What is and isn't established about a commercial peptide catalogue
An evidence map of a class of widely-used compounds, graded on three independent axes, with every claim carrying a source and a retrieval status — and every gap stated rather than smoothed over.
01Scope corrections
The working assumptions going in were wrong in four ways. Recorded because scope discovered mid-project invalidates batching.
| Assumption | What the catalogue actually shows |
|---|---|
| ~40 catalogue items | 51 SKUs, resolving to 43 distinct compounds and blends. Under-counted by roughly a third. |
| A full Khavinson bioregulator line — Chonluten, Pinealon, Cortagen, Prostamax, Vilon, Thymalin | None of these are stocked. The Khavinson-type line is exactly three: Epitalon, Cartalax, Bronchogen. |
| Adamax and Bronchogen were missed by the prior pass | Correct. Both present, both added July 2026 per image upload paths. |
| — | 13 compounds went unanticipated entirely: LL-37, VIP, ARA-290, Snap-8, AHK-cu, AOD-9604, Thymosin α1, Survodutide, Sermorelin, Cagrilintide, DSIP, Melanotan I, Lipo-C. |
Semax itself is not sold — only N-acetyl semax amidate and Adamax, both analogues that inherit nothing. So FDA's Semax review, one of the seven, attaches to no catalogue item directly. That is the reverse of the assumption going in.
02Key findings so far
From one completed record and the catalogue enumeration. Ironic findings held to a higher evidentiary bar, not a lower one.
DSIP is emideltide — the one compound the committee rejectedverified
The catalogue sells DSIP 10mg at $178/kit. FDA's meeting page states that the substance it reviewed as Emideltide is "also referred to as delta sleeping inducing peptide (DSIP)."
So the only one of the seven the PCAC declined to recommend is stocked here under its non-INN name. A direct identity link from a primary source, not an inference.
FDA found no clinical studies of BPC-157 in tendonitisload-bearing
BPC-157 was nominated for ulcerative colitis, Crohn's, celiac and tendonitis. FDA evaluated only UC, stating in footnote 4:
"FDA did not evaluate the proposed uses of Crohn's disease, Celiac disease, and tendonitis because the nomination did not include sufficient information … In addition, FDA did not identify clinical studies using BPC-157 in these populations."
A dated, primary-source statement of absence covering the musculoskeletal indication that drives essentially all demand.
Popularity runs inverse to clinical footprintpartly corroborated
FDA's outsourcing-facility database recorded zero compounded BPC-157 products between January 2017 and June 2025. Over the same period, per a paper FDA quotes, June 2024 search volume hit an all-time high with more than 50 million tagged video views on each of YouTube and TikTok.
Exactly the kind of finding a writer is drawn to, so: the database figure is a direct regulatory record; the social figure is attributed to Vasireddi et al. 2025 which has not been opened. The regulatory half stands; the social half is provisional.
Every nomination was withdrawn before the meetingverified
All nominations — LDT Health Solutions on behalf of the International Peptide Society, and Wells Pharmacy Network — were withdrawn by the nominator before the July meeting. FDA proceeded with all seven evaluations anyway, on its own initiative.
Bears on Source Independence, and withdrawing a nomination is an unusual posture for a substance its proponents are said to believe in.
03Grade table
Three independent axes, never averaged. The Class · sold for column carries what the compound is actually used for anecdotally, against what was actually studied — the gap between those two is the highest-yield finding in the project. Click a column to sort.
