What has cleared publication, and what was waived to get here
84 of 84 graded peptide-outcome pairs are published and reachable, and all 84 of the published ones are offered to search engines along with every peptide and explore page. 38 vendor records are published and reachable, and all of them are offered to search engines — a decision argued below. Two launch gates were waived rather than met, and this page names them, names who waived them, and states what is still unmet about each.
State
Read from the corpus on this request- Peptides 10
- Graded pairs 84
- Published 84
- Graded, withheld 0
- Vendors published 38
- Gates waived 2
- Gates outstanding 0
- Correction requests open 0
- Intake this hour 0/60
| Authorised by | Pairs | From |
|---|---|---|
| Duncan Lockard (founder) | 84 | 2026-07-30 |
What publication did and did not change. Every grade you can see was computed from recorded inputs by the published rule, and every citation links to its source. That was true before publication and is unchanged by it. What publication changed is which records a reader can reach and which a search engine may index — not what any record says about itself. The standing disclosure on every record, the statement that judgment fields are extracted by agents and are not human-audited, and every stated limitation on the methodology pages are all exactly as they were.
No human has audited the corpus record by record. Publication is an editorial decision that these records may be seen, made by the person named above. It is not a claim that each one has been checked by hand. The two gates that would have asserted that — citation accuracy and stance accuracy on manual audit — were removed from the requirements long before this and replaced with a published disagreement rate, which is read live and says plainly that no person has verified the judgment fields.
“Published” is now a control rather than a label. Until this change nothing filtered on it: all 84 graded pairs rendered identically whether the flag was true or false, so this page could report a count of zero beside a site serving every record. The record page and the peptide record now filter on it inside the query. A withheld pair returns 404, is absent from its peptide’s outcome list, and is absent from the sitemap — which is what makes the zero above a fact about the site rather than a note about the database.
Launch gates
9 met · 2 waived · 0 outstanding2 gates were waived, not met
A waived gate is a gate the founder decided not to wait for. It is not a gate that passed, and the table below never says it is: the status column reads waived, the note column still carries the live measurement showing what is unmet, and the status cell names the person who waived it and the date. Every one of those numbers is queried on this request, so a waiver cannot suppress a measurement and cannot survive the measurement passing — a gate whose underlying check starts holding reads met on its own.
Nothing in the code that produces this table can turn a waiver into met. Waivers are applied after every status is computed and can only move a gate from outstanding to waived.
| Gate | Status | Note — measured on this request |
|---|---|---|
| Methodology pages published and linked from every score | met | — |
| Stated-limitation copy live | met | — |
| Standing disclaimers on every peptide and vendor record | met | — |
| Affiliate disclosure at every link | met | Component built; no vendor on the site has an affiliate relationship, so no paid link exists to disclose. |
| Adverse-event routing tested end to end | met | — |
| Every alias on every peptide accounted for | waived waived by Duncan Lockard (founder) on 2026-07-29 | 9 of 10 peptides have a fully accounted-for alias set. "Accounted for" means each alias is attested in the literature, appears in a regulatory registry, or is a punctuation variant of an alias that is — not that a person has read a paper for every one. Outstanding: CJC-1295. Still unmet. Authorised publication with one peptide’s alias set incomplete. CJC-1295 already carries a published alias caveat and its DAC/non-DAC hazard is rendered on the record itself, so the unaccounted names are disclosed on the page a reader lands on rather than only here. |
| Vendor entity resolution signed off before any transparency score publishes | met | 38 signed off, 0 pending, 0 refused, across 38 recorded decisions each naming who made it and why. No transparency score has been computed for any of them, so nothing this gate governs has published yet: scoring needs documents fetched from the vendor, which is a later surface. Every vendor page is nonetheless offered to a search engine, by a founder decision recorded against the argument put to it — see the indexing band below. |
| Legal review of vendor scoring, mention attribution, and submission handling | met | Completed by the founder, 2026-07-29. Covers the corpus and surfaces as they stood on that date; material changes to vendor scoring or mention attribution warrant a fresh look. |
| Backfill uniformity confirmed across peptides published with trending | waived waived by Duncan Lockard (founder) on 2026-07-29 | 0 of 10 peptides have a uniform 12-month backfill; 10 are suppressed for partial baseline. GDELT and the Reddit archive returned a recent suffix rather than the full year, and YouTube search cannot be backfilled uniformly within its quota at all. No trending signal displays for any peptide as a result, so the unmet baseline cannot reach a reader as a wrong number. Still unmet. Authorised publication with trending suppressed for every peptide. The gate governs publishing a trending signal, and no signal is published, so the unmet backfill cannot reach a reader as a wrong number. It stays unmet and trending stays withheld until the archives deepen. |
| Vendor correction and right-of-reply process live | met | Live: /corrections/vendor renders and the request table accepted a probe insert, rolled back. 0 open requests on this route. A request is recorded and reviewed; nothing on the public path can change a published record. Measured structurally — queue volume is reported under State, and does not move this status. |
| Voice correction and removal process live | met | Live: /corrections/voice renders and the request table accepted a probe insert, rolled back. 0 open requests on this route. A request is recorded and reviewed; nothing on the public path can change a published record. Measured structurally — queue volume is reported under State, and does not move this status. |
| Basis | Aliases |
|---|---|
| literature | 72 |
| registry | 11 |
| unattested | 6 |
| variant | 4 |
“Accounted for” is a weaker claim than “verified against the literature”, and the breakdown above is why the distinction is published rather than folded into a status word. An alias attested by a paper has been shown to name this compound. An alias accounted for as a variant differs from an attested one only by punctuation. An unattested alias is one nothing has yet confirmed, and it is the case that matters: an alias naming a different compound puts that compound’s studies behind this peptide’s grade.
