Chapter 36: What's Actually Coming — Peptides in the Next Decade, and How to Read Anyone's Prediction Including This One

"Forecast, measure, revise: it is the surest path to seeing better." — Philip Tetlock and Dan Gardner, Superforecasting (2015)

Overview

Thirty-five chapters have taught one discipline: rate the claim, not the molecule, and rate it against the evidence that exists.

This chapter breaks that rule, and it has to say so before doing anything else.

A chapter about the future cannot rate claims against evidence, because the evidence does not exist. That is not a limitation of my research. It is true by construction. If a compound is in phase 2 as of this writing, in 2026, there is no phase 3 result to weigh, and no amount of care conjures one. Every other chapter could be checked against a literature. This one cannot be — which makes it the chapter most likely to be wrong, and the one where a confident tone would do the most damage.

So it does two other things instead.

First, it reports what is actually in development and at what stage. That is a checkable fact rather than a prediction. "Retatrutide is in late-stage trials as of 2026" is a claim about the present that you can verify and I can be caught getting wrong. "Retatrutide will transform obesity treatment" is a claim neither of us can check, and I will not be making claims of that kind.

Second, and more importantly, it teaches the reading skill. Section 36.2 is the longest section here and the only part I expect to be useful in ten years. Everything else has a shelf life. The six questions do not.

Now the warning the chapter is built around. By the time you read this, some of it will be wrong. A compound named here as promising will have failed; a phase will have advanced or a program will have been quietly dropped. The chapter is written to remain useful anyway, because its real subject is not what is coming but how to evaluate anyone who tells you what is coming — including me, in what follows.

One more thing, because it governs everything after. The 🔬 rating is not a promise. It means a serious question is being asked seriously. It carries no timeline and no probability of success. Most 🔬 becomes ❌. A future-facing chapter that treated the frontier as a preview of coming attractions would discredit the thirty-five chapters before it, and I would rather this chapter be dull than do that.

In this chapter, you will learn to:

• Explain why no evidence can exist for a claim about the future, and what a responsible source does instead
• Apply six questions — phase, endpoint, population, comparator, source, base rate — to any pipeline claim
• Use the attrition base rate as a prior, and refuse to let mechanism move it
• Describe where the oral peptide problem stands, including the possibility that peptides lose
• Explain why monthly dosing is a trade-off and not an upgrade
• Separate "more weight loss" from "better outcomes" in the multi-agonist race
• State what would have to be measured to settle the muscle question
• Distinguish acting on the brain from crossing the blood-brain barrier
• Separate cost-and-access arguments from efficacy arguments when patents expire
• Write a falsifiable, date-stamped forecast — and recognize that almost nobody else does

Learning Paths

All five paths should read §36.1, §36.2, and §36.10 in full. Those three are the chapter; the rest is worked application.

💊 GLP-1 — §36.3 through §36.6 are about drugs you or someone close to you may be taking. Read §36.6 twice. 🏋️ Performance — §36.6 is your section, and the one place in the book where a muscle question is genuinely open rather than settled against the marketing. Note what "open" means and what would close it. 🔬 Science — read straight through. §36.7 through §36.9 are the technical frontier. 💄 Cosmetic — a lighter chapter for you, but §36.2 applies unmodified to ingredient claims. 🏥 Clinical — §36.2 is the section to hand a colleague. §36.4's safety counterweight and §36.6's endpoint discipline are most likely to change a conversation you actually have.

36.1 Why this chapter is different, and how to read it

Every previous chapter had a literature to argue with. When Chapter 10 rated semaglutide for cardiovascular event reduction, there was a trial. When Chapter 17 rated BPC-157 for tendon healing in humans, the honest finding was that no completed randomized human trial existed — and that absence was itself the evidence, a checkable fact about the published record. Even a ❌ in this book rests on something.

This chapter has none of that. Suppose I want to tell you whether triple agonists will prove better than dual agonists at preventing heart attacks. The trials that would answer that are, as of this writing in 2026, either running or not yet started. There is no result and there cannot be one. Anything I say is a guess dressed in the vocabulary of a field that has spent thirty-five chapters teaching you to distrust guesses dressed that way.

This chapter can report status. "Compound X is in phase 3 as of this writing" is a present-tense factual claim, checkable against trial registries, regulatory filings, and the literature. If it is wrong, that is an error I made rather than a prediction that failed.

It can teach reading. The six questions in §36.2 are not about any specific compound. They are a procedure you can run on a press release published in 2031 about a molecule that does not exist yet.

It will not predict. §36.10 comes closest — it lists what would and would not surprise me — and even there the point is not the content of the lists. It is that they are falsifiable and date-stamped, so a reader in 2032 can grade them. That property separates a forecast from a prediction, and essentially every claim this book has taught you to distrust turns out to lack it.

The failure mode I am trying to avoid

Thirty-five chapters of hedged, evidence-anchored analysis build credibility. Then the future chapter arrives, the constraint of having to cite something disappears, and the author cashes that credibility in on exciting speculation. The reader quite reasonably transfers their trust: this author was careful about everything else, so when they say the next decade will be transformative, that must be a careful judgment too.

It is not. This is the one chapter where carefulness has nothing to grip.

A future-facing chapter that hypes does not just fail on its own terms — it retroactively discredits everything before it, because it reveals that the discipline was a style rather than a commitment. If I abandon the evidence standard the moment evidence becomes unavailable, the standard was never doing the work; the availability of evidence was.

So this chapter is deliberately unexciting. Where I know something I say it flatly. Where I do not, I say that instead, and the number of times I say it should feel slightly excessive. That is the correct amount.

🔍 Check Your Understanding

1. Why can this chapter not rate a claim like "triple agonists will reduce cardiovascular events more than dual agonists"? Answer in terms of what a rating requires, not in terms of difficulty.
2. What is the difference between "compound X is in phase 3" and "compound X will be approved"? Which one can catch me in an error?
3. Why would a hype-heavy future chapter damage the thirty-five chapters before it, rather than just being a weak chapter on its own?

36.2 How to read a pipeline claim

This is the most valuable section in the chapter. Everything else here expires. This does not.

Pipeline claims arrive in characteristic forms: a news article about a breakthrough, a press release, a clinic email describing what will be available "soon," a conference slide on social media, a friend's excited summary. Every one can be run through the same six questions, and they take about ninety seconds once memorized.

SIX QUESTIONS FOR ANY "COMING SOON" CLAIM

  1. WHAT PHASE, ACTUALLY?     preclinical / 1 / 2 / 3 — and how far from an answer
  2. WHAT ENDPOINT?            a surrogate, or something a person would notice
  3. WHAT POPULATION?          who was enrolled, versus who will receive it
  4. AGAINST WHAT COMPARATOR?  placebo / standard of care / nothing at all
  5. WHO IS CLAIMING, AND WHEN? press release / abstract / peer-reviewed paper / earnings call
  6. WHAT IS THE BASE RATE?    most candidates fail — start there, and make the claim move you

  If you can only ask one, ask 6. If you can ask two, add 1.
  Together those two resolve most overheated coverage before you read a single result.

They are ordered by how much time they take, not by how much they matter. Question 6 is the most useful and costs nothing, because you know the answer before you read the claim.

Question 1 — What phase, actually?

The phases are not points on a smooth continuum. They differ in kind, and the differences are where most misreading happens.

Preclinical means the compound has not been given to a human for this indication. Chapter 5 established the problem: a great many things work in a mouse. Preclinical is not "almost in trials"; it is a different category of knowledge.

Phase 1 is primarily safety and dose-finding in a small number of participants, often healthy volunteers. It speaks to tolerability and pharmacokinetics, almost never to whether the drug helps anyone, because it is generally not designed to answer that. When coverage of a phase 1 says "showed promise," ask what the trial was capable of showing.

Phase 2 is where efficacy signals first appear, at modest sample sizes, usually on surrogate endpoints. Chapter 9 established phase 2 optimism, and it is one of the most important things in this book: phase 2 results are systematically more favorable than the phase 3 results that follow. This is not fraud. It arises from small samples, selected populations, sensitive endpoints, and the ordinary statistics of which programs get reported enthusiastically. A striking phase 2 result is genuinely informative and, on average, an overestimate.