| Compound | Class · sold for | Evidence | Indication | Independence | Flags |
|---|---|---|---|---|---|
| BPC-157open record ↗ | RepairTendon, ligament and joint healing; gut repair; "wolverine compound"FDA studied: ulcerative colitis. No clinical studies in tendonitis | L4 UC only L5 as marketed |
Mismatched | Concentrated | IDENTITYROUTE/DOSESELECTION |
| KPV | Repair · immuneGut inflammation, IBD, skin conditions, wound healingFDA studied: wound healing and inflammatory conditions — closest alignment of the seven | pending | — | — | |
| TB-500 / TB4 | RepairTendon and muscle injury recovery, flexibility, "stacked" with BPC-157FDA studied: wound healing — and reviewed "TB-500", while the vendor sells "TB4" | pending | — | — | IDENTITY |
| MOTS-c | MetabolicMitochondrial energy, fat loss, "exercise mimetic", enduranceFDA studied: obesity and osteoporosis — not energy or endurance | pending | — | — | |
| DSIP (emideltide) | SleepSleep onset and quality, insomniaFDA studied: opioid withdrawal, chronic insomnia, narcolepsy. Rejected by the committee | pending | — | — | |
| Epitalon | Khavinson · longevityTelomere lengthening, anti-aging, lifespan, pineal/sleepFDA studied: insomnia only — not longevity or telomeres | pending | — | — | |
| Semax (not sold) | NootropicFocus, cognition, neuroprotection, ADHDFDA studied: cerebral ischemia, migraine, trigeminal neuralgia. Not stocked — only analogues | pending | — | — | IDENTITY |
| N-Acetyl Semax Amidate | Nootropic · analogueFocus, cognition, mood, verbal fluencyInherits nothing from Semax — graded on its own evidence | pending | — | — | IDENTITY |
| N-Acetyl Selank Amidate | Nootropic · analogueAnxiety reduction, calm focus, stress toleranceInherits nothing from Selank | pending | — | — | IDENTITY |
| Adamax | Nootropic · analogueCognition, memory, learning — premium positioning at $388/kitIdentity unconfirmed. Adamantane-conjugated Semax-type, per vendor framing only | pending | — | — | IDENTITY |
| PDA | Repair · analogueMarketed as a more stable "BPC-157 upgrade" — tendon and gut healingInherits nothing from BPC-157. Vendor-stated identity only | pending | — | — | IDENTITY |
| TB500 Frag 17-23 | Repair · fragmentInjury recovery — sold as the "active fragment" of TB4Inherits nothing from thymosin β4 | pending | — | — | IDENTITY |
| LL-37 | Immune · antimicrobialChronic infection, gut biofilm, SIBO, "mold illness" | pending | — | — | |
| Cartalax | KhavinsonCartilage and joint support, connective tissue | pending | — | — | |
| Bronchogen | KhavinsonLung and respiratory function, chronic bronchitisSequence conflict unresolved: AEDL vs ADEL are different molecules | pending | — | — | IDENTITY |
| Tesamorelinopen record ↗ | GH-axis · approvedVisceral fat loss, body recomposition, anti-agingApproved for: HIV-associated lipodystrophy only. Label states "not indicated for weight loss management" | L1 HIV lipo L5 as marketed |
Mismatched | Concentrated | ROUTE/DOSESELECTION |
| PT-141open record ↗ | Sexual function · approvedLibido and arousal, both sexes, on-demandApproved for: HSDD in premenopausal women, 1.75mg/dose. Vial is 10mg. Desire moved; satisfying events did not | L1 HSDD L5 as marketed |
Mismatched | Concentrated | ROUTE/DOSESELECTION |
| Melanotan I | Melanocortin · approvedTanning, sun tolerance, photoprotectionApproved as: a subcutaneous implant for erythropoietic protoporphyria — not a vial | pending | — | — | |
| Melanotan II | MelanocortinTanning, libido, appetite suppressionHas adverse-event literature — gets a dedicated safety section | pending | — | — | |
| Thymosin α1 | Immune · approved abroadImmune support, chronic infection, "immune resilience", long COVIDApproved for: hepatitis B and as a vaccine adjuvant, ~35 countries, not the US | pending | — | — | |
| Sermorelin | GH-axis · withdrawnGH support, sleep quality, body composition, anti-agingUS approval withdrawn 2008. Was approved for paediatric GH deficiency | pending | — | — | |
| VIP | Immune · neuroCIRS, "mold illness", chronic inflammatory response, usually intranasalAviptadil approved for: pulmonary arterial hypertension (EU) — not CIRS | pending | — | — | |
| NXP-1P → semaglutide | Metabolic · approvedWeight lossIdentity inferred from image filename, not a COA. Trial evidence attaches to a GMP formulation a research vial does not inherit | pending | — | — | IDENTITY |