Two gates in the original requirements were removed rather than met: citation accuracy on manual audit, and mention stance accuracy on manual audit. They were replaced by multi-pass extraction with the disagreement rate published live. That page states plainly that no human has verified the judgment fields, which is the point of publishing it.
The two correction gates are measured, not declared. A process counts as live when its route renders and the request table accepts a probe insert, which is rolled back — so the row you are reading was produced by trying it on this request rather than by anyone remembering to update a table. Vendor right of reply · Correction or removal of a mention. Both record a request; neither applies one. Nothing reachable without the console secret can change a grade, a transparency score, a mention’s stance, or delete a row, and that is asserted end to end against this corpus rather than described here.
Indexing
122 records offered · 0 withheld- Outcome records indexed 84
- Outcome records noindex 0
- Vendor records indexed 38
- Vendor records noindex 0
- Below the completeness threshold 38
The completeness threshold is measured and not currently enforced
The PRD requires that records below a completeness threshold are noindex. On 2026-07-30 the founder decided to index all corpus content instead, so a published record is offered to a search engine whether or not it clears that bar. The threshold is still computed on every request and its output is the figure above, because a gate that has been relaxed and a gate that has been deleted look identical unless the site keeps reporting what it has stopped enforcing.
It costs nothing today: every published record clears the threshold on its own, so the decision changes no page’s status at the moment it was made. What it changes is what happens to the next record that does not clear it — that record will be indexed, and this figure will say so.
Vendor records are indexed, over a recorded objection
All 38 of them are offered to search engines. That was a founder decision taken on 2026-07-30, after the case against it was put explicitly. The case is kept here rather than deleted, because it is what a future reader needs if one of these companies ever objects.
The legal review recorded above was completed on 2026-07-29, when the corpus held no vendor records at all. Its own note says it covers the corpus as it stood on that date and that material changes warrant a fresh look. Indexing pages that name real companies — including manufacturers and compounding pharmacies read out of FDA recall notices — is materially outside what was reviewed.
There is also a difference in kind. Every other page here is our own computation over public literature, shown with its inputs. A vendor page with no transparency score, no product listings and, for most of them, no website asserts little more than this company is in our vendor list, sourced from a recall notice. That is a statement about a named company rather than about a body of evidence, and a cached search result is not something we can retract.
Both points were stated before the decision and neither was resolved by it — they were accepted as costs. A fresh legal look at the vendor layer is outstanding, and any company that believes its record is wrong can contest it here.
| Shortfall | Records |
|---|---|
| no transparency score computed yet | 38 |
The site-wide exclusion is gone. Search engines were previously refused everything, by a noindex in the root layout and a blanket disallow in /robots.txt. Both are removed. /robots.txt now permits the corpus and names this site’s sitemap, which lists exactly what the pages themselves declare indexable — so a crawler is never sent to a URL we then tell it to ignore.
What the threshold measures. A record clears it when it has a grade, a stored decomposition that can be shown, at least one cited source that actually contributed to that grade, and at least one of those sources the reader can follow to the document. Membership is the same definition the grader runs over, not a second one written for this purpose. Nothing about the letter is part of the test: a D counts exactly as an A does, because the honest distribution is the product and a threshold that quietly preferred the flattering grades would misrepresent the corpus.
Why it exists at all. The requirement is one sentence in the design spec and the reason is the next one: the first indexed pages set this domain’s baseline quality signal in a category that is scrutinised heavily. A thin page is not neutral there — it is a sample of the site, and the cost lands on the records that are complete. That reasoning is the case against the current setting, and it is stated here rather than left out because the founder’s decision to index everything was taken against it, not in ignorance of it.
Vendor scoring is what would change the vendor position. A transparency score is computed from documents fetched from the vendor, which is a separate later surface; until it exists, a vendor page is reachable here and offered to nobody. How vendors will be scored
Removing the exclusion was an editorial decision, not a deployment step — and it changed only what a crawler may read. Nothing on any record was reworded to make it more presentable, no limitation was softened, and no absence was filled with a guess.