Phase 3 is the large, adequately powered trial regulators rely on, where most of the cost sits and most surprises happen.

Chapter 10 established the figure that anchors all of this: most compounds entering human trials never reach approval. The exact percentage varies by therapeutic area, by how you count, and by which decade you sample, and I am deliberately not quoting one number, because direction and magnitude matter more. The direction is that the great majority fail. The magnitude is that this holds even for compounds that reached phase 3.

So when you read that a compound is "in trials," the first reaction is not interest. It is: which trial, how far from an answer, and what fraction of compounds at this stage historically arrive anywhere?

A compound "in trials" may be years from an answer and more likely than not to fail. Popular coverage routinely drops both halves of that sentence.

⚠️ Hype Check — "it's already in clinical trials"

The claim, in its usual form:

"This isn't speculative — it's already in human clinical trials."

What's true in it. Being in human trials is a real milestone, and it separates a trialed compound from most of Part III's gray-market catalog, which has been through none of it.

Where it fails. First, the phrase flattens an enormous difference. A first-in-human study in two dozen healthy volunteers and a 15,000-participant outcome trial are both "clinical trials," and treating them as the same evidence is like treating a sketch and a finished building as both being architecture.

Second, it inverts the base rate. "It's in trials" is offered as reassurance, but the record says a compound in trials is more likely to fail than succeed. The sentence raises your confidence when the underlying fact should lower it relative to the excitement being generated.

Third, "in trials" is often true of a different indication than the one being discussed — a true sentence deployed to create a false impression.

Verdict: "in trials" tells you a compound cleared an entry barrier. It says nothing about the probability it works, and read against the base rate it should make you more cautious, not less.

Question 2 — What endpoint?

Chapter 16 established the distinction, and it does more work here than anywhere else in the book. A surrogate endpoint is a measurement believed to predict something people care about. A hard outcome is the thing itself: dying, having a heart attack, being hospitalized, climbing stairs, living alone.

Surrogates are legitimate and often necessary — faster, cheaper, and how a phase 2 decides whether phase 3 is worth attempting. The problem is that surrogate results get reported in the language of hard outcomes.

A drug that moves a biomarker has not been shown to help anyone. It has been shown to move a biomarker. Sometimes those are the same thing; historically, often they have not been.

Three pairs recur in this chapter:

In every row the surrogate is easier to measure, faster, and more photogenic. That is exactly why it appears in the press release.

Question 3 — What population?

Trials enroll people who meet entry criteria, and those criteria are narrow on purpose: a homogeneous population gives a cleaner signal with fewer participants.

The population that actually receives an approved drug is almost never the population studied. It is older, sicker, on more concurrent medications, less adherent, more varied in every way that matters. This is not a scandal; it is how development necessarily works. But "reduces X by 30 percent" honestly expands to "reduced X by 30 percent, on average, in people who met roughly forty entry criteria, over the trial's duration, against whatever the control arm received." Ask specifically: were older adults enrolled in meaningful numbers? Were the usual comorbidities included or excluded? Is baseline severity anything like that of people who would actually be prescribed the drug?

Question 4 — Against what comparator?

This is the question that separates sophisticated readers from enthusiastic ones.

A result is always a difference, measured against something. That something can be nothing at all — a single-arm or before-and-after comparison, the weakest form, universal in clinic marketing. It can be placebo, the standard for establishing that a drug does something, and limited: beating placebo tells you the drug works, not that it works better than what already exists. Or it can be an active comparator — a treatment already known to work, the hardest test and the most useful result.

Chapter 28's sacubitril/valsartan result carries the weight it does partly because the comparator was an active drug already known to work. That trial did not show the new agent beat nothing. It showed it beat a well-established therapy, on a hard outcome, in a large population — which is why it changed practice rather than merely generating headlines.

Hold that standard for §36.5. A new multi-agonist against placebo on weight loss is a far weaker claim than the same compound against tirzepatide on cardiovascular events, and the two will be reported in nearly identical language.

Question 5 — Who is making the claim, and when?

Chapter 6 established the source hierarchy. Applied to pipeline claims it gives a ladder whose rungs are routinely conflated:

THE PIPELINE CLAIM LADDER — weakest at the bottom

  PEER-REVIEWED PUBLICATION, full results, prespecified analysis
      ↑  methods visible, data tables available, reviewers pushed back
  REGULATORY REVIEW DOCUMENTS
      ↑  an adversarial reader with statistical training saw everything
  CONFERENCE PRESENTATION with a full slide deck
      ↑  more detail than an abstract; not peer-reviewed
  CONFERENCE ABSTRACT
      ↑  a few hundred words, no methods, frequently never published in full
  COMPANY PRESS RELEASE, "topline results"
      ↑  written by people with a legal duty to shareholders, not to you
  EARNINGS CALL or investor presentation
      ↑  a specific audience, a specific purpose, a specific incentive
  A NEWS ARTICLE SUMMARIZING ANY OF THE ABOVE
      ↑  add one layer of compression and one layer of headline pressure

Three distinctions do most of the work.

A press release announcing topline results is not a publication. "Topline" means a handful of headline numbers, selected by the sponsor, without the tables that would let you check them — frequently accurate, never sufficient. The gap to full publication is often a year or more, and things appear in the full paper that were not in the release: adverse event detail, subgroup behavior, discontinuation rates, prespecified analyses that did not make the summary.

A conference abstract is not a peer-reviewed paper. Abstracts are short, lightly reviewed, and a substantial fraction never appear as full publications at all — yet one presented at a major meeting will be covered as though it were a study.

A company statement at a quarterly earnings call has an audience and a purpose. The audience is investors; the purpose is to shape expectations about revenue. That does not make the statements false — they are subject to securities law and companies are generally careful. It makes them selected. Framing, emphasis, and which number leads are optimized for a goal that is not your understanding.

Add the "when." A 2023 claim about what would arrive "within two years" can be checked in 2026, and checking it calibrates a source fast. Sources that made specific, dated, checkable claims and were wrong are more trustworthy than sources that made vague claims and can never be graded.

Question 6 — What is the base rate?

The single most useful correction available, and it costs nothing.

Before you read a word, you already know something: most drug candidates fail. They fail in preclinical, in phase 1, in phase 2 most of all, and a meaningful fraction fail in phase 3 after enormous investment. That prior is your starting position, and the claim in front of you has to move you off it.

Here is the hardest discipline in the chapter: a mechanism story does not move the base rate. This is rating rule 3 — never upgrade with mechanism — applied to the future instead of the present. Same rule.

Every failed drug had a mechanism story; it is why the compound entered trials at all, since nobody funds a phase 1 with no rationale. So a compelling mechanistic account carries essentially zero information about whether this compound succeeds, because it is equally present in the failures. Chapter 2 gives you the vocabulary to find any mechanism plausible; Chapter 5's central discipline — knowing how something would work is not evidence that it does — is what stops plausibility becoming belief.

What does move you off the base rate: stage, since a phase 3 compound has cleared filters a preclinical one has not; a hard outcome already demonstrated for the same molecule in a related indication; precedent in the same mechanistic class — not the story but the track record of approved drugs working that way; and a large trial with an active comparator that has already read out. What does not: elegance of the mechanism, enthusiasm of investigators, preclinical effect size, volume of press coverage, or your own sense that it ought to work.

📊 Evidence Rating

Claim form — "the next generation will be far better than what exists now."

Claim: In its standard form — press coverage, clinic marketing, conference keynotes, investor materials — the assertion that forthcoming compounds will represent a large improvement over currently approved ones. Rating: ❌ — hype outpaces evidence, as usually stated. Reason: The claim names no endpoint, no population, and no comparator, which means it cannot be false. Better at what, for whom, against which existing option, measured how? Absent those, the sentence is unfalsifiable at the moment it is made — and rating rule 5 requires falsifiability for a claim to be rated at all. The ❌ describes the evidence available for the claim as constructed, not a prediction that future drugs will disappoint. Some will be better. The sentence still tells you nothing. What would change it: A specified version. "Compound X produces a greater reduction in cardiovascular events than tirzepatide in adults with obesity and established cardiovascular disease over four years" is rateable — currently 🔬, and a trial could settle it. The instant someone supplies endpoint, population, and comparator, the claim becomes gradeable and the ❌ no longer applies.

