depressive disorder — with human PET data confirming that the drugs were occupying the target receptor at the doses tested. That is not an absence of knowledge. That is one of the most informative negative results in modern neuropharmacology. It rules out an entire mechanistic hypothesis, and it does so with the receptor-occupancy data that forecloses the usual escape hatch of "the drug never reached the target."

Now look at what Chapter 5 and Chapter 17 record for BPC-157: no completed, peer-reviewed, randomized controlled human trial in the published literature. The ❌ is fully earned — the confident claims made for the compound are not supported — but it is an ❌ about a blank page. The honest sentence is we do not know, and the people selling it do not know either. That is a different sentence from we looked, and it isn't there.

Both facts are usable, and they are usable in opposite directions. If you are trying to predict which rows in this table will move, evidence-absent ❌s are the volatile ones and evidence-present-and-negative ❌s are close to settled. If you are trying to decide how much weight to put on the current state of knowledge, it is exactly reversed.

There is also a third case, rarer and worth noticing so it does not confuse you when you meet it in the table. Chapter 7 rates native GLP-1, administered as such, is a viable treatment for type 2 diabetes or obesity as ❌ — and the reason is not that the evidence is absent, and not that the trials failed. Continuous infusion of native GLP-1 in humans does what you would expect it to do. The claim fails because native GLP-1 survives in circulation for a couple of minutes, which makes "administered as such" not a viable treatment regardless of what the receptor does. That is an ❌ earned on pharmacokinetics rather than on efficacy, and it is a reminder that the rating attaches to the whole claim — including the parts of the claim that are not about whether the molecule works.

In the master table, an ❌ marked (N) is an ❌ of the second kind, in the cases where the source chapter made it unambiguous. Unmarked ❌s should be read as the first kind or as mixed, and the source chapter will tell you which.

⚠️ Hype Check — "there's no evidence either way, so it might work"

The claim, in its usual form:

"Sure, there aren't big trials yet. But there's no evidence against it either. The animal data looks great and thousands of people report benefit. Absence of evidence isn't evidence of absence."

What's true in it. The Latin tag is real and it is correctly applied in some settings. If nobody has looked, a null result has not been produced, and the claim has not been refuted. A compound with no human trials genuinely might work. Several compounds in this table probably do something.

Where it fails. In three places.

First, it treats "might work" as though it were a finding. It is the starting position — the state every compound is in before anyone investigates. Arriving at it is not progress, and presenting it as a result is a rhetorical move, not an evidentiary one.

Second, absence of evidence is weak evidence of absence when a search would have been likely to find something. That is not a subtle statistical point; it is why you conclude there is no elephant in your kitchen without a randomized trial. For compounds that have been circulating for twenty years, in a market with real money in it, the persistent absence of a single completed randomized human trial is itself informative — usually about the absence of a sponsor willing to risk finding out, which Chapter 38 §38.10 examines in detail.

Third, and most important here: the argument is deployed almost exclusively for compounds where the evidence is absent, and almost never for compounds where the evidence is present and negative. That asymmetry is the tell. Nobody markets nesiritide with "absence of evidence isn't evidence of absence," because the evidence is not absent. The slogan is available precisely when nothing has been looked for.

Verdict: true as a logical point, misleading as an argument, and useful mainly as a signal that you have found an evidence-absent ❌ rather than a reason to upgrade it.

37.4 NOT RATED — the four claims the system refuses

Four claims in this book carry no tier at all. They are recorded as NOT RATED, and they are the most interesting entries in the table.

Notice where they are. All four are in Part VIII — the part that turns the book's apparatus on society rather than on molecules. That clustering is not a coincidence and it is not a failure. It is what a rating system behaving correctly looks like. A method that has a domain will hit the edge of that domain, and a good method says so at the edge instead of producing output anyway. Parts I through VII rate 132 claims without ever needing to decline one, because those claims are of the form X does Y in population Z and evidence settles that form. Part VIII asks different questions, and four of them turn out not to be that form at all.

The fourth entry is the one worth sitting with, because it is where the system is turned on itself.

Read the claim again slowly: society should treat obesity primarily as a medical condition to be pharmacologically managed, rather than primarily as a product of a food environment to be structurally changed.

The claim has real empirical components, and this book rates several of them elsewhere. Whether GLP-1 receptor agonists produce substantial weight loss: ✅, repeatedly. Whether they reduce cardiovascular events in a defined population: ✅. Whether widespread use will change what the packaged food industry sells: 🔬. Those are all X does Y in population Z, and evidence settles them.

But the load-bearing word in the claim is "primarily." It is not an empirical term. It encodes a judgment about how a society should allocate attention, money, political capital, and moral seriousness between two approaches that are not mutually exclusive — you can do both, most countries are doing some of both, and the trade-off between them is not a quantity anyone measures. No trial has "primarily" as an endpoint. No dataset has a column for it. Someone who tells you the evidence settles this question has smuggled a value judgment inside an empirical-sounding sentence, and the smuggling is invisible unless you are looking for it.

This is why NOT RATED has to be a named category rather than a quiet omission. A rating system that always returns a tier is not a rating system; it is a machine that converts any sentence into a verdict, and the danger of building fluency with such a machine is that you start feeding it sentences it was never competent to judge. The four entries above are the inoculation. They are the places where the apparatus you have spent thirty-six chapters learning to use stops and says: this one is not mine.

And a refusal to rate is not a rating. There is no fifth tier. NOT RATED does not mean "unrated for now, like 🔬" and it does not mean "so weak it isn't worth a glyph, like ❌." 🔬 says the answer is coming and we are waiting for it properly. NOT RATED says this question does not have the kind of answer this instrument produces. Two of the four are values questions, one is a prediction about collective behavior, and one is an empirical question whose direction — not merely whose magnitude — is unforecastable from what is known.

If you find yourself wanting to add a fifth tier to accommodate them, notice what that impulse is: the desire for the machine to always return something. Resist it. The refusal is the output, and it is a more honest output than anything a fifth tier would produce.

