Corrections

Entry price shown as $2.42/mo (readings dated 2026-07-05 through 2026-07-16)

Instrument error, our side. The price parser matched a dollar figure near the word 'annually' and divided it by 12. The bad readings are quarantined and the last verified price ($9/mo, read 2026-07-04) is restored with its date. A sanity gate now quarantines any overnight price swing above 60% that has no confirmed event on tape. The wrong readings stay in this ledger — that is the point of the ledger.

PAGE: /tools/photoai/

Entry price shown as $8/mo

Two errors in one cell. The $8 on file is BetterPic's per-edit add-on fee, not a package price — the instrument's context-free fallback grabbed the smallest dollar figure on the page. And BetterPic sells one-time packages (from $35, live check 2026-07-16), not a monthly subscription, so the '/mo' suffix was wrong too. Fixed everywhere: prices now render with their real billing model (one-time / monthly / usage), and the context-free fallback has been removed from the parser — a reading it can't justify is now an empty cell, not a guess.

PAGE: /tools/betterpic/

Entry price shown as $384/mo

The $384 was Chatbase's ANNUAL total for its Hobby plan ($32/mo billed annually), misread as a monthly price by an early parser version. Live pricing check 2026-07-16: Hobby $32/mo, Standard $120/mo, Pro $400/mo. The $99 readings that followed match neither and stay quarantined. Published price corrected to $32/mo with a dated, human-verified reading.

PAGE: /tools/chatbase/

Five traction figures contradicted by their own cited sources: OpusClip shown as $10.3M ARR / $68M raised (Sacra states $20M revenue 2025 / $50M raised); Submagic 4M+ users (source states 1M+); Aragon 2M users (founder interview states 1.3M+); Supermemory $3M raised (TechCrunch states $2.6M); BetterPic bootstrapped to $3M (announcement says $3.2M projected)

Found during a source-by-source audit on 2026-07-16: every cited source was re-fetched and the on-page figure compared against it. Where our roster disagreed with its own source, the source wins. All five figures now match what their sources actually say. This is what the source links on every page are for — they let you catch us.

PAGE: / (multiple boards)

Death rule stated as '3 unreachable days → confirmed dead, no exceptions'

That overstated the rule we actually run. The real rule, now stated everywhere: 3 consecutive silent days puts a board on the docket; a death is confirmed only after we rule out our own instrument (see the VOID ledger — it has two entries precisely because our instrument does fail). Days our instrument fails never advance anyone's clock.

PAGE: /docket/

CALL-0001 graded UNRESOLVED — instrument outage on verdict day (2026-07-27)

The desk's own network failed during the 2026-07-27 morning sweep, so no reading dated the deadline exists and the frozen default applied: UNRESOLVED. For the record, the bracketing readings: 29,343 stars on 2026-07-26 and 29,469 on 2026-07-28 (manual instrument run, same GitHub API source) — both above the 28,500 line. The desk does not claim the HIT; the rule is the rule, and the outage is ours. Filed so nobody has to wonder what the missing day would have shown.

PAGE: /calls/call-0001/

CALL-0004's frozen baseline — "No API pricing public as of 2026-07-13" — is not supported by the desk's own record

The call graded HIT, and the reading behind that is solid: on the deadline date Submagic's pricing page carried a plan named 'Business + API' at $41/member/mo and a public 'API Credits' pack table from $0.15/min down to $0.10/min. The problem is at the other end of the call. The frozen baseline says no API pricing was public on 2026-07-13, and the desk cannot show that. Our own pricing-page diff log has the line 'API & Integrations (10min/mo)' back on that page on 2026-07-11, and the earliest independent copy we can produce — a third-party archive dated 2026-07-23, seventeen days before the line — already carries the full credit-pack table. No archive of the page exists between the issue date and 2026-07-23, so what the page said on 2026-07-13 is UNKNOWN and stays UNKNOWN; the desk does not fill that gap with an inference. On the balance of what is on file, this call was more likely than not already true the day it was written. The verdict is not withdrawn — a frozen call is graded against its deadline reading and nothing else — but a HIT bought this way carries no forecasting value, and the track record should not be read as if it did. What went wrong here is ours and it is specific: the baseline was written from a homepage headline diff ('Submagic API is live', 2026-07-10) by someone who never opened the pricing page. The rule in calls.json now requires a manual-type call's baseline to name the page that was read and the date it was read. Separately on file: this call sat past its deadline for twenty days before anyone graded it. Manual-type calls are resolved in a weekly session, and that session did not happen between 2026-08-09 and 2026-08-29. A desk whose masthead reads 'graded by morning' left an overdue ticket on its own front page for three weeks.

PAGE: /calls/call-0004/