Why rating a claim form matters. This is the single most common sentence in coverage of this field, and it is almost never challenged, because it sounds like a summary rather than an assertion. Rating it teaches what molecule-by-molecule ratings cannot: a claim can fail before you look at any evidence, purely on its construction. Hence the Claim form — prefix, marking a rating about how something is said rather than about what was studied.

36.3 The oral problem, and where it stands

Chapter 4 established why peptides are injected, and Chapter 1 gave the two-line version: your digestive system contains enzymes whose evolved purpose is breaking peptide bonds, and a surviving peptide is generally too large and too polar to cross the intestinal wall. Two barriers, in series.

Oral semaglutide is the existence proof that this can be solved. It is a real, approved, prescribed medicine — approved in 2019 in the United States — co-formulated with an absorption enhancer called SNAC.

🧬 The Molecule — SNAC, and what one percent actually means

SNAC (sodium N-[8-(2-hydroxybenzoyl)amino] caprylate) is not a peptide and is not the drug. It is an excipient that travels with the drug and changes the local environment in the stomach.

Two things happen where the tablet dissolves. SNAC transiently raises local pH, reducing the activity of pepsin — the stomach's principal protease — in that microenvironment and protecting the peptide from being cut. And it promotes absorption across the gastric epithelium, letting some fraction of intact molecule enter circulation from the stomach rather than the intestine, which is not where peptide absorption would normally be attempted at all.

Both effects are local, transient, and dependent on the tablet dissolving in a specific way in a specific place. That is why the administration conditions on the label are unusually strict: not fussiness, but the conditions under which the mechanism functions.

The honest number is that bioavailability remains on the order of 1 percent. Roughly ninety-nine percent of the drug substance never reaches systemic circulation.

Sit with what that implies. An oral tablet must contain far more drug substance than an injection delivering a comparable effect — a large multiple, not a modest one. Every milligram was made by the process Chapter 32 describes: linear, stepwise, solvent-intensive solid-phase synthesis. The oral formulation therefore consumes vastly more manufacturing capacity per patient-year than the injectable does — a direct input to §36.9, and one reason oral and injectable versions of the same molecule can differ so much in availability and price.

Higher-dose oral formulations have been developed and studied — the expected engineering response: if one percent gets in, put more in the tablet. It works, and it makes the manufacturing arithmetic more demanding rather than less.

📊 Evidence Rating

Claim: Oral peptide delivery via absorption enhancers works as a general solution to the oral peptide problem. Rating: ⚠️ — promising but preliminary, as a general solution. (The specific case, oral semaglutide for its approved indications, is not preliminary at all; it is approved and in use.) Reason: It demonstrably works, in an approved product, in millions of people — at roughly 1 percent bioavailability, with strict administration requirements, and with real consequences for cost, manufacturing capacity, and supply. Whether the approach generalizes to peptides of different size, charge, and stability is not established, and the number of approved oral peptides remains very small relative to injected ones. What would change it: Approvals of orally delivered peptides in other classes using similar enhancer strategies, at bioavailability substantially above the current order of magnitude, would move this toward ✅ as a platform. A decade with no additional enhancer-based oral peptide approval would suggest semaglutide is closer to a special case than a template. Date-stamped: as of this writing, 2026.

The other route: stop using a peptide

Here is the part most coverage misses, and it is the more likely ending.

The other route to an oral drug is to stop using a peptide at all.

Orforglipron is a non-peptide, small-molecule GLP-1 receptor agonist in late-stage development as of this writing. Not a modified peptide, not a shortened one, not a peptide with a protective group — a small molecule that happens to activate the same receptor.

The consequences are obvious once stated. No absorption enhancer is needed, because small molecules of the right physicochemical profile are absorbed from the gut as a matter of course. No cold chain, which changes distribution economics profoundly, especially outside wealthy countries with reliable refrigeration. And manufacturing by conventional organic chemistry, typically at far lower cost per dose.

Chapter 33 §33.7 and Chapter 35's captopril story both end here. Chapter 35 told you how a peptide from pit viper venom led, through deliberate medicinal chemistry, to an orally active non-peptide drug in use for decades. The peptide was the lead; the drug was not a peptide. Chapter 33 §33.7 gave the general strategy — peptidomimetics, the redesign of a peptide lead into something no longer a peptide that retains the binding interaction that mattered.

So, plainly, because it cuts against the framing of nearly every article you will read about oral peptides:

The most likely solution to the oral peptide problem may be that the successful oral drugs are not peptides.

That is not a defeat for peptide science. It is arguably its highest achievement — the peptide identified the target, characterized the interaction, and proved the pharmacology in humans, and then chemistry built something swallowable. But a reader waiting for oral versions of the injected peptides they know may be waiting for the wrong thing.

36.4 Longer and longer: monthly dosing and implants

The duration trajectory is one of the cleanest engineering stories in this book.

DURATION OF ACTION — the trajectory so far

  Native GLP-1              minutes         degraded by DPP-4 almost immediately
  Exenatide                 twice daily     a protease-resistant natural analog
  Liraglutide               once daily      fatty acid acylation, albumin binding
  Semaglutide               once weekly     further acylation plus a DPP-4-resistant substitution
  Under investigation:      monthly+        depot formulations, implants, novel scaffolds

  Each step required a specific engineering intervention (Chapter 33), and each
  delivered a real improvement in adherence. The arrow points right.
  It does not follow that further right is better.

Monthly and longer intervals are under investigation, along with implantable and depot approaches that establish a reservoir releasing slowly over weeks or months. The convenience argument is real: adherence to weekly injection is imperfect, adherence to daily is worse, and much of the real-world gap between trial and clinical results is an adherence gap.

Now the counterweight, and this section exists for the counterweight.

You lose the ability to stop. If an adverse effect appears on a weekly drug, discontinuation produces declining exposure over the following weeks. If it appears on a drug engineered for a three-month duration, or delivered from an implanted depot, exposure continues regardless of what anyone decides. For a well-tolerated drug in a healthy population this is minor. For a drug with meaningful gastrointestinal effects, in an older population, with surgery or an intercurrent illness in the picture, it is not. Some depots can be removed; many cannot, and removal means a second procedure.

Dose adjustment becomes coarse. Chapter 13 described the titration schedules used with GLP-1 receptor agonists; titration exists because tolerability improves with gradual escalation. An interval measured in months makes titration slow, imprecise, or impossible — and makes individualization, which is most of what clinical practice consists of, much harder.

And the pharmacology itself pushes back. This is the one engineering cannot design around. Chapter 2 §2.8 established that sustained receptor occupancy is exactly what drives desensitization and downregulation. A continuously occupied receptor is one whose cell reduces responsiveness — by phosphorylation and internalization on the fast timescale, by reduced receptor expression on the slow one. Chapter 33 §33.10 named these the two termination mechanisms engineering cannot defeat, because they sit downstream of the receptor rather than being properties of the drug. Build a molecule that resists every protease in the body and escapes renal filtration, and the cell will still adjust.

Native peptide signaling is pulsatile for a reason. Chapter 3 made this the central lesson of endogenous versus exogenous signaling: the body's own peptides arrive in bursts, terminated on cue, and a drug's flat sustained exposure is a fundamentally different stimulus. Extending duration moves further from the physiological pattern, not closer to it.

Duration is a trade-off being optimized, not a scale being climbed. The optimum is not "as long as possible." It is wherever adherence benefit stops outrunning loss of control, and where receptor biology has not yet started charging rent. Where that sits is empirical, differs by drug and patient, and is not answered by announcing a longer interval.

🩺 Safety and Risk — what a longer interval changes about a conversation

If a longer-acting formulation becomes available, the questions worth raising with a clinician are not about the technology. They are about reversibility.