| NXP-2P → tirzepatide | Metabolic · approvedWeight lossIdentity inferred from image filename, not a COA | pending | — | — | IDENTITY |
| NXP-3P → retatrutide | Metabolic · investigationalWeight loss — the most aggressive of the incretin classIdentity inferred. Evidence is largely topline releases → grades will carry (provisional) |
pending | — | — | IDENTITY |
| NXP-ATP | UnknownUnstated. $440/kit — the second-priciest SKU in the catalogueNo identity evidence of any kind. Recorded as unresolved rather than guessed | pending | — | — | IDENTITY |
| Survodutide | Metabolic · investigationalWeight loss, metabolic liver disease | pending | — | — | |
| Cagrilintide | Metabolic · investigationalAppetite suppression, weight loss — usually stacked with a GLP-1Sponsor programme studies it in combination, not alone | pending | — | — | |
| ARA-290 | Neuro · investigationalSmall-fibre neuropathy, neuropathic pain, sarcoidosis | pending | — | — | |
| Ipamorelinopen record ↗ | GH-axisGH pulse, sleep quality, recovery, lean mass, anti-agingOnly clinical endpoint ever tested: post-operative ileus — 25.3 vs 32.6 h, p=0.15, not advanced. Zero trials in any marketed use | L2 GH release L5 as marketed |
Mismatched | Concentrated | ROUTE/DOSE |
| CJC-1295 no DAC | GH-axisGH pulse, body composition — usually stacked with ipamorelin | pending | — | — | |
| AOD-9604 | GH-axis · fragmentFat loss without GH side effectsFailed obesity trials exist — must appear in counterbalancing findings | pending | — | — | |
| GHK-Cu | CosmeticSkin firmness, collagen, hair growth, wound healing — topical and injectedReference case for a broken chain: biology is solid, molecule largely does not cross intact skin | pending | — | — | |
| AHK-Cu | CosmeticHair growth and follicle stimulation, topical | pending | — | — | |
| Snap-8 | CosmeticExpression-line softening — marketed as a topical "botox alternative" | pending | — | — | |
| NAD+ (injectable) | Non-peptideEnergy, "anti-aging", cognition, addiction recoveryPrecursor evidence (NR/NMN) is not injectable-NAD+ evidence — routinely merged, not the same claim | pending | — | — | |
| 5-Amino-1MQ | Non-peptideFat loss, metabolic rate — an NNMT inhibitor, taken orally or injected | pending | — | — | |
| Lipo-C with B12 | Non-peptide"Fat-burning" injection — methionine, inositol, choline, B12 | pending | — | — | |
| KLOW Blend | BlendSkin, hair, healing, inflammation — GHK-cu / BPC-157 / TB4 / KPVNever studied as a blend. Combination effects untested | L5 | Mismatched | — | |
| Glow Blend | BlendSkin and aesthetic "glow" — GHK-cu / TB4 / BPC-157Never studied as a blend | L5 | Mismatched | — | |
| Deadpool Blend | BlendInjury recovery and joints — BPC-157 / TB4 / CartalaxNever studied as a blend | L5 | Mismatched | — | |
| Beauty Blend | BlendSkin and anti-inflammatory — GHK-cu / KPVNever studied as a blend | L5 | Mismatched | — | |
| BPC-157 / TB4 | BlendThe standard "injury stack" — 5/5 and 10/10 strengthsThe FAERS hyperpigmentation case involved this exact combination | L5 | Mismatched | — | |
| Tesamorelin / Ipamorelin | BlendVisceral fat loss plus GH pulseNever studied as a blend | L5 | Mismatched | — | |
| CJC / Ipamorelin | BlendThe most common GH-axis stack — sleep, recovery, body compositionNever studied as a blend | L5 | Mismatched | — | |
| N-Acetyl Selank / Semax | BlendCalm focus — anxiolytic plus nootropicNever studied as a blend, and both components are analogues | L5 | Mismatched | — |
Showing 45 of 45 rows · 4 fully graded · 9 blends at protocol default
These are market facts, not endorsements — what the compound is marketed and discussed for, recorded so the Indication Match axis has something to compare against. For BPC-157 the claims are quoted from clinic and pharmacy pages collected by FDA. For the rest they are currently characterisations of the market positioning and get verified against vendor pages during each compound's Phase 2 pass (step 6 of the search order, market facts only). Until then treat the column as provisional.
04Catalogue inventory
Full pagination of the peptides, aminos and accessories categories, 8 Aug 2026. Prices are the vendor's displayed discounted price per 10-vial kit.