🔍 Check Your Understanding

1. What is the difference between 🔬 and NOT RATED? Give one claim that belongs in each, and say what would move the 🔬 claim to a tier while leaving the NOT RATED claim exactly where it is.
2. All four NOT RATED entries are in Part VIII. Why should that reassure you about the rating system rather than worry you about Part VIII?
3. In the Chapter 44 claim, identify the single word that makes the sentence unrateable. Then rewrite the sentence, keeping its subject matter, so that it is rateable — and say what tier your rewritten version would probably get.

37.5 The master table

Here it is: all 140 molecule-and-indication ratings, organized by therapeutic area.

How to read the columns. Compound lifts the molecule, class, or product out of the claim sentence so the table can be scanned by name. Claim (population and endpoint) preserves what the source chapter actually rated — read it carefully, because the population and the endpoint are doing all the work, and two rows with the same compound and different populations will have different ratings. Rating gives the tier; a slash means a split (§37.2), and (N) marks an ❌ where adequate trials were run and answered no rather than an ❌ where the trials were never run (§37.3). Ch is where the reasoning lives. The rating is the summary; the chapter is the argument.

What the table cannot show you. Effect size, evidence quality within a tier, cost, availability, and — above all — whether any of this applies to you. Two ✅ rows can differ by an order of magnitude in how much good the treatment does. Nothing here is a recommendation, and nothing here is a substitute for a conversation with someone who knows your history.

Metabolic and weight

The largest block in the table and the one most readers came for. Twenty-eight ratings, and the notable feature is the tier distribution: fourteen ✅, six ⚠️, four ❌, three 🔬, one split. This is the most heavily studied area in the book by a wide margin, and it shows — not because these molecules are better, but because a very large commercial incentive has been pointed at them for fifteen years and trials followed the money. Hold that thought until §37.8.

Three things the columns show that a single glance does not. First, semaglutide appears nine times in this block alone, at three different tiers, which is §37.7's whole subject. Second, the four ❌s here are unlike the ❌s in the next block: two of them (native GLP-1, compounded semaglutide) are not about whether a molecule works at all, and one (leptin in common obesity) is marked (N) because the trials were run and the answer was no — circulating leptin is already elevated in common obesity, and adding more does not fix a resistance problem.

Third, and most usefully, this block contains three claims that two different chapters rated independently, and all three pairs agree. Semaglutide's weight-loss claim is rated in Chapter 5 and again in Chapter 8; its cardiovascular-event claim is rated in Chapter 5 and again in Chapter 8; retatrutide's weight-loss claim is rated in Chapter 9 and again in Chapter 36. A reference table that merged those duplicates would look tidier and would destroy the only self-check the book contains. If two chapters, written to different purposes and reasoning from the same literature, land on the same tier, that is weak evidence the tier is not an artifact of how one chapter framed the question. Leave the duplicates in and run the check yourself; it costs ten seconds and it is the only place the book audits itself in public.

Growth, repair, and performance

Sixteen ratings, nine of them ❌ and two more split with an ❌ arm. This is the densest block of negative ratings in the book, and it is also — read alongside §37.3 — the block where almost every ❌ is the weaker kind. These are largely compounds nobody has tested, not compounds that failed.

The column that does the most work here is population. Growth hormone appears twice, at ✅ and ❌, and the difference between the rows is the phrase with documented growth hormone deficiency. Same molecule, same endpoint family, opposite ratings, and the entire distinction lives in five words in the population field. The same structure governs mecasermin (✅ in children with severe primary IGF-1 deficiency) against IGF-1 LR3 (❌ in healthy adults), and myostatin inhibitors (⚠️ in muscular dystrophy) against follistatin-344 (❌ in healthy trained adults). In four separate cases, this block rates the same pathway ✅ or ⚠️ in people with a documented deficiency or disease and ❌ in healthy people who want more of something they already have enough of. That pattern is the most transferable finding in Part III, and you can see it here in a way you cannot see it chapter by chapter.

The ⚠️ on MK-677 deserves a second look for the opposite reason. It sits in a block of ❌s, sold through the same channels as the compounds around it, and it is rated higher because randomized human trials exist showing it durably raises growth hormone and IGF-1 and increases fat-free mass. Rule 4 from §37.2 in action: the company a compound keeps is not evidence about the compound.

Neuropeptides and behavior

Twenty-two ratings, and the most balanced block in the table: nine ✅, three ⚠️, eight ❌, one 🔬, one split. The neuropeptide field contains both some of the best-evidenced drugs in this book and some of its most durable popular myths, frequently attached to the same molecule.

The melanocortin family's cosmetic claim — Melanotan II for tanning — is filed in Cosmetic and topical below, where readers will look for it.

Two column-level observations. First, oxytocin occupies four rows spanning ✅ to ❌, and the ✅ is the oldest and least glamorous of them: intravenous administration in an obstetric setting for uterine contraction, an indication approved in every major jurisdiction and in continuous clinical use for decades. The famous claim — the love hormone — is the ❌. The gap between which oxytocin claim is best evidenced and which is best known is probably the largest such gap in the book. Second, the Semax/Selank ⚠️ and the Cerebrolysin ⚠️ are worth reading as a pair, because they are ⚠️ for opposite reasons: one because a substantial literature exists in a language and publication ecosystem most readers of this book cannot assess, the other because an accessible literature exists and the studies disagree with each other. Both are "preliminary." Neither is preliminary in the way a phase 2 result is preliminary. The tier is the same and the situations are not, which is exactly why every row carries a chapter number.

Infectious disease and immunity

Ten ratings, evenly spread: three ✅, three ⚠️, two ❌, two 🔬. This block contains the book's clearest example of a mechanism that is real, important, and nonetheless does not license the claim usually built on it.

Look at the two antimicrobial-peptide class rows together with the two approved-drug rows. Colistin and daptomycin are ✅ — peptide antibiotics have been curing people for decades. The claim that antimicrobial peptides cannot generate resistance is ❌ and marked (N), because multiple resistance mechanisms are well characterized and at least one, plasmid-borne mcr, is transferable between organisms. And the claim that this class will deliver a new generation of broad-spectrum systemic antibiotics is 🔬 — a real prospect with specific, named obstacles, proceeding properly. Three different tiers on one molecular class, distinguished entirely by what the claim asserts. That is rule 1 doing its job in the space of a four-row block.