Can it be stopped, and how fast? For a depot or implant, is removal possible, and does it require a procedure?

What happens if something else comes up? Surgery, pregnancy, a new diagnosis, a new medication, an acute illness with vomiting or dehydration. A weekly drug can be paused. A three-month depot has already been given.

How will the dose be individualized, if titration is unavailable? And what adherence problem is actually being solved? If someone reliably takes a weekly injection, the convenience gain is smaller than it sounds, and the control given up is real.

None of this argues against long-acting formulations, which solve a genuine problem for the many people who do not take weekly injections reliably. It argues that "longer" is a design choice with two sides, and only one appears in the announcement. The decision belongs with a clinician who knows the history — Chapter 39 is about making that conversation productive.

36.5 The multi-agonist race and what it is actually optimizing

As of this writing, in 2026, the state of play is roughly this.

Dual agonism is approved. Tirzepatide activates both the GLP-1 and GIP receptors; Chapter 9 covered the pharmacology and the trial program.

Triple agonism is in late-stage development. Retatrutide targets the GLP-1, GIP, and glucagon receptors. Its phase 2 program reported mean weight reductions that were, at publication, larger than any previously reported for a pharmacological agent. Phase 3 is underway as of this writing. I am deliberately not stating an approval date, because I do not have one and inventing one would violate the standard this book has held for thirty-five chapters.

Amylin is the other axis. Chapter 13 introduced amylin as a genuinely separate satiety pathway rather than another route into GLP-1 biology. Cagrilintide is a long-acting amylin analog, and CagriSema combines cagrilintide with semaglutide — two mechanisms, one product — and has been through a late-stage program as of this writing.

🔬 Read the Study — when "disappointing" and "ineffective" are different words

When topline results from a large CagriSema trial were announced, the reported weight loss was below what the company had previously guided investors to expect. The commercial reaction was severe.

Read the same number two ways. As an investor: the figure missed guidance, changing revenue models and valuation — a rational disappointment. As a patient or clinician: it was a degree of weight loss that would have been considered extraordinary at any point before roughly 2021, achieved pharmacologically, in a large trial.

Both readings are correct, and they answer different questions. The market asked "did this beat expectations?" A clinician asks "does this help people, and how does it compare to alternatives?" The coverage merged them, producing headlines a general reader could easily read as "the drug did not work."

This is question 5 doing real work. Who is making the claim, and to whom? The number was not wrong; the frame was built for somebody else. Neither triumph nor failure framing is a clinical judgment.

The question nobody asks loudly enough: what is being optimized?

More weight loss is the headline of every multi-agonist story — the number in the press release, compared across compounds as though they were lap times.

But weight loss is itself a surrogate.

Return to question 2. Anyone cares about weight loss in a medical context because it is believed to predict things people actually care about: heart attacks, strokes, diabetes, joint replacement, sleep apnea, mobility, death. Weight is the measurement. Those are the outcomes.

Chapter 10's SELECT result is the model. That trial did not report kilograms as its primary result. It reported cardiovascular events, in people with established cardiovascular disease and overweight or obesity without diabetes, against placebo, over years. The claim that mattered was cardiovascular events, not kilograms — which is why it changed guidelines and coverage decisions in a way weight-loss data alone had not.

Now apply that standard:

A drug that produces more weight loss than tirzepatide has not thereby been shown to prevent more heart attacks.

That should feel almost pedantic, and it is not. The inference can fail several ways. The relationship between additional weight loss and additional risk reduction may not be linear — the first fifteen percent may buy most of the benefit. Mechanisms may contribute independently of weight; glucagon receptor agonism does things to energy expenditure and hepatic metabolism not reducible to the scale. Compounds may differ in effects that outcome trials capture and weight-loss trials do not. And composition matters — §36.6 is about the possibility that not all weight lost is equally good to lose.

The outcome trials that would settle this are longer, larger, and slower, because hard outcomes are rarer events than kilograms and you need years and thousands of participants to count enough of them. That is why weight-loss data always arrives first, why it dominates coverage, and why the gap between "more weight loss" and "better outcomes" persists for years after the first number is known.

📊 Evidence Rating

Claim: Triple agonists (e.g. retatrutide) produce clinically meaningful weight loss in adults with obesity. Rating: ⚠️ — promising but preliminary. Reason: There is substantial late-stage weight-loss data as of 2026, and the magnitude reported in earlier-phase work was large enough that it is unlikely to be entirely a phase 2 optimism artifact — though Chapter 9 warns that some regression toward a smaller effect in phase 3 is the norm rather than the exception. Approval is pending as of this writing and I am not predicting it. What would change it: Completed, published phase 3 results consistent with the earlier data would move this toward ✅ for the weight-loss endpoint specifically. Substantially smaller phase 3 effects, or a tolerability profile producing high real-world discontinuation, would hold it at ⚠️ or worse.

And the separate, unanswered claim: Triple agonists reduce cardiovascular events more than currently approved agents. 🔬 — a serious question under investigation, with no answer as of this writing. Cardiovascular outcome evidence is a separate claim from weight-loss evidence, requires a separate and much larger trial, and cannot be inferred from the weight number however large. This is rating rule 6 in its purest form: one molecule, many ratings. Retatrutide holds ⚠️ for one claim and 🔬 for another, simultaneously, and any source giving you a single verdict on "retatrutide" has thrown away the distinction you need.

36.6 The muscle question — the field's most important open problem

This is the section that makes this chapter worth reading rather than skipping. The honest answer is genuinely "we do not know," the question matters to millions of people right now, and — unusually for this book — the uncertainty is not a polite way of saying the claim is unsupported. It is real, load-bearing uncertainty about something important.

What is established

Substantial weight loss from any cause includes loss of lean mass, not only fat mass. Dietary restriction does it. Bariatric surgery does it. Illness does it. Pharmacological weight loss does it.

This is not a novel finding and it is not specific to GLP-1 drugs. It has been documented across every weight-loss modality studied, for decades. The proportion varies by study, measurement method, rate of loss, and what the person was doing with their body during it — variation that is part of why the question is hard. But the phenomenon is not in dispute, and coverage presenting it as a newly discovered hazard of GLP-1 drugs specifically is getting the history wrong.

What is genuinely open

"Open question" gets used to mean both "nobody has checked" and "people have checked carefully and it is not yet clear." This is the second kind.

One: how much of the lean mass lost is functional muscle? Lean mass as measured by body composition methods is not the same as contractile skeletal muscle. It includes fluid, connective tissue, organ mass, and glycogen with its associated water. Someone losing a large amount of weight loses some of the structural and fluid load that supported a larger body, and it is not obvious that this is harmful — some may be appropriate remodeling. Distinguishing "lost muscle you needed" from "lost supporting tissue you no longer need" is technically difficult, and it is not what a standard scan reports.

Two: does it matter, and for whom? The concern is not the number. It is function: strength, walking speed, ability to rise from a chair, falls, and the outcome that matters most, independence. The concern is sharpest in older adults — who have less reserve, in whom sarcopenia is already a clinical concern, and who are also a population in whom substantial weight loss has documented benefits. Both are true at once, which makes this a real clinical question rather than a slogan.

Three: do resistance training and adequate protein intake mitigate it? For weight loss by conventional means this is generally believed and reasonably supported. Whether it works equally well during pharmacologically driven weight loss, where appetite suppression may make adequate protein intake harder, is less established than the confident version of the advice suggests. A reasonable default; not a settled finding in this context.

Four: should any pharmacological agent be added to prevent it? This is the question with money behind it, and where the pipeline lives.

The pipeline, and a detail most coverage omits

Agents targeting the myostatin/activin pathway are in trials for this indication. Chapter 16 introduced myostatin as a negative regulator of muscle mass — block it and muscle mass increases, a finding vivid enough to have fueled two decades of supplement marketing. Candidates work at various points in that pathway: neutralizing myostatin or its precursor, or blocking the activin type II receptors through which it signals. Bimagrumab, an antibody against those receptors, is among the compounds studied in this context.