Approved drugs sold as research chemicals8 SKUs
| Product | mg | Kit | Compound | Identity |
|---|---|---|---|---|
| Tesamorelin | 10 | $288 | Tesamorelin (Egrifta) | verified |
| PT 141 | 10 | $168 | Bremelanotide (Vyleesi — 1.75mg/dose) | verified |
| Melanotan I | 10 | $188 | Afamelanotide (Scenesse — an implant) | verified |
| Thymosin Alpha 1 | 10 | $288 | Thymalfasin — approved ~35 countries, not US | verified |
| Sermorelin | 5 | $170 | Sermorelin — US approval withdrawn 2008 | verified |
| VIP | 10 | $338 | Aviptadil | verified |
| NXP-1P | 5–? | $143–198 | Semaglutide | inferred |
| NXP-2P | 30/60 | $168–488 | Tirzepatide | inferred |
NXP-1P, -2P, -3P carry no compound name anywhere on the product page. Identity leaks through image upload filenames — Sema-5mg, Tirz-30mg, Reta-12mg. A filename is not a certificate of analysis, so these stay inferred. NXP-ATP has no identity signal of any kind and is recorded as unresolved.
Investigational · repair · Khavinson · nootropic19 SKUs
| Product | mg | Kit | Compound | Note |
|---|---|---|---|---|
| NXP-3P | 12/24/48 | $198–828 | Retatrutide | inferred |
| Survodutide | 10 | $323 | BI 456906 | verified |
| Cagrilintide | 5 / 10 | $188 / $332 | Cagrilintide | verified |
| ARA-290 | 10 | $248 | Cibinetide | verified |
| BPC 157 | 10 | $152 | BPC-157 | PCAC-reviewed |
| TB4 | 10 | $251 | Thymosin β4, full-length | IDENTITY |
| TB500 Frag 17-23 | 10 | $161 | Tβ4 fragment 17–23 | IDENTITY |
| PDA | 10 | $134 | "Pentadeca-Arginate" | IDENTITY |
| KPV | 10 | $143 | KPV tripeptide | PCAC-reviewed |
| LL-37 | 10 | $378 | Cathelicidin LL-37 | |
| Epitalon | 10 | $143 | Epitalon (AEDG) | PCAC-reviewed |
| Cartalax | 10 | $161 | Cartalax (AED) | inferred |
| Bronchogen | 10 | $158 | Bronchogen tetrapeptide | IDENTITY |
| N-Acetyl Semax Amidate | 10 | $152 | Not Semax | IDENTITY |
| N-Acetyl Selank Amidate | 10 | $152 | Not Selank | IDENTITY |
| Adamax | 10 | $388 | Unconfirmed | IDENTITY |
| DSIP | 10 | $178 | = emideltide | PCAC-rejected |
TB4 vs TB-500. The vendor sells both as separate SKUs, implying TB4 is full-length thymosin β4. FDA reviewed "TB-500." If those differ, the FDA review does not cleanly attach to this SKU.
Bronchogen sequence. AEDL (Ala-Glu-Asp-Leu) vs Ala-Asp-Glu-Leu (= ADEL). Different molecules, not two spellings of one. Flagged, not resolved by preference.
GH-axis · melanocortin · cosmetic · non-peptide12 SKUs
| Product | Amount | Kit | Compound |
|---|---|---|---|
| Ipamorelin | 10mg | $178 | Ipamorelin |
| CJC-1295 no dac | 5mg | $170 | Mod-GRF(1-29) |
| AOD-9604 | 2mg | $125 | hGH fragment 176–191 |
| Melanotan II | 10mg | $188 | Melanotan II — adverse-event literature |
| GHK-cu lyophilized | 50mg | $125–170 | GHK-Cu |
| GHK-cu Cosmetic | 5 × 1g | $116 | GHK-Cu topical |
| AHK-cu Cosmetic | 5 × 1g | $170 | AHK-Cu topical |
| Snap-8 | 20mg | $198 | Acetyl octapeptide-3 |
| NAD+ Buffered | 250 / 500mg | $168 / $228 | A dinucleotide, not a peptide |
| 5-Amino-1MQ | 50mg | $198 | NNMT inhibitor |
| Lipo-C with B12 | 10 × 10mL | $188 | MIC + cyanocobalamin |
| NXP-ATP | 30mg | $440 | Unresolved |
Blends — all L5 / Mismatched by default9 SKUs
| Product | Composition as stated | Kit |
|---|---|---|
| KLOW Blend 80mg | GHK-cu 50 / BPC-157 10 / TB4 10 / KPV 10 | $458 |
| Glow Blend | GHK-cu 50 / TB4 10 / BPC-157 10 | $358 |
| Deadpool Blend | BPC-157 10 / TB4 10 / Cartalax 10 | $398 |
| Beauty Blend | GHK-cu 50 / KPV 20 | $288 |
| BPC-157 / TB4 | 5 / 5 · 10 / 10 | $188 · $323 |
| Tesamorelin / Ipamorelin | 10 / 3 | $328 |
| CJC-no DAC / Ipamorelin | 5 / 5 | $198 |
| N-Acetyl Selank / Semax | 10 / 10 | $238 |
Counts
- SKUs
- 51 enumerated
- Records needed
- 34 compounds + 9 blends = 43
- Identity
- 27 verified · 9 inferred · 1 unknown
- Standing flags
- 7 carry ⚠ IDENTITY
- PCAC coverage
- 7 compounds have an FDA evidence review
05FDA PCAC docket — FDA-2025-N-6895
Verified against primary sources. Agency-authored reviews of seven catalogue compounds, ~20 MB, free, written by reviewers with no commercial stake.