Oncology

Nine ratings, and the block with the highest proportion of ✅ anywhere in the book except the approved pharmacopeia: six ✅, one 🔬, two splits. Peptides in oncology are almost entirely a story of receptor-targeting done well, and the evidence base reflects an area where trials are the norm rather than the exception.

The instructive contrast is between the two vaccine rows and the six approved-therapy rows. The approved therapies carry populations of extraordinary specificity — read the PSMA row's population field, which specifies the imaging method used to select patients and the two prior treatments they must have received. That specificity is not bureaucratic fussiness; it is the population in which the trial was run, and the ✅ does not extend one inch past it. Meanwhile the therapeutic-cancer-vaccine class carries the book's clearest example of a downgrade earned by trials that ran and failed: a long record of large randomized studies missing primary endpoints across multiple tumor types, antigens, and platforms — often while demonstrably inducing the intended immune response. That last clause is the most important half-sentence in this block. The mechanism worked and the patients did not benefit, which is the exact scenario rule 3 exists to guard against.

Cardiovascular

Six ratings — four ✅ and two ❌, both ❌s of the (N) kind. This is the highest concentration of evidence-present-and-negative anywhere in the table, which makes it the best short course in §37.3 available in the book. The block follows the natriuretic peptide family rather than the specialty, which is why its last row is about children's growth rather than about hearts.

Read the first three rows in order and you have the whole argument of Chapter 16 in miniature. The biomarker is genuinely useful for the thing it was validated for — diagnosis, and especially exclusion. Treating the biomarker as a target to be driven downward did not improve outcomes, and the trial testing that strategy was stopped for futility. And infusing the peptide itself, approved on hemodynamic surrogates and short-term symptom measures, produced no meaningful effect on death or rehospitalization when a randomized placebo-controlled outcome trial of roughly 7,100 patients finally asked. A molecule can be a good measurement, a bad target, and an ineffective drug, all at once. Nothing about the mechanism predicted which would be which. Only the trials did.

The last row is the family's one approved indication outside cardiology, and it is worth a moment for what its endpoint column says. The claim is annualized growth velocity over one year of treatment — how much faster a child grows in twelve months. That is precisely stated, it was measured, and the ✅ covers it. It is not a claim about final adult height, about mobility, about pain, or about anything else a family would eventually want to know, and the ✅ does not reach those questions because the claim did not make them. This is rule 1 at its most disciplined: the rating is exactly as wide as the sentence it attaches to, and a reader who widens the sentence has left the evidence behind without noticing. When you meet this row in the wild — and growth-velocity results are reported as though they settled the larger question routinely — the useful reflex is to ask what the trial's clock was and what it measured when the clock stopped.

The approved pharmacopeia

Seven ratings, six ✅ and one split. This block is a reminder that the peptide pharmacopeia is older and duller and more successful than the peptide market, and that most of the peptides doing the most good for the most people are ones nobody argues about on the internet.

The endpoint column is what distinguishes this block. Fracture risk. Urine output and osmolality. Wet nights. Blood glucose in an emergency. Bowel frequency. Relief of an acute swelling attack. These are hard, directly measured, patient-relevant outcomes, and four of the six ✅s rest on endpoints so unambiguous that the effect is visible in one patient within hours, or across a single night. There is no surrogate anywhere in this block, which is exactly why §37.9 will call these among the most durable rows in the table.

Desmopressin is worth pausing on, because it now appears three times across two chapters — twice for central diabetes insipidus and once for bedwetting in children. One hormone, one physiological action (concentrating urine), three separately evidenced claims, and two clinical presentations that look nothing alike from the outside. If you wanted a single example of why this book refuses to give a molecule an overall rating, that is it. The calcitonin split is the exception that proves the point: its fracture-endpoint evidence came from a single supportive trial with an inconsistent dose–response and substantial attrition, never adequately replicated — an endpoint of the right kind, measured once, unconvincingly.

💊 In the Clinic — what a table like this is actually for in a consultation

Clinicians do not look things up in tiers. They look up indications, populations, interactions, and monitoring, and none of those are in this chapter. So it is worth being precise about the one job this table can do in a clinical conversation, because it is a real job and it is narrow.

It lets a patient arrive with a specific claim instead of a molecule name. The difference between "I want to ask about semaglutide" and "I want to ask about semaglutide for kidney protection, because I have type 2 diabetes and my kidney function is declining" is the difference between a conversation that starts from zero and one that starts from a shared premise. The second version names a population and an endpoint. It is, in the vocabulary of this book, a claim — and claims are things two people can reason about together.

It also lets a patient say the harder sentence. Chapter 39's ✅-rated claim form is that disclosing all substance use to a treating clinician improves care, and the commonest obstacle to that disclosure is not shame but the expectation of a lecture. Arriving with "I have been taking this, I know the evidence for it is thin, and I want to talk about monitoring" changes the shape of the encounter entirely. The table can supply the middle clause.

What it cannot do is settle anything. A row is population-level evidence. Whether a given treatment is right for a given person depends on their history, their other medications, their other conditions, what they are trying to achieve, and what they are willing to tolerate — none of which appears in any column here.

Cosmetic and topical

Eight ratings, and three of the book's ten split ratings live here — the highest concentration anywhere. That is not a coincidence, and it is the single most useful thing this block teaches.

Why do splits cluster here? Because cosmetics is the part of the peptide world where the distance between the modest thing that was measured and the extravagant thing that was claimed is widest, and where both versions of the claim circulate under the same product name. A ⚠️/❌ on GHK-Cu is not fence-sitting. It says: instrument-measured effects on skin appearance, modest, plausible, ⚠️ — and "rebuilds collagen," "reverses skin aging," ❌. The split is the consumer advice, in the only form this book gives consumer advice, which is a statement about evidence.

Note also that the two ✅s in this block belong to an injected neurotoxin, and the ❌ next to them belongs to a topical peptide marketed as its alternative. Chapter 1 predicted that outcome from molecular weight alone, before any evidence was consulted: acetyl hexapeptide-8 is far above the 500-dalton heuristic for passive skin penetration and would need to reach neuromuscular junctions beneath the dermis. Chemistry narrowed the space; evidence closed it.