And here is the detail that belongs in a peptide book: several of the leading candidates are antibodies rather than peptides.

Look at that name with Chapter 1 §1.8 in hand. Bimagrumab. The -mab stem means monoclonal antibody: roughly thirty times the mass of a peptide, produced in living cell culture rather than by chemical synthesis, with the cost structure and long half-life that follow. A reader who has learned to decode drug names can see that from six letters.

It is a reminder that the interesting frontier is not always peptide-shaped — §36.3's orforglipron was the other example — and that a book organized around peptides will systematically under-notice the antibody and small-molecule answers to the same problems.

The discipline: what would actually answer the question

Adding a second drug to manage the effects of a first is a claim that requires its own outcome evidence. It is not a correction. It is a new intervention with its own risk profile, cost, and potential for harm, and it must clear the same bar as any other. That it is added to fix a problem caused by something else earns it exactly zero credit.

And the endpoint has to be right. The endpoint that matters is function — strength, mobility, falls, independence — not a body-composition scan.

A change in a DEXA reading is a surrogate (Chapter 16). A trial can demonstrate convincingly that adding a myostatin-pathway agent preserves lean mass during weight loss; that result can be entirely real; and it can still fail to mean anything for the people taking it. The chain from "preserved lean mass on a scan" to "better function" to "fewer falls, more independence" has multiple links, each an assumption until measured.

One wrinkle makes this surrogate especially treacherous: an agent that adds lean mass to the scan has, mechanically, improved the thing being measured. The intervention acts directly on the yardstick, and a drug that raises the number you are using to judge it demands more care than usual, not less.

📊 Evidence Rating

Claim: Pharmacological prevention of lean mass loss during weight loss improves functional outcomes — strength, mobility, falls, independence — in adults losing substantial weight on GLP-1-based therapy. Rating: 🔬 — frontier. Reason: A serious question, under active trial as of this writing in 2026, with no completed evidence on functional endpoints. Agents targeting the myostatin/activin pathway can plausibly preserve lean mass on body-composition measurement; whether that translates into function is unestablished, and body composition is a surrogate (Chapter 16). Several leading candidates are antibodies rather than peptides. What would change it: An adequately powered trial in the relevant population — realistically including older adults, where the concern concentrates — reporting functional endpoints rather than composition alone: measured strength, gait speed, chair-rise, falls, or independence outcomes, against an appropriate comparator, ideally including a resistance-training-plus-protein arm. A trial reporting only DEXA results would not move this rating however favorable it was.

And the caveat this chapter keeps repeating: this is not an endorsement and not a forecast. 🔬 means a serious question is being asked seriously. It does not mean coming soon, and it carries no probability of success. Most 🔬 becomes ❌. This is the question I would most want answered in the next five years, and I do not know how it comes out.

💊 In the Clinic — what this means for someone taking one of these drugs now

A reasonable person reading the last three pages might conclude they should be worried. That is not what the evidence supports, and the honest framing is narrower.

Established: substantial weight loss of any kind includes lean mass; this has been true of every modality studied; and the benefits of weight loss in people with obesity-related disease are supported by hard outcome data (Chapter 10).

Not established: that lean mass lost on these drugs is functionally worse than lean mass lost by other means, that it produces meaningful impairment in most people, or that any drug should be added to prevent it.

Uncontroversial and available now: resistance training and adequate protein intake are recommended during weight loss by essentially every clinical source, for exactly this reason, and carry no pharmacological risk.

Worth raising with a clinician, particularly for older adults or anyone with existing mobility concerns: whether functional status should be tracked, and what that would look like. Grip strength and gait speed are cheap, quick, and measure the thing that matters. A scan measures the surrogate.

As everywhere in this book: this describes an open question, not advice about your treatment. Chapter 39 is about having that conversation well.

36.7 Peptide-drug conjugates and targeted delivery

Here is a frame that inverts one of the book's central assumptions. Everywhere else, a peptide's properties are problems to engineer around: short half-life, poor tissue distribution, inability to cross membranes.

Peptide-drug conjugates invert this. Here, selective binding is the entire product, and the peptide is a delivery vehicle rather than the active agent.

Take a peptide that binds one receptor with high selectivity. Attach a payload — something that does damage, or emits, or would be too toxic to give systemically. The peptide's selectivity becomes a targeting system: the payload goes where the receptor is, and less of it goes everywhere else.

ANATOMY OF A PEPTIDE-DRUG CONJUGATE

   [ TARGETING PEPTIDE ] —— [ LINKER ] —— [ PAYLOAD ]
      binds one receptor      holds them     does the work
      with high selectivity   together; may  (cytotoxic agent,
      = the address           be designed    radioisotope,
                              to release at  or other)
                              the target

   The peptide's job is NOT to have an effect. Its job is to arrive somewhere
   specific. Everything the book usually calls a limitation — short half-life,
   restriction to cell-surface targets, rapid clearance — can be an asset here.

That last line deserves emphasis. Rapid clearance of unbound drug is a feature here: whatever fails to find its target should leave promptly, because it is carrying something you did not want distributed. A property that is a liability for a hormone analog is a design advantage for a delivery vehicle.

Peptide receptor radionuclide therapy (PRRT) is the approved existence proof, covered in Chapter 27. The architecture is exactly the diagram. A somatostatin-analog peptide provides targeting — somatostatin receptors are densely expressed on many neuroendocrine tumor cells, and octreotide-family peptides bind them selectively. A chelator is chemically attached, caging a metal ion, and a therapeutic radioisotope sits in the cage. The construct circulates, concentrates where the receptors are, and delivers radiation locally. Lutetium-177 dotatate is the approved example, used in somatostatin-receptor-positive gastroenteropancreatic neuroendocrine tumors, with a randomized trial behind it.

There is an elegant corollary: the same targeting peptide carrying an imaging isotope produces a scan showing where the receptors are, so a patient can be imaged for target expression before receiving the therapeutic version. The scan and the therapy share an address.

📊 Evidence Rating

Claim: Peptide-drug conjugates work as a general targeted-delivery platform across indications and payload types. Rating: 🔬 — frontier. Reason: The approach is approved and established in one specific application — PRRT with a somatostatin-analog targeting peptide in somatostatin-receptor-positive neuroendocrine tumors. Everything beyond that is early. Cytotoxic and other payloads are under investigation, and generalizing from one successful receptor-payload pairing to a platform is exactly the inference this book keeps warning against. What would change it: Approvals in additional indications, with different targeting peptides and payload classes, would move this toward ⚠️ and eventually ✅ as a platform claim. A decade in which radionuclide conjugates remain the only approved application would suggest the specific case does not generalize — plausibly because the payload matters: a radioisotope does its damage from a distance and need not be released inside the cell, an easier engineering problem than delivering a cytotoxic molecule intact to an intracellular target.

Keep the two claims visibly separate. PRRT for somatostatin-receptor-positive neuroendocrine tumors is ✅ — approved, trialed, in use. Peptide-drug conjugates as a platform is 🔬. Same technology, two claims, two ratings, and collapsing them is how a single approval becomes a marketing story about a revolution in targeted medicine.

36.8 Crossing the blood-brain barrier

Chapter 20 established the barrier: a specialized endothelium with tight junctions and active efflux transporters that excludes the overwhelming majority of large, polar molecules from the central nervous system. Peptides are large and polar. The barrier is why Part IV reads as a series of near-misses.

Chapter 35 showed the brute-force answer. Ziconotide — the cone snail venom peptide — is an approved analgesic of extraordinary potency that cannot cross the barrier at all, so it is delivered directly into the cerebrospinal fluid by an implanted pump. That works, and the fact that it requires neurosurgical hardware tells you how hard the problem is.

Two research routes aim to do better.

Receptor-mediated transcytosis exploits the fact that the barrier is a checkpoint rather than a wall. Certain endothelial receptors actively transport their cargo across — the transferrin receptor is the most studied, moving iron-bound transferrin into the brain as ordinary physiology. Attach a therapeutic peptide to a ligand for such a receptor and in principle it rides across as a passenger. The strategy is real, it has produced constructs that measurably increase brain exposure, and it has been difficult to convert into approved medicines. The obstacles are unglamorous: binding tightly enough to be picked up but loosely enough to release on the far side, avoiding interference with the receptor's physiological job, and getting enough material across to matter.

