The claims register reports; it does not edit

What was decided

Roughly a hundred assertions rendered on this site are not true of this company. They are now written down — every one with a file, a line, a severity and the evidence that would make it true — in docs/legal/FLAGGED-CLAIMS.md. Not one of them was changed.

That restraint is the decision, and it is worth defending because the opposite instinct is strong. Half of these could be softened in an afternoon: "aligned" to "in progress", "41 Enterprise trade accounts" to nothing at all, "$697K+ ARR" deleted. It would feel like progress and it would be the wrong move three times over.

It would destroy the record. A survey is only useful if what it surveyed still exists. Edit first and the next reader sees a clean site and a memo describing a site that no longer matches it, with no way to tell which claims were removed because they were false and which were removed because they were awkward.

It would substitute my judgement for the owner's on questions that are not technical. Is the division seven years old or seven weeks? Is the audit-log retention 24 months or 13? Is the RTO sixty seconds or four hours? Fourteen of these (Part 4 of the memo) are two published sentences that disagree, and picking one is asserting a fact about the business. I do not have that fact.

And it would exceed the brief in the direction that is hardest to undo. PRD row 15 asks for a memo "with no factual claim altered by me", and §7 lists changes to claims about business history, customers or revenue as a non-goal. Both were written before the survey found how much there was to find, which makes them more binding rather than less — the constraint that only holds while the findings are small is not a constraint.

Reported, not ranked by embarrassment

Ordering is by consequence: what a regulator, a procurement reviewer or opposing counsel reaches for first. That puts the credit-backed 99.99% SLA at the top — an unobservable trigger on a financial commitment — and puts the ™ on the division badge near the bottom, though the second is more visible than the first.

Three severities, and "borderline" is not a synonym for "fine". It means the claim might survive with context the site does not currently give.

Two exits, and deletion is usually the worse one

Each entry carries what would substantiate it, because most of these have a fix that is not "remove the sentence". The SLA stops being a defect the day a monitor is connected — lib/uptime-monitor.ts is written and waiting on an account. "Entirely on AWS ca-central-1" stops being a defect when reworded to describe the Supabase region and the edge tier in front of it, which is a true sentence about a genuinely Canadian-resident database.

The exception is the security-review packet. A3, A8 and A9 promise to send documents — an engagement letter, a pentest summary, a control mapping. Those cannot be reworded into truth. Either they exist or the offer goes.

The three admissions

lib/changelog.ts:12-16 calls the release history illustrative. config/site.ts:216-221 concedes five of twelve homepage integration tiles have no written-up record. config/site.ts:448-450 concedes the two surviving homepage metrics are uncited.

All three are code comments. None is rendered. That gap is the memo's organising argument, because it is the difference between an oversight and a decision: a comment is a note to whoever edits the file, a disclosure is a note to whoever reads the page, and anyone reconstructing intent later will find that the codebase knew.

config/site.ts:448-450 also promised that its two metrics were "tracked as such in the flagged-claims memo" — a pointer to a file that did not exist until now. They are F1 and F2. The comment is true retroactively.

The directory drafts inherit a stricter rule

docs/directory-listings.md holds Crunchbase and Clutch copy and is submitted nowhere. The standard there is tighter than the site's: everything in it can be checked by a stranger with a browser, today, without asking us for anything — because a directory field is a *structured* claim that gets scraped, aggregated and quoted back years later with no surrounding paragraph to hedge it.

Applying that rule removed revenue, customer counts, every named operator, both compliance badges, the uptime figure, the two homepage metrics, the benchmark comparison, headcount, the residency claim and the 555 phone number. What is left is a smaller and duller company, which is the correct trade for this artefact.

The Clutch draft carries a recommendation against filing it yet, on structural grounds rather than honesty ones: Clutch is a verified-review marketplace, its registration flow asks for client references, and its ranking runs on review volume. A profile with no reviews is an empty storefront, which a buyer reads as a bad sign rather than a new one.

What would make this wrong

If the owner reads the memo and asks for the claims to be corrected, that is the next task and it is a large one. This decision is about sequence, not about whether the site should keep saying these things. Survey, then decide, then edit — in that order, with the memo as the record of what the site said on 3 August 2026.