Veterinary

Four ratings — one ✅ and three ❌ — and the whole block is about a single inferential move.

The Rating column here is less interesting than the Claim column, and specifically than the phrase on the basis of. Two of these rows rate an inference, not a molecule. Veterinary GnRH agonists work, in approved species, for approved indications, supported by controlled studies — hence the ✅. What fails is the extrapolation: different species, different indication, and outcome measures (average daily gain, feed conversion ratio, carcass composition in young food animals) that cannot capture anything a human cares about. The fourth row generalizes the failure into a claim form, which is where this table's other fourteen entries live.

Science, technology, and society

Thirty ratings, covering claims about the field itself — its methods, manufacturing, analytics, economics, and social consequences — rather than about what a molecule does in a body. This is also where all four NOT RATED entries sit.

Methods, manufacturing, and analysis (15).

Society, culture, and access (15).

Two things the columns show here that matter more than any individual row.

In the first table, notice that every ✅ is a technical claim — synthesis chemistry, albumin binding, mass spectrometry, structure prediction — and every ❌ is a quality or economic claim about the consumer market. The methods of peptide science are extremely well established. The assurances offered to consumers of peptide products are not. Those are two different questions and the table separates them cleanly, which is worth remembering next time a vendor cites the sophistication of modern peptide chemistry as though it were a statement about their vial.

In the second table, notice how many rows specify their measurement instrument inside the claim: "measured by validated weight-bias instruments," "measured as shifts in portion size, protein and fiber content, and product mix." Part VIII is the part of the book most exposed to the temptation to opine, and the discipline it adopts in response is to make every rateable social claim name the thing that would be counted. When a claim in that part could not name one — when the operative word was should, or inevitable, or a direction that two opposed mechanisms both plausibly determine — it went into the NOT RATED column instead. That is the boundary of the method, drawn from the inside.

🔍 Check Your Understanding

1. Growth hormone appears twice in the master table with opposite ratings. Quote the phrase in the population field that produces the difference, and name two other rows in the same block that work the same way.
2. Colistin is ✅, "antimicrobial peptides cannot generate resistance" is ❌ (N), and "antimicrobial peptides will provide a new class of broad-spectrum systemic antibiotics" is 🔬. Explain to someone who has not read Chapter 25 how all three can be true simultaneously.
3. Find the two rows in the master table that rate an inference rather than a molecule. What do they have in common with the fourteen entries in §37.6?

37.6 Claim-form ratings

Fourteen ratings in this book attach to a claim form — a sentence shape that recurs across compounds — rather than to any particular molecule. They are filed separately for a practical reason: a reader scanning for tirzepatide should not have to filter past "it isn't on the banned list" to find it.

These are probably the most useful fourteen rows in the chapter, and here is why. A molecule-specific rating helps you exactly once, with exactly that molecule. A claim-form rating helps you every time the sentence shape appears, and these shapes appear constantly — attached to compounds this book covers, compounds it does not, and compounds that do not exist yet. Learning that BPC-157 is ❌ prepares you for one conversation. Learning what "third-party tested" does and does not establish prepares you for every conversation about every product for the rest of your life.

Read the Rating column and notice the eleven ❌s, then notice what unites them. Every one is an inference that skips a step: from a certificate to a vial, from an approval to a benefit-risk judgment about you, from a modification to a clinical effect, from a regulator's silence to a conspiracy, from a prescription to an evidence base. Each is a real thing being used to vouch for a different thing it does not actually vouch for. That is the shape. Once you can see it, the specific words stop mattering.

The three non-❌ rows are worth as much as the eleven. ✅ for "disclosing all substance use to a treating clinician improves care," which is rated ✅ on structural rather than statistical grounds: differential diagnosis, interaction checking, monitoring, and pre-operative assessment each take a medication history as an input and produce a wrong output when that input is incomplete. ⚠️ for "medical supervision reduces the risk of using an unapproved compound," which is the honest tier: supervision genuinely adds baseline and interval monitoring, interaction checking against a full medication list, and a defined response pathway — and it adds no evidence whatsoever about whether the compound works. And ⚠️ for "AI has revolutionized drug discovery," where the structure- prediction advance is real and enormous but the claim conflates producing candidate molecules with producing approved drugs.

Notice, finally, that several claim forms of the same shape are filed in the master table rather than here — "third-party tested" in Chapter 19, "used in animals for years" in Chapter 31, "98% purity" and "pharmaceutical grade" in Chapter 32. They are recorded where the chapter that issued them sits, because those chapters attached them to a specific product category rather than stating them as general forms. Chapter 34 then re-rates two of them as general claim forms, which is why "third-party tested" and the purity claim appear in both tables. That duplication is not an error; it is the same sentence being evaluated twice, once about a product type and once as a form of reasoning, and both times it gets ❌.

⚠️ Hype Check — "third-party tested"

The claim, in its usual form:

"We're not like the sketchy vendors. Every batch is third-party tested and we publish the certificate of analysis. You know exactly what you're getting."

What's true in it. Testing is better than no testing, publishing a certificate is better than hiding one, and a vendor who does neither is worse than one who does both. This is a real, if small, quality signal, and the ❌ is not a claim that testing is worthless.

Where it fails. The phrase specifies no scope. Identity, purity, content, sterility, endotoxin, and residual solvents are separate determinations requiring separate methods, and most programs run two or three of them. A certificate showing HPLC purity says nothing about whether the vial is sterile or endotoxin-free — and per Chapter 19, contamination and endotoxin are where the documented harm in this market has actually occurred.

More fundamentally, a certificate describes a sample the supplier provided, at a moment in the past. It is not a property of the container in your hand. Nothing in the phrase establishes that the tested material and the sold material came from the same batch, or that anything downstream of the test — fill, storage, shipping — preserved what was tested.

And per Chapter 34's separate rating, independent testing cannot convert a gray-market product into a pharmaceutical one, because pharmaceutical equivalence is constituted by documented process control, batch traceability, chain of custody, and validated aseptic processing. Testing samples the output. It does not create the system.