Molecular shuttles are peptides designed as carriers rather than borrowed from an existing ligand. Chapter 35 §35.10's computational design methods are directly relevant, since a shuttle is a well-posed target: a short sequence with one required property.

As of this writing, neither route has produced an approved central nervous system peptide drug that works by crossing the barrier. 🔬 — and note that this frontier has resisted competent effort for decades, which is itself information: a problem that has held out that long is usually harder than the current proposal accounts for.

The complication most coverage omits

The GLP-1 drugs' central effects on appetite do not require the drug to cross the blood-brain barrier in bulk.

Some brain regions do not have a complete barrier. The circumventricular organs — including the area postrema, in the brainstem adjacent to nuclei involved in nausea and appetite — have fenestrated capillaries specifically so their neurons can sample the blood. The brain needs places where it can detect circulating signals directly, positioned next to the circuits that respond. A peptide that never crosses an intact barrier anywhere can still act on neurons there.

Vagal afferent signaling provides an indirect route. Chapter 7 described the gut-brain axis and Chapter 13 applied it: GLP-1 receptors on vagal afferent terminals in the gut and hepatic portal region transmit signals centrally without the molecule going near the brain. The information crosses; the molecule does not.

Which produces the distinction this section exists to make:

"It acts on the brain" and "it crosses the blood-brain barrier" are different claims, and conflating them is one of the most common errors in popular coverage of this drug class.

The confusion runs both ways. Someone observes that GLP-1 drugs affect appetite or reported cravings, concludes the drug must be entering the brain in quantity, and infers things that do not follow. Or someone observes correctly that these are large peptides that do not readily cross, and concludes the central effects must be imagined — equally wrong, because the effects are real and the routes above explain them.

The correct statement is narrow: these drugs produce genuine central effects through anatomically specific routes that do not require bulk penetration of an intact blood-brain barrier. Longer and less quotable than either error, which is roughly why you will not see it in a headline.

36.9 Manufacturing, supply, and what happens when patents expire

The least glamorous section here, and possibly the most consequential, because it governs whether any of the above reaches anyone.

Chapter 32 established how peptides are made, and the structural fact is that solid-phase synthesis is linear, stepwise, and solvent-intensive. Each residue is added in a cycle — deprotect, couple, wash, repeat — so a thirty-residue peptide takes thirty cycles, each carried out on the full quantity of material. Yield losses compound multiplicatively, and solvent volumes vastly exceed product mass.

This means scale is a genuine constraint rather than a formality. You cannot answer a demand surge by running the reactor harder. You need more reactor capacity, more solvent handling, more purification, more of everything downstream — and building any of it takes years plus regulatory qualification.

The shortages of GLP-1 drugs in the mid-2020s were a real manufacturing and fill-finish story, not only a demand story. Two bottlenecks, and the second surprises people. The first was synthesis capacity for drug substance. The second was fill-finish — sterile filling into the final container and assembly of injection pens. It is entirely possible to have drug substance in a warehouse and be unable to ship, because there is no capacity to put it into pens. Device assembly is specialized, capital-intensive, heavily regulated, and it was binding.

Recall §36.3's arithmetic. If an oral formulation delivers roughly one percent of what it contains, every oral patient-year consumes substantially more synthesized peptide than an injectable one. Oral formulations relieve the fill-finish bottleneck and worsen the synthesis bottleneck.

Research directions all target the same thing — fewer steps, less solvent, better yield. Enzymatic and chemoenzymatic synthesis uses engineered ligases to join fragments in water rather than organic solvent, letting a long peptide be built from a few pieces rather than one residue at a time; that changes the arithmetic from linear to convergent. Greener solvents address traditional solid-phase solvents now under regulatory pressure. Hybrid recombinant-plus-chemical routes produce the backbone biologically and perform the modification synthetically. None of this is speculative the way §36.7 and §36.8 are; it is process development, and its pace is set by regulatory qualification of manufacturing changes, which is slow by design.

On patents: be careful

I am going to be deliberately imprecise here, and the imprecision is the point.

Compound patents on semaglutide expire on different dates in different jurisdictions, and the spread is wide. Patent terms, extensions, and the outcomes of individual national filings differ, and the result is that several major markets outside the United States are expected to see expiry considerably earlier than the United States does. The earliest expiries are expected in the second half of the 2020s in some markets.

I am not giving you specific dates. Dates shift with litigation and administrative decisions, and inventing precision about a patent expiry is exactly the kind of fabrication this book forbids itself. For a specific country, ask a current legal or regulatory source, not a textbook.

What generic entry changes: price, and therefore access. Chapter 12 laid out the access landscape — coverage decisions, out-of-pocket costs, the gap between who could benefit and who can obtain. A drug at a small fraction of its current price is a different policy object: coverage calculations invert, health systems that could not consider population-level provision can consider it, and the gray market's principal selling point — cost — erodes considerably. Chapter 19's economics are downstream of this.

What it does not change: any evidence about whether the drugs work.

That distinction is this section's teaching point. Chapter 5 established that a rating attaches to a claim about a molecule's effect in a population on an endpoint. The price of the molecule is not in that sentence. When semaglutide becomes inexpensive it will not have become more effective. When a company defends its patent position, that says nothing about whether its drug works. Arguments about cost, access, profit, and fairness are real and important — this book takes no position on them — and they are a different category from evidence about efficacy and safety.

You will see the categories mixed in both directions: people who dislike pharmaceutical pricing casting doubt on the clinical evidence, and people defending the clinical evidence dismissing pricing concerns as unserious. Errors of the same shape. Keep the two ledgers separate.

36.10 What would actually surprise me

I am going to write this in the first person, and it is the most honest section in the book.

Everything above was either a checkable statement about the present or a piece of reading technique. This is neither. This is my judgment, with no evidence behind it — and the only thing that makes it worth including is its structure. Every item is specific enough that you can tell, later, whether it happened.

Things that would not surprise me

If any of these occurs by the mid-2030s, I will not consider it a notable development, and neither should you:

• More multi-agonists. The pharmacology is understood, the path is established, the incentive is enormous.
• A monthly or longer-interval formulation reaching approval. (§36.4 stands: I would expect the trade-offs to become clinically visible after approval rather than before.)
• An approved oral small-molecule GLP-1 receptor agonist. Late-stage programs exist as of this writing.
• Additional approved indications for existing GLP-1 drugs. This class keeps producing outcome data in conditions adjacent to metabolic disease.
• A designed peptide binder entering clinical trials. Chapter 35 §35.10 described the methods, and reaching phase 1 is a far lower bar than reaching approval.
• Further supply constraints. §36.9's manufacturing physics has not changed and demand has not stopped growing.

None of that is a prediction these will happen. It is a statement that if they do, my model of the field is unchanged.

Things that would surprise me

These would require me to revise something I currently believe:

• An oral peptide with high bioavailability achieved without an absorption enhancer or a peptidomimetic redesign. If someone gets a genuine peptide across the gut wall at anything like small-molecule bioavailability using neither known strategy, I would want to understand what I have been wrong about since Chapter 4.
• Any of the gray-market compounds in Part III producing a positive, well-powered outcome trial. Not a positive small trial, not a surrogate result, not an animal study — a properly powered human trial on an endpoint that matters. I would update and say so. The reason for the surprise is not that these compounds are impossible; it is that the ones with the loudest claims have had decades and considerable commercial motivation and have not produced such a trial (Chapters 17 through 19).
• An antimicrobial peptide approved as a systemic antibiotic. Chapter 25 explained why: the properties that make antimicrobial peptides effective against bacterial membranes are hard to separate from toxicity to human membranes, and systemic administration is where that concentrates. Topical and local uses are a different matter.
• An AI-designed molecule reaching approval materially faster end-to-end than a conventionally discovered one. This is specifically about the end-to-end timeline, and Chapter 35 §35.10 argued that discovery was never the constraint. Trials are — their duration, size, and the rate at which the outcomes you are counting occur. Compressing discovery from four years to four months moves a small fraction of a fifteen-year path. If one arrives dramatically faster overall, either the trial process changed or I was wrong about where the time goes.