| Claim checked | Status |
|---|---|
| Docket FDA-2025-N-6895 exists | Confirmed — Federal Register 2026-07361 |
| PCAC met 23–24 July 2026 | Confirmed — FDA meeting page, updated 6 Aug 2026 |
| Covers all seven compounds | Confirmed — briefing document introduction |
| Reviewers had no commercial stake | Confirmed in structure — FDA proposed NOT to list all 14 substance forms |
FDA proposed rejection of every one of the seven — 14 substance forms, free base and acetate each. The committee then voted against the Agency's own reviewers on six of them.
Briefing documents
| Compound | Size | Extracted |
|---|---|---|
| Introduction | 268 KB | done |
| BPC-157 | 3.64 MB | done |
| Emideltide | 3.57 MB | pending |
| Epitalon | 2.97 MB | pending |
| KPV | 2.05 MB | pending |
| MOTS-c | 2.23 MB | pending |
| Semax | 2.94 MB | pending |
| TB-500 | 1.88 MB | pending |
06Vote record — a provenance problem
The most important epistemic caveat in the project so far.
FDA had not posted minutes or a transcript as of 6 Aug 2026, and the primary record is two YouTube webcasts that cannot be watched with available tools. Every tally below is ⚠ learned-from-summary and may not appear in any synthesis, table or summary until upgraded.
The figures came from the project brief rather than from a vendor. That does not change their status — same failure mode, no exception for trusted sources.
| Vote | Figure supplied | Independent corroboration |
|---|---|---|
| BPC-157 | 8–6 in favour | Yes — trade press |
| KPV | 8–6 in favour | Yes — trade press |
| TB-500 | 8–6 in favour | Yes — trade press |
| MOTS-c | 7–5 in favour | Yes — trade press |
| Emideltide | rejected 7–6 | Direction only; tally uncorroborated |
| Epitalon | 7–4 | Not corroborated |
| Semax | 8–5 | Not corroborated |
A PCAC recommendation is non-binding and is not evidence of effectiveness — FDA's own reviewers recommended against listing every one.
07Queue — 42 records pending
Batching order confirmed 8 Aug 2026: the seven PCAC compounds first, since the FDA reviews are located and they calibrate the rubric before harder cases.
| Block | Compounds | Blocked on |
|---|---|---|
| 1 — done | BPC-157 | — |
| 1b — done | PT-141, Tesamorelin, Ipamorelin — the approved-drug and GH-axis cluster, taken out of order | — |
| 2 — next | KPV, TB-500/TB4, MOTS-c | Briefing PDFs; ClinicalTrials.gov access |
| 3 | DSIP/emideltide, Epitalon, Semax | Semax record must state it is not sold |
| 4 | Identity cluster — PDA, TB500 Frag, N-acetyl Semax/Selank, Adamax, NXP-ATP | Janoshik COA retrieval would resolve several |
| 5 | Approved drugs sold as research chemicals | — |
| 6 | GH-axis, melanocortin (MT-II gets its own safety section), cosmetic | — |
| 7 | Non-peptides, Khavinson trio, 9 blends | Bronchogen sequence conflict |
| 8 | Audit — separate session, records-only context | All records complete |
08Method & rubric
Enforced search order
Not reorderable. A previous pass wrote this order then searched conversationally instead, which returned vendor SEO content and contaminated the dataset.