Verdict: ❌ as usually stated. The useful question is not "is it tested?" but "tested for what, by whom, on which batch, and how do I know this vial came from it?" Most sellers cannot answer the second half.

37.7 One molecule, many ratings: semaglutide

If you take one page from this chapter, take this one.

Semaglutide occupies ten rows in the master table, resolving to eight distinct claims across four different tiers. (Two of the claims are rated independently by two chapters each; both pairs agree, which is the self-check described in §37.5's metabolic block.) Every one of them is correct. They do not conflict, they are not a sign that the evidence is confused, and no two of them can be substituted for each other.

And one adjacent row that belongs in the same conversation: GLP-1 receptor agonists reduce alcohol consumption, smoking, or other addictive behaviors — 🔬 (Chapter 10), rated at the class level rather than for semaglutide specifically, which is itself a distinction worth noticing.

So: is semaglutide good?

The question has at least eight answers, and the reason it produces so much argument is that people asking it are usually asking about different rows without knowing it. A cardiologist and a neurologist and someone who bought a vial from a website are all asking "does semaglutide work?" and they are asking about a ✅, a 🔬, and an ❌ respectively. All three will find support for their position. None of them is wrong. The question is the problem.

Watch how the tiers are earned, because the pattern is transferable. The four ✅s each rest on adequately powered randomized trials with a defined population and a prespecified endpoint — including one, the cardiovascular row, resting on a hard outcome rather than a surrogate, which §37.9 will identify as the most durable kind of ✅ in the table. The two ⚠️s rest on real human data that does not yet settle the question — histological improvement in a disease where histology is a surrogate for a long-term outcome, and symptom-and-function improvement in a condition where symptom scores are meaningful but limited. The 🔬 rests on trials that have been run in a field proceeding properly, with the answer genuinely not in.

And the ❌ is not about the molecule at all. It is about whether a compounded preparation is equivalent to the branded product — a category spanning everything from a registered facility using genuine semaglutide base under inspected quality systems to material of unclear origin. The molecule's evidence base is irrelevant to that question, which is precisely why it needs its own row.

🧬 The Molecule — the same eight claims, read from the chemistry

Chapter 1 taught you to read a peptide's structure. Reread the eight rows above with that in hand and something appears.

Semaglutide is a GLP-1 analog with two backbone modifications and a fatty-acid chain attached at a lysine — the substitution at position 8 defeats DPP-4, and the acylation promotes reversible albumin binding, which is the ✅-rated technical claim in §37.5's methods table. Those changes buy one thing: time in circulation. They do not change what the molecule does at the receptor.

Which means every row in the table above is downstream of the same pharmacology — a GLP-1 receptor agonist present at a sustained concentration — landing in eight different physiological contexts. The receptor is expressed in the pancreas, the gut, the kidney, the cardiovascular system, and the brain. The molecule does not know which claim it is being asked about.

This is exactly why mechanism cannot set a rating. The same mechanism is fully present in the Alzheimer's row and in the cardiovascular row. One is 🔬 and one is ✅, and the difference is entirely a matter of which trials have been completed. If mechanism could settle it, both rows would carry the same glyph — and if you had rated them from mechanism in 2016, you would have been wrong about at least one.

37.8 What the distribution shows

Here is the whole table, counted.

DISTRIBUTION OF ALL 154 RATINGS                     (140 molecule/indication + 14 claim form)

  ✅  Strong clinical evidence     54   ██████████████████████████████████
  ❌  Hype outpaces evidence       50   ████████████████████████████████
  ⚠️  Promising but preliminary    26   ████████████████
  🔬  Frontier                     10   ██████
  ⚠️/❌, ✅/⚠️, ⚠️/🔬  split         10   ██████
      NOT RATED                     4   ██
                                  ───
                                  154

Fifty-four against fifty. More claims in this book turn out to be well supported than turn out to be unsupported — and that is the first thing worth saying about the ledger: this is not a debunking book. If you arrived expecting a takedown of the peptide field, it does not deliver one. If you arrived expecting vindication of the peptide market, it does not deliver that either.

Do not lean on the margin. Four ratings is well inside the range that a different but equally defensible set of chapters would have shifted, and nothing in this book's argument depends on ✅ outnumbering ❌ rather than the reverse. The direction is worth stating; the size of the gap is not. What is worth leaning on is the structure underneath, and that structure is very legible. Sort the same 154 ratings by where they came from.

Read down the ✅ column. It concentrates in metabolic and weight (14), neuropeptides (9), oncology (6), and — proportionally the densest of all — the approved pharmacopeia (6 of 7) and cardiovascular (4 of 6). Read down the ❌ column. It concentrates in growth, repair, and performance (9 of 16), in claim forms (11 of 14), in veterinary extrapolation (3 of 4), and in the product-quality entries of the science block.

Now say what those two lists have in common internally.

✅ clusters where a compound is dispensed by prescription, generally after regulatory approval, in a defined population, for an indication somebody paid to study. ❌ clusters where a compound is sold directly to consumers without a prescription, or where the claim is a marketing sentence rather than a trial result.

That is not a fact about peptides. It is a fact about which peptides get studied — which is a fact about who pays for trials.

This deserves to be stated carefully, because it is easy to read it as either more or less conspiratorial than it is. A phase 3 program costs a great deal of money and takes years. Somebody has to fund it, and the only entities that reliably do are companies that can recoup the cost through a period of exclusive sale. That mechanism has produced an extraordinary amount of good — every ✅ in the approved pharmacopeia block exists because of it. It also has a structural blind spot exactly where you would predict: compounds that cannot be protected are compounds nobody funds trials for, and compounds nobody funds trials for accumulate ❌s that mean "untested" rather than "disproven." BPC-157 has no completed randomized human trial not because a trial was run and buried, but because no sponsor has a route to recovering the cost of running one. Chapter 38 §38.10 works through the economics in detail, and it is the single most important section in the book for interpreting this table correctly.

So the headline at 54–50 is real and not very interesting, and the structure under it is the finding. The evidence base for peptides is not distributed according to which compounds are promising. It is distributed according to which compounds are ownable. Both halves of that sentence should change how you read a row.

Two smaller observations from the table that are worth having.