Things I would bet against but cannot rule out

Lower probability in my estimation, and I hold them loosely enough to want them written down:

• A peptide reversing a neurodegenerative disease. Not slowing, not modifying, not improving a biomarker — reversing. §36.8's barrier is only the first obstacle; the deeper one is that dead neurons are dead. I would bet against and would be delighted to lose.
• A compound producing muscle growth in healthy adults comparable to what the marketing already claims. Note the comparison class. Part III's growth-hormone-axis compounds are marketed with implied effects no approved drug in the category has demonstrated in healthy adults.
• A growth-hormone-axis intervention extending healthy lifespan in humans. The oldest promise in the peptide space, with the weakest evidence relative to its persistence. The trial that would test it has never been run at adequate scale — and note that "never properly tested" is a different statement from "tested and failed," which is exactly why it sits in this list rather than the previous one.

Why the lists are the point

Those lists are falsifiable. They are date-stamped to 2026. And a reader can check them.

That is the difference between a prediction and a forecast. A prediction says what will happen. A forecast says what will happen and tells you what would prove it wrong. It can be graded. It exposes the forecaster to being caught. It creates a record that accumulates into calibration — the thing that distinguishes people worth listening to about the future from people who merely sound worth listening to.

If most of the third list happens, I was badly wrong, and the record will say so in my own words. I made that easy on purpose, because the alternative — a chapter of unfalsifiable enthusiasm — is precisely what this book has spent thirty-five chapters teaching you to detect.

Every claim this book has taught you to distrust has lacked exactly this property. "Peptides are the future of medicine." "The next generation will be far better." "This compound has enormous potential." "Trials are underway and results are expected to be exciting." Not one of those sentences can be wrong. No observation would refute any of them, which is why they are so comfortable to say and so useless to hear. They are not claims. They are moods with the grammar of claims.

So here is the test to apply to the next futurist you encounter, including me: ask what would make them wrong. If they answer specifically, listen carefully — whatever their track record, they are engaged in the right activity. If they cannot, or treat the question as hostile, you have learned that whatever they are doing, it is not forecasting.

📋 Your Evidence Dossier

Adding 🔬 frontier entries — without letting them creep upward.

This chapter gives you reason to add a new kind of entry. Until now your dossier has held compounds you had reached some conclusion about. Now you have met things genuinely in motion: oral delivery strategies, longer-acting formulations, multi-agonists, muscle-preserving agents, conjugates, barrier-crossing constructs.

Adding them is legitimate and useful. A dossier containing only settled questions will go out of date and will not know it. A frontier entry is how you track something you are watching rather than something you have decided.

It is also the single most dangerous thing you can do to your own dossier.

A 🔬 entry has all the emotional properties of a promise. It is exciting. It is about the future. And nothing in it can yet be disproved, which means that unlike every other entry, it faces no resistance. Your ⚠️ entries are held in place by trials that came in smaller than hoped. Your ❌ entries are anchored by an absence you can point at. A 🔬 entry is anchored by nothing, and it will float upward unless you tie it down.

Three rules. Appendix C has the blank workbook.

Rule 1 — A 🔬 entry must name the readout that would move it

Before anything else, write the specific result that would change the rating: what trial, what population, what endpoint, what comparator.

If you cannot name it, it is not a frontier entry — it is an enthusiasm. Delete it or downgrade it, because an entry no conceivable result would change is not tracking anything.

This is the discipline Chapter 6 introduced and Chapter 35 sharpened, applied to a claim about the future rather than the present. It is only harder here, because the future is where specifying a disconfirming result feels least necessary and is most necessary.

Rule 2 — A 🔬 entry must carry a date

Write the date you created it, and revisit on a schedule.

Frontier status decays. A compound that has been "promising" for ten years is telling you something, and what it is telling you is usually not that the science is hard. Sometimes it genuinely is — §36.8's barrier work has resisted competent effort for decades. More often, a long-running 🔬 means the readout that would have settled it either came back unfavorably and was not publicized, or was never seriously attempted.

An undated 🔬 entry cannot age, and an entry that cannot age can never be found stale. That is how people carry hopeful beliefs about specific compounds for entire decades without noticing they have.

Rule 3 — A 🔬 entry may never be upgraded by news

The hard one. A press release does not move a rating. Neither does a conference abstract, a funding round, a partnership announcement, an acquisition, a new mechanism paper, an encouraging animal result, or an investigator interview.

Only the readout you named in rule 1 moves the rating.

This is rating rule 3 — never upgrade with mechanism — applied to your own file. You already accept it when evaluating someone else's claim. It is considerably harder to apply to a compound you have followed for two years and have started, without deciding to, to want to work.

The failure mode, named plainly

Ratings drift upward over time and almost never downward.

The reason is a selection effect — the same one Chapter 6 described for testimonials, operating on your own attention rather than on a product page.

Once you are watching a compound, essentially all information reaching you about it is information somebody chose to publicize. Companies announce progress and go quiet about setbacks. Investigators publish positive findings faster and more prominently than null ones. Your feed learns what you clicked. The stream of news about any compound you follow is filtered, and filtered in one direction.

So you receive twelve encouraging items and zero discouraging ones and conclude the case has strengthened. It has not. You have observed what the filter passes. A discontinued program produces no headline; a failed phase 2 is often buried in one sentence of a quarterly update, if reported at all. Absence of bad news about a compound you are watching is close to uninformative, and it feels exactly like good news.

The correction: re-read your 🔬 entries against your own rule 1, not against the latest headline. The question is never "has there been encouraging news?" It is always "has the readout I named arrived, and what did it say?" If no, the rating does not move. If the readout was due and has not appeared, note that too — quietly overdue is one of the most reliable signals available to an outside observer.

Worked demonstration — two frontier entries from this chapter

🔬 ENTRY — ORAL PEPTIDE DELIVERY as a general platform    [worked demonstration]
  Rating            ⚠️ for the approved specific case (oral semaglutide, ~1%
                    bioavailability, in use); 🔬 for the platform claim.
                    Two claims, two ratings, one technology (rule 6).
  Readout that      Approvals of enhancer-based oral peptides in OTHER classes,
  would move it     at bioavailability substantially above the current order of
                    magnitude. Watch the competing outcome too: an approved oral
                    small-molecule agonist would NOT raise this rating — it would
                    make the platform claim less important, which is a different
                    thing and must not be scored as support.
  Date created      2026
  Review by         2029
  Do NOT count      new enhancer chemistry papers; preclinical bioavailability
                    figures; partnership announcements; coverage of oral
                    semaglutide's existing approvals, which are already priced
                    into the ⚠️ and are not new evidence for the platform.

🔬 ENTRY — PEPTIDE-DRUG CONJUGATES as a general platform  [worked demonstration]
  Rating            ✅ for PRRT in somatostatin-receptor-positive neuroendocrine
                    tumors (approved, trialed, in use); 🔬 for the platform.
                    Keep the two lines visibly apart — the approval is the thing
                    most likely to be used to inflate the platform claim.
  Readout that      Approval in a DIFFERENT indication, with a DIFFERENT
  would move it     targeting peptide and ideally a different payload class.
                    A non-radioisotope payload counts for more, because a
                    radioisotope acts at a distance and needs no intracellular
                    release — the easier engineering problem.
  Date created      2026
  Review by         2029
  Do NOT count      additional radionuclide products using the same targeting
                    family; preclinical conjugate papers; imaging-agent
                    approvals; a company describing a "conjugate platform."

Notice the line doing the most work in both: Do NOT count. Naming in advance what will not move a rating is the practical form of rule 3, and it is far easier to write now — before you have any attachment to the outcome — than in 2029, with an exciting headline in front of you.

