Peptide Record

Corrections and removal

If you are a vendor and a record about you is wrong

A transparency score is computed from inputs we have recorded, each linked to the document it was read from. The productive form of a dispute is therefore not "the score is unfair" but "this specific input is wrong, and here is the document that shows it".

Why contesting an input works and contesting the score does not

A transparency score is not a judgment about your business. It is arithmetic over five components — third-party testing, CoA quality, manufacturing transparency, business transparency, and regulatory record — each weighted by a published rule and each computed from inputs recorded against the document they were read from. Given the same inputs, the same score comes out, every time, for every seller. The rule, in full.

Two consequences follow, and they are the whole shape of this process. There is nothing to appeal in the weighting, because it is not applied to you specifically and there is no discretion in it to exercise. And there is something to correct in an input, because an input is a factual claim about a document that either holds or does not.

So the productive request names one input, says what we recorded, says what the document actually says, and links the document. Undisclosed data scores zero and displays as “Not disclosed” — never imputed — so if the answer is that a disclosure exists and we missed it, that is a correction we will make and the score recomputes from it.

38 vendor records are published on this site, and all of them are offered to search engines. If one of them names your company, there is a record here to correct and this is the route for it.

None of them carries a transparency score yet, so there is no score to contest today. What a published record currently asserts is narrower and worth stating plainly: that we hold this company in our vendor list, and where that came from. Several were drawn from FDA enforcement notices. If the entity is wrong — a misresolved name, a merger, two businesses collapsed into one, or a recall attributed to the wrong firm — that is a correction we will make now, before any score is computed on top of it.

What we will do

  • Correct a recorded input that a linked document does not support.
  • Record a disclosure you already published that we failed to find, and recompute from it.
  • Split or re-resolve an entity where we have merged your business with another, or attributed somebody else’s record to you.
  • Publish the correction against the source and collector that produced the record, so the change is traceable rather than silent.
  • Tell you the outcome either way, with the reason, and keep that reason on the record.

What we will not do

  • Adjust a score for any reason other than a factual error in an input. There is no editorial discretion in the arithmetic and no exception for a commercial relationship.
  • Remove a regulatory record — a warning letter, an import alert, a recall — because it is unflattering. If an agency published it, it is a public fact about your business.
  • Treat a disclosure you publish after we scored you as a correction. That is a re-score: recollect and the score recomputes from the new document.
  • Apply anything you ask for automatically. A request is recorded and reviewed by a person.
  • Verify that you are who you say you are. We cannot, so standing is recorded as asserted.

How long it takes

We aim to acknowledge a request within 5 working days.

We aim to reach a decision within 20 working days, and the decision is recorded with its reason whichever way it goes.

Both figures are targets set by one person with no team behind them, not commitments. Every request stores the date it arrived, so how long it actually took is a fact rather than a recollection.

Your request

Fixed choices, and one field for the specific claim

Submitting records a request. It does not change any record on this site. At most 3 undecided requests may be open against one record, and at most 5 may be sent from one address in 24 hours.

What this form accepts, and what it does not

The honest version
  • Bounded, because it is unauthenticated. The claim field takes at most 2,000 characters, every other field is capped, and the caps are enforced by the database rather than by this page. At most 3 undecided requests may be open against one record, at most 5 may come from one address in 24 hours, and at most 60 arrive in any 1-hour window across everybody.
  • The last of those can refuse a real request. It is a ceiling on rows per hour, so a flood spends it for everyone until the window rolls forward. That cost is accepted: an unbounded write into this database from the open internet is worse. Nothing is queued when it trips — you are told plainly, and nothing was stored.
  • The per-record limit is a queue depth, not a defence. Where you supply a record reference it is keyed on that exact row. Where you do not, it is keyed on the page URL you give us — normalised, so a query string or a trailing slash does not make a second record, but a different path does. Somebody determined to file more can, and what actually bounds that is the per-address and hourly limits above, neither of which depends on a value you choose.
  • No identity check, and none claimed. The relationship you state is recorded as stated. Your address is not verified. A request purporting to come from someone else is possible, which is one more reason a person decides.
  • No attachments. Link a document instead. We do not accept uploads, which removes a whole class of abuse from an unauthenticated form.
  • Your text is stored as text. It is never rendered as markup anywhere, on any surface. A correction request is attacker-authored by definition and is treated that way regardless of who sent it.

The whole corrections policy · What is still gated