⚠️ is the rarest of the main tiers, at 26 of 154, and that is slightly surprising. You might expect the middle of a scale to be crowded. It is not, and the reason is that "real human data that does not settle the question" is an expensive state to be in — it requires that somebody ran a trial at all. Most claims in the consumer peptide space never reach ⚠️ because they never reach a trial. ⚠️ is not a way station that every compound passes through; it is a distinction earned by having been studied.

Adding splits to their component tiers changes the picture. Eight of the ten splits are ⚠️/❌. Count those arms and the ❌ population grows to fifty-eight claim-versions while ⚠️ grows to thirty-four — because in eight separate cases, a modest claim was supportable and the marketing version of the same claim was not. That is the most common single failure mode in this entire ledger, and it is invisible if you collapse splits into one glyph, which is why §37.2 refuses to.

🩺 Safety and Risk — what a rating never tells you

Every glyph in this chapter answers one question: does the claimed effect occur in the stated population? None of them answers the question people most often think they are answering.

A ✅ is not a safety verdict. It means efficacy is established and the safety profile is characterized — characterized, not benign. Several ✅ compounds here carry serious risks that are managed rather than absent. Ziconotide's ✅ is for a drug delivered into spinal fluid under specialist supervision. Colistin's ✅ is for an agent used when better options have run out. Botulinum toxin is the most potent biological toxin known and carries two ✅s.

An ❌ is not a safety verdict either, in either direction. It does not mean dangerous, and — far more commonly misread — it does not mean harmless. An evidence-absent ❌ means nobody ran the efficacy trial, and the same absence usually applies to systematic safety data. "No documented harms" for a compound nobody has studied is a statement about the study, not about the compound. Chapter 19's rating of "nobody has reported any problems" is ❌ for exactly this reason: no reporting pathway exists for unapproved products, so the absence of reports reflects the absence of collection.

And no rating in this table applies to you specifically. Population-level evidence is the input to a clinical decision, not the decision. A ✅ in a population you do not belong to is not your ✅, and a ⚠️ in a condition you have may still be the best available option for you. Those judgments belong with a clinician who knows your history, your other medications, and what you are trying to achieve. Chapter 39 is entirely about making that conversation useful — and its ✅-rated claim form is disclose everything, which is the one recommendation this book does make.

37.9 How this table will age

Snapshots age. This one will age unevenly, and the pattern is predictable enough to be useful.

Some ❌s will become ⚠️ or ✅ when somebody finally runs the trial. This is the most common direction of movement in a field where the ❌s are dominated by evidence-absence. If a sponsor emerges for one of the compounds in the growth-and-repair block, or if an academic group secures funding for an adequately powered trial, the row moves the day the result publishes — and it can move a long way, because there is nothing anchoring it.

Some ⚠️s will become ❌ when a larger study fails to replicate — and this direction is commoner than readers expect. Chapter 9's discussion of phase 2 optimism is the reason. Effects measured in small, early, often unblinded studies in carefully selected participants regress when the study gets larger, longer, better blinded, and more representative. This is not fraud and it is not usually even error; it is the predictable behavior of a filter that lets encouraging early results through more readily than discouraging ones. Two rows in the current table are already the visible residue of that process: intranasal oxytocin for autism and calcitonin for fracture risk are both recorded as ⚠️/❌ downgrades, and both moved in this direction after larger studies arrived.

So when you are guessing which ⚠️ will move and where, the base rate does not favor promotion. A ⚠️ resting on a single phase 2 result is more likely to shrink than to grow.

✅ on hard outcomes is the most durable thing in the table. A result showing fewer deaths, fewer fractures, fewer cardiovascular events, or fewer hospitalizations in an adequately powered randomized trial is close to permanent. It can be refined — the population can be narrowed, a better comparator can arrive — but it is very rarely reversed, because the endpoint is not a proxy for the thing that matters. It is the thing that matters.

✅ resting on surrogates is the least durable ✅. Chapter 16 is the reference here, and Chapter 28 supplies the object lesson: nesiritide was approved on hemodynamic surrogates and short-term symptom measures, and the outcome trial that eventually ran found no meaningful effect on death or rehospitalization. Bone mineral density, HbA1c, visceral fat on imaging, histological scores, biomarker levels, and body composition are all surrogates. Each is a reasonable proxy. None is a guarantee, and the graveyard of drugs that improved a surrogate and did nothing for patients is large enough to have its own literature. When you see a ✅ in this table, look at the endpoint in the claim column and ask which kind it is.

❌ evidence-absent entries move most easily in either direction, precisely because nothing anchors them. This is the corollary of §37.3 and the most practically useful forecast in this section. If you want to know which rows in this table will look different in 2031, look for the ❌s that exist because nobody ran a trial. They are the volatile ones. Some will be vindicated in part. Some will be demolished. The one thing that will not happen is that they stay interesting while remaining untested — the moment a real trial reports, the row acquires an anchor and stops moving freely.

🔬 will resolve, and that is what 🔬 means. Every frontier entry here is a question somebody is actively working on with methods capable of answering it. Ten rows, and by 2031 most will have become ✅, ⚠️, or ❌. That resolution is the frontier tier working as designed, and a 🔬 that is still 🔬 a decade later should be reread with suspicion — at some point, "too early to say" becomes its own kind of answer.

NOT RATED will not move at all. Three of the four are not the kind of claim that evidence settles, and no volume of new data will make should into an empirical term. The fourth — whether pharmacological obesity treatment reduces population-level weight stigma — could in principle acquire data, since it names its measuring instrument. But its problem is that two well-supported mechanisms push in opposite directions through the same variable, so even good data may only tell us which happened, not which was predictable.

🔬 Read the Study — how to update a row yourself

A new result arrives. Here is the procedure for deciding whether it moves a row in this table, compressed from Chapters 5 and 6.

1. Is the population the same? If a trial in adults with established cardiovascular disease reports a result, it does not move a row whose population is healthy adults. Population mismatch is the most common way a genuine result gets attached to the wrong claim.

2. Is the endpoint the same, and is it hard or surrogate? A result on a surrogate does not move a row whose claim names a clinical outcome. It may create a new ⚠️ row of its own.