- ClinicalTrials.gov
- FDA PCAC 2026 docket extracts
- PubMed, Species: Human filter ON, then RCT / systematic review / meta-analysis filters
- PubMed, no species filter — only if 3 returns nothing
- PMC / DOI full-text resolution
- Vendor pages — last, market facts only, never for an evidence claim
Stopping rule. If 1–4 return no human study of any design, cap at L5, record the dated absence, move on.
Evidence Level
| L1 | Systematic review of RCTs, or multiple large consistent RCTs |
| L2 | ≥1 adequately powered RCT with a prespecified primary endpoint |
| L3 | Non-randomised controlled or cohort study |
| L4 | Case series, case-control, uncontrolled observational, or conference abstract |
| L5 | In vitro, animal, or mechanistic reasoning only |
(provisional) is appended where a grade rests on sponsor announcements rather than peer-reviewed publication.
Provenance — two fields on every claim
retrieval: retrieved | abstract-only | ⚠ learned-from-summary
⚠ learned-from-summary means the figure was encountered in someone else's description of a study that was not then opened. It may stay in the dataset; it may not be used in any synthesis until upgraded. This applies to figures supplied in the project brief itself.
Working rules
- Absence is a finding. "No registered trials as of <date>" is a result.
- Never smooth over a discrepancy. Flag and move on rather than resolving by preference.
- Watch for narrative pull. Ironic findings get a higher evidentiary bar.
- No recommendations. If a sentence would work as marketing copy, rewrite it.
09Limitations
What could not be reached, so the next pass knows where the soft spots are.
| Issue | Effect | Status |
|---|---|---|
| Literature retrieval — SOLVED | Direct PubMed page fetches hit a reCAPTCHA on the second consecutive request. NCBI E-utilities (esearch.fcgi / efetch.fcgi) returns clean abstracts with no rate-limiting and should be the default from now on. Five primary sources retrieved this way on 9 Aug 2026, clearing every outstanding efficacy figure in the Tesamorelin and Ipamorelin records | resolved |
| ClinicalTrials.gov unreachable | Returns an empty response body on both the v2 API and the legacy full_studies endpoint; study pages are client-rendered. Step 1 of the search order still cannot be executed directly. Workaround: where a trial is published, the paper carries its own NCT identifier and is a better source than the registry entry — that path resolved NCT00672074 and NCT00123253. It does not substitute for a current search of unpublished or ongoing trials, which remains undone | Open — partially mitigated |
| Corrections made 9 Aug 2026 | Three claims were wrong or over-stated and have been fixed in place, each with an inline note: (1) a safety statement attributed to Gobburu 1999 that the paper does not contain — struck; (2) tesamorelin glucose intolerance presented as an established liability when both pivotal trials found no significant glycaemic difference — now presents label warning and trial result together; (3) tesamorelin VAT "reverts toward baseline on withdrawal" — not supported by the retrieved abstract, withdrawn and reframed as an open question | remediated |
| Ipamorelin Phase 2 characterisation | Originally described as having "failed." The published result is a 7.3-hour median advantage at p=0.15 in 114 patients — underpowered rather than flatly negative. Reframed throughout | remediated |
| PCAC vote record is video-only | Primary record is two YouTube webcasts; no minutes as of 6 Aug 2026 | Open — tallies excluded |
| Large FDA PDFs exceed response limits | Each must be fetched to disk and read in chunks — slow for the remaining six | Managed |
Structural gaps in the evidence base itself
Findings about the field, not failures of this project.
- No observational tier. For most of this catalogue there is nothing between rodent studies and marketing copy — no registries, no cohorts, no post-marketing surveillance, because there is no marketing authorisation to surveil.
- FAERS undercounts structurally. 503A compounders generally do not report to FDA; research-chemical vendors are outside the system entirely. Counts are floors of unknown tightness, never rates.
- Certificates of analysis report purity only. Impurity profiles, aggregation state and endotoxin are typically untested — "99% pure" constrains far less than it appears to.
Each completed compound ships as a standalone HTML file in compounds/ — self-contained, no shared dependencies, so it can be saved or shared on its own. Markdown working copies, FDA source extracts and the protocol are in working/.
Records are kept as separate files because Phase 4 requires auditing in a session holding only the records — same-session auditing shares the writer's blind spots.
Protocol v1.0, frozen 8 August 2026.