Do this now

Add two to four 🔬 entries from this chapter, choosing the frontier questions you actually care about. For each, write all four lines: rating, readout, date, and do not count.

Then audit the entries you already have. Any compound with a "coming soon" story attached should have that story run through §36.2's six questions and recorded: phase, endpoint, population, comparator, source, base rate. Some will not survive the exercise, and the ones that do will have shrunk.

Chapter 40 assembles the completed dossier. Chapters 37 and 38 continue Part VI. When you reach Chapter 40, the 🔬 entries you write today are the ones you will be gladdest you dated.

Conclusion

This chapter had a structural problem and said so in its first paragraph: you cannot rate claims against evidence that does not exist yet.

So it did two other things. It reported status — what is in development and at what stage, as of this writing in 2026, in a form you can check and I can be caught getting wrong. And it taught the reading skill, which is the part that lasts.

The six questions are the chapter. What phase, actually. What endpoint. What population. Against what comparator. Who is claiming, and when. And what is the base rate — which you know before reading anything, which is that most candidates fail, and which mechanism does not move.

The specific content will age. Oral delivery works at one percent through an absorption enhancer, and the winning oral drugs may not be peptides at all. Longer duration is a trade-off with receptor biology, not a ladder. The multi-agonist race is optimizing a surrogate. The muscle question is genuinely open and would be answered by measuring function. Conjugates are approved in one application and speculative as a platform. Acting on the brain and crossing the blood-brain barrier are different claims. Patent expiry will change access arguments without changing one piece of evidence about whether the drugs work.

Some of those sentences will be wrong within a few years. The chapter was built to survive that, because its content is the procedure rather than the conclusions.

And §36.10 turned the procedure on my own work, in specific and checkable terms, dated to 2026, so a reader in the 2030s can grade it. That is the difference between a prediction and a forecast: a forecast tells you what would prove it wrong.

It is also the difference between the claims this book rates ✅ and the claims it rates ❌. Not confidence — the ❌ claims are usually more confident. Not plausibility — the mechanism stories are often lovely. The difference is that one kind of claim specifies what would refute it and submits to the answer, and the other does not.

Chapter 37 continues Part VI. But carry one question out of this chapter into the rest of your life as a reader of this field: what would change your mind? The answers, and the silences, will sort your sources faster than any credential.

Key Terms

Preclinical — development before human administration for the indication in question: cells, tissue, animals. Not "almost in trials."

Phase 1 — first-in-human studies, usually small, primarily assessing safety, tolerability, and pharmacokinetics. Generally not designed to demonstrate efficacy.

Phase 2 — mid-stage trials assessing efficacy signals and dose, at modest sample size and usually on surrogate endpoints. Subject to phase 2 optimism (Chapter 9).

Phase 3 — large, adequately powered confirmatory trials, usually the basis for approval.

Base rate — the historical frequency of an outcome in a reference class. Here: the low proportion of compounds entering human trials that reach approval, serving as the prior any claim must move you off.

Topline results — a sponsor-selected summary of headline findings released before full data. Not a publication.

Surrogate endpoint — a measurement believed to predict a clinical outcome, used because it is faster or cheaper. Weight loss and lean mass on a scan are surrogates.

Hard outcome — an event a person would care about independently of any measurement: death, heart attack, hospitalization, a fall, loss of independence.

Active comparator — a control arm receiving a treatment already known to work, rather than placebo. The hardest and most informative comparison.

Absorption enhancer — an excipient co-formulated with a drug to increase absorption across an epithelial barrier. SNAC is the example in oral semaglutide.

SNAC — the absorption enhancer in oral semaglutide; transiently raises local pH and promotes gastric absorption.

Bioavailability — the fraction of an administered dose reaching systemic circulation intact. On the order of 1 percent for oral semaglutide.

Peptidomimetic — a molecule designed to reproduce a peptide's binding interaction without being a peptide. Captopril is the classic case (Chapter 35).

Small-molecule agonist — a low-molecular-weight non-peptide compound activating a receptor normally addressed by a peptide. Orforglipron is a GLP-1 receptor example.

Desensitization — reduced receptor responsiveness following sustained stimulation, on a fast timescale (Chapter 2 §2.8).

Downregulation — reduced receptor number following sustained stimulation, on a slower timescale. With desensitization, one of the two termination mechanisms engineering cannot defeat (Chapter 33 §33.10).

Multi-agonist — a single molecule activating two or more receptors: dual (GLP-1/GIP), triple (GLP-1/GIP/glucagon).

Lean mass — non-fat body mass as measured by body composition methods. Includes contractile muscle, fluid, connective tissue, and organ mass, which is why it is not a synonym for muscle.

Sarcopenia — age-associated loss of muscle mass and function; the concern that makes the lean mass question sharpest in older adults.

Myostatin — a negative regulator of skeletal muscle mass, signaling through activin type II receptors; blocking either increases muscle mass in animal models (Chapter 16).

DEXA — dual-energy X-ray absorptiometry, a body composition measurement. A surrogate, not a functional outcome.

Peptide-drug conjugate — a construct joining a targeting peptide, a linker, and a payload, using the peptide's receptor selectivity as a delivery address.

PRRT — peptide receptor radionuclide therapy: an approved conjugate approach using a somatostatin-analog peptide, a chelator, and a therapeutic radioisotope (Chapter 27).

Receptor-mediated transcytosis — active transport of cargo across the blood-brain barrier via an endothelial receptor, exploited by attaching a therapeutic to a ligand for it.

Circumventricular organ — a brain region with fenestrated capillaries and an incomplete blood-brain barrier, letting its neurons sample circulating signals directly. The area postrema is the example most relevant to appetite and nausea.

Vagal afferent — a sensory nerve fiber carrying information from the gut to the brain; a route by which a peptide's signal reaches the brain without the peptide doing so.

Fill-finish — sterile filling of drug product into its final container and device. A bottleneck distinct from synthesis, and a real cause of shortage.

Compound patent — a patent covering the molecule itself, as distinct from formulation, device, or use patents. Expiry dates differ by jurisdiction.

Generic entry — market entry of non-originator versions after patent expiry, changing price and access without changing efficacy evidence.

Falsifiability — the property of a claim that specifies what observation would show it false. Required by rating rule 5.

Forecast — a statement about the future that names what would prove it wrong and can therefore be graded. Distinguished from a prediction, which merely asserts.

Spaced Review

1. (Ch 36 §36.2) A press release states: "Our novel peptide showed a statistically significant improvement in patients with metabolic dysfunction." Run all six questions on that sentence. For each, state whether the release answers it and what the omission implies. Then decide what, if anything, you have learned about the compound.

2. (Ch 36 §36.6 + Ch 16) A trial reports that adding a myostatin-pathway agent to a GLP-1 drug preserves lean mass on DEXA, with a large and highly significant effect. Using Chapter 16's framework, explain why this does not establish that patients are better off, and name three measurements that would move the claim toward a functional conclusion. Why is this surrogate more treacherous than most?

3. (Ch 36 §36.3 + Ch 35) Chapter 35 described how a peptide from snake venom became captopril, a non-peptide oral drug. §36.3 describes orforglipron. State what is structurally the same about the two stories, then state what §36.3 concludes about the future of oral peptides. Is that conclusion a defeat for peptide science? Defend your answer.

4. (Ch 36 §36.5 + Ch 9 + Ch 10) A new triple agonist reports phase 2 weight loss substantially exceeding tirzepatide's approved-label results, and a colleague concludes it will be the better drug. Identify two separate errors: one from Chapter 9's phase 2 optimism, one from the surrogate-versus-outcome distinction that Chapter 10's SELECT trial illustrates. Then write the two ratings you would assign this compound — one for weight loss, one for cardiovascular outcomes — and justify the difference.

5. (Ch 36 dossier + Ch 6 + Ch 5) Pick one 🔬 entry from your dossier. Write the readout that would move it — naming population, endpoint, and comparator — then write the "do not count" line: three kinds of news you can already foresee that must not move the rating. Using Chapter 6's selection effect, explain why the second list is the one that protects you. Is what you wrote a prediction or a forecast?