3. Was it adequately powered and prespecified? A subgroup finding, a secondary endpoint, or a post-hoc analysis is a hypothesis. It moves a row toward 🔬 or ⚠️ at most, never to ✅.

4. Which direction does it push, and does that match the base rate? A larger trial shrinking an earlier effect is the expected pattern (§37.9). A larger trial enlarging one is unusual and deserves more scrutiny, not less.

5. What did the source chapter say would change it? Every rating in this book names this explicitly in its fourth line. If the new result is the thing the chapter named, the row moves. If it is something else, ask why you wanted it to move.

Run those five and you are doing what this book has been teaching since Chapter 5 — which is the skill that outlives the table.

37.10 Finding your own drift

Now the part that is about you rather than about peptides.

You have a dossier. It has your ratings in it. This chapter has the book's ratings. Put them side by side, entry by entry, and count the disagreements.

But the count is not the point, and this is where almost everyone stops too early. Getting thirty-one of forty right tells you almost nothing useful. What tells you something is the direction of the nine you got wrong.

There are two ways to be systematically wrong, and they are mirror images.

Drift toward generosity. Your ratings are consistently a tier higher than the book's for compounds you wanted to work. The peptide your training partner swears by; the thing you already bought; the mechanism you found elegant; the compound whose story is more interesting than its data. Generous drift usually feels like open-mindedness from the inside. It has a characteristic fingerprint: you find yourself rating ⚠️ where the book says ❌, and your reasoning leans on mechanism or on animal data or on the sheer volume of people reporting benefit.

Drift toward severity. Your ratings are consistently a tier lower than the book's for anything that pattern-matched to marketing. The compound with an aggressive advertising campaign; the one sold by people you find embarrassing; the one that is popular. Severe drift feels like rigor from the inside, and it produces a characteristic error too: rating ⚠️ or ❌ where the book says ✅, on evidence you did not actually examine because the packaging told you what to expect. Rule 4 from §37.2 exists because I have made this error and it took me an embarrassingly long time to notice, since being wrong in the skeptical direction never feels like being wrong.

Neither bias is more respectable than the other. This matters, because one of them has better cultural standing than the other right now, and that standing is not earned. Credulity and reflexive cynicism are the same failure — substituting a prior about the source for an examination of the evidence. They just point different directions, and the cynical version has the advantage of being right more often in a market full of nonsense, which is exactly what makes it hard to detect. Being right for the wrong reason is still not knowing.

And a disagreement is not automatically an error. This is the last thing to say before the dossier exercise, and the most important. If you rated something ⚠️ where the book says ❌, and you can name the specific evidence that moved you, and you can say what would move you back — that is correct work. That is what Chapter 5 trained you to do, and the fact that a book disagrees with you does not make you wrong. Several rows in this table are ones reasonable people rate differently, and a few are ones I expect to be revised.

The error is not disagreeing. The error is disagreeing without being able to say what evidence moved you — and, above all, disagreeing in a consistent direction without having noticed.

📋 Your Evidence Dossier

Assemble every rating, and find your drift.

This is the final analytical pass over the dossier you have been building since Chapter 1. Chapter 40 closes it out; this is where you find out what it taught you. The procedure is mechanical, and it is supposed to be — the discipline is in doing it exactly, not in doing it thoughtfully.

Step 1 — Complete your ratings first

Before you look at anything in §37.5, go through your dossier and make sure every claim you have tracked has your rating on it: a tier and one sentence of reasoning. If you have entries you never rated, rate them now. If you have entries where your Chapter 1 belief was never revisited, revisit it now.

Do not skip this. Everything downstream is worthless if your column is contaminated by the book's.

Step 2 — Build the comparison

For each claim in your dossier, write a line with five fields.

DRIFT AUDIT — one line per claim

  CLAIM            the claim as you stated it, with population and endpoint
  MY RATING        the tier you assigned, and the date you assigned it
  BOOK RATING      the tier in §37.5 or §37.6, and the chapter
  DIRECTION        HIGHER if yours is more favorable, LOWER if less, MATCH if equal
  MY REASON        one sentence: what evidence drove your rating

Two mechanical notes. If your claim does not exactly match a claim in the table — different population, different endpoint — do not force the comparison. Write "no matching claim" and move on; that is itself a finding, and usually it means your claim was less precisely specified than the book's, which is worth knowing. And if the book gives a split rating, compare against the arm that matches the version of the claim you rated.

Step 3 — Sort by direction, not by count

Now count three numbers: how many MATCH, how many HIGHER, how many LOWER.

The first number is the least interesting. Look at the ratio of the second to the third.

READING YOUR OWN NUMBERS

  HIGHER ≈ LOWER          Your errors are noise. This is the good result. It means
                          you are miscalibrated in magnitude but not in direction,
                          which is a fixable problem and a much smaller one.

  HIGHER >> LOWER         Generous drift. You are rating up on compounds you have
                          a stake in. Go back and look at what those compounds have
                          in common — usually it is that you own some, use some,
                          or find the mechanism beautiful.

  LOWER >> HIGHER         Severe drift. You are rating down on presentation. Check
                          whether your LOWER entries cluster on compounds that are
                          heavily marketed, and whether you actually read the
                          evidence for those or inferred it from the advertising.

Those three patterns are the whole diagnostic. A rough guide: if one direction outnumbers the other by more than about two to one across ten or more comparisons, treat it as a real signal rather than noise.

Step 4 — Examine the disagreements you can defend

Go back through your HIGHER and LOWER entries and sort them into two piles.

Pile one: disagreements where you can name the evidence. You rated Semax ⚠️ where the book says ⚠️ but for a different reason; you rated a cosmetic peptide ❌ where the book says ⚠️/❌ because you weighted the penetration argument more heavily. Fine. These are not errors. Write down what would change your mind, and you have done exactly what this book asked.

Pile two: disagreements where your reason sentence is vague. "Seems overhyped." "The mechanism makes sense." "Everyone says it works." "It's obviously marketing." These are the ones that carry your bias, and the size of pile two relative to pile one is a better measure of your progress than your match count.

Step 5 — Write one sentence about yourself

Not about peptides. About you.

One sentence, in your dossier, dated, naming the pattern in your errors and what you think produced it. Something with the shape of: I rated four compounds higher than the book, all four were things I had already spent money on, and my reasons for three of them were mechanism rather than evidence.

That sentence is the deliverable. Everything else in this exercise exists to make it possible to write. Chapter 1 asked you to record what you believed before you knew anything; this asks what your believing is systematically like. Very few people can answer the second question about themselves, and the ones who can are noticeably harder to sell things to.

Conclusion

One hundred and fifty-four ratings. Fifty-four ✅, fifty ❌, twenty-six ⚠️, ten 🔬, ten splits, and four claims the system declined to rate at all.

Slightly more of this field is well supported than unsupported, and that near-balance — not the margin, which is too small to lean on, but the fact that the two ends of the scale are anywhere near each other — is the most honest summary of the peptide field available: it contains extraordinary medicines and extraordinary nonsense, frequently overlapping, occasionally in the same molecule. Insulin analogs and BPC-157 are both peptides. Botulinum toxin and Argireline are both sold for facial lines. Tesamorelin and CJC-1295 are both growth-hormone-axis compounds, and one has an imaging-based randomized trial behind it while the other has a forum thread.

Underneath the balance is a structure, and the structure is the finding. The ✅s cluster where trials were funded. The ❌s cluster where they were not. That is a fact about capital, regulation, and patentability, and it is very nearly orthogonal to which compounds are worth studying — which is why the correct response to an evidence-absent ❌ is neither dismissal nor enthusiasm, but the specific, uncomfortable, accurate sentence: we do not know, and the reason we do not know is that nobody paid to find out.

Underneath that is the point of the whole book. This table exists because the ratings in it were produced by a method, and the method is the part you keep. Every row here is downstream of five questions — what population, what endpoint, what evidence exists, what kind of evidence would settle it, and what would change my mind — applied to one claim at a time. Those five questions do not expire in 2031. This table does.

So the test of whether the last thirty-six chapters worked is not whether you can recall that tirzepatide is ✅ for weight loss in adults with obesity. It is whether, when someone tells you about a compound invented after this book was printed, you find yourself asking in whom, measured how, compared to what — before you have any opinion at all.

Chapter 38 takes up the question this chapter kept deferring: why the evidence is distributed the way it is, and who decides which peptides get studied.

Key Terms

Master table — the reference table in §37.5 reproducing all 140 molecule-and-indication ratings issued across Chapters 2–44, organized by therapeutic area, with the source chapter given for each. Chapters 37 and 40 contribute none, because they restate ratings rather than issue them.

Claim form — a sentence shape that recurs across many compounds, rated as a form of reasoning rather than as a statement about any particular molecule. Fourteen are rated in this book (§37.6).

Evidence absent — the state behind an ❌ where adequate human trials were never run. The claim is unsupported and untested, and the row can move in either direction as soon as somebody studies it.

Evidence present and negative — the state behind an ❌ where adequate trials were run and answered no. A far stronger state of knowledge than evidence-absent, and routinely mistaken for a weaker one. Marked (N) in the master table where the source chapter made it unambiguous.

Split rating — a rating carrying two glyphs (⚠️/❌, ✅/⚠️, ⚠️/🔬) because the evidence supports a modest version of a claim and not its strong version, or supports it in one sub-population and not another, or has moved between tiers over time. Ten appear in this table, and the split is the finding, not a hedge.

NOT RATED — a claim the rating system declines to rate because it is not the kind of claim evidence settles. Four appear in this book, all in Part VIII. Not a fifth tier and not a synonym for 🔬.

Values question — a claim whose load-bearing terms encode a judgment about what ought to be valued rather than a measurable state of the world. Words like should, owes, and primarily are the usual markers.

Date-stamping — attaching an explicit assessment date to a rating so that it can become visibly rather than invisibly out of date. Every rating in this book is stamped, and this table is stamped 2026.

Falsifiability — the property, required of every rating in this book, that the rater can state in advance what evidence would change the rating and to what.

Surrogate endpoint — a measurement used as a stand-in for an outcome patients care about: bone density for fractures, HbA1c for diabetic complications, biomarker levels for survival. A ✅ resting on a surrogate is the least durable kind of ✅ (§37.9, Chapter 16).

Rating drift — the movement of a rating over time as evidence accumulates. The commonest direction for a ⚠️ is downward, because of the phase 2 optimism effect described in Chapter 9.

Directional bias — a systematic tendency to rate consistently higher or consistently lower than the evidence supports. Generous drift and severe drift are the two forms, and neither is more respectable than the other (§37.10).

Snapshot — a reference assembled at one moment, valid as a description of that moment and progressively less valid thereafter. This chapter is one, and says so.

Spaced Review

1. (Ch 37 + Ch 5) A friend tells you a compound is "unproven, so it's basically fifty-fifty." Using §37.3, explain why that sentence could describe two completely different situations, and name the one question you would ask to find out which one you are in. Then say which of the two states is more likely to change in the next five years.

2. (Ch 37 + Ch 16) Two rows in the master table are ✅. One rests on reduction in vertebral fractures; the other rests on improvement in a histological score. Explain which is the more durable ✅ and why, using the concept of a surrogate endpoint — and name the compound in the cardiovascular block whose history is the cautionary example.

3. (Ch 37 + Ch 9) You read that a compound currently rated ⚠️ in this table has just reported a large phase 3 result. Before looking at the numbers, what is the base-rate expectation for how the phase 3 effect compares with the phase 2 effect, and why? What would have to be true of the phase 3 design for the result to move the row to ✅?

4. (Ch 37) Growth hormone appears twice in the master table with opposite ratings, and so does bremelanotide, and so does oxytocin — four times. Choose one of the three and explain to someone who has read none of this book how one molecule can be simultaneously well supported and unsupported without any contradiction. Do it in four sentences.

5. (Ch 37 + Ch 5) Take one claim from your dossier where you disagreed with the book's rating. Write out (a) the claim with its population and endpoint, (b) your tier, (c) the book's tier, (d) the specific evidence that moved you, and (e) what would move you back. If you cannot complete (d) and (e), what does that tell you about which of the two piles in §37.10 Step 4 your disagreement belongs in?