The claim log: every dated claim, recorded once

Our pipelines publish dated claims: an earnings date confirmed from a company's filing, a CEO's share count from a Form 4 or a proxy, an AI explainer for one session's move. The pages that show them are rebuilt in place, so a corrected figure used to leave no trace. This log records each claim once, with the time it was recorded and its source. When a pipeline later publishes something different for the same claim, the change is a new entry that names the one it corrects. Nothing is edited in place, and each entry's SHA-256 digest folds in the one before it, so an edit anywhere changes every digest after it.

The count (read 2026-11-10)

93 entries in 1 monthly shard; 44 earnings date confirmed from a company filing, 36 ceo shares held, from a filing, 13 ai move explainer for a session. 2 corrections appended. 13 recorded on a later day than they were published (flagged late below); for 80 the first publish time was not stored by the pipeline, so no late flag can be computed and none is shown. Recomputed now: 93 of 93 digests match.

Corrections

• Entry 79 (2026-11-06) corrects entry 47: was “KINS reports on 2026-11-05 at 8:30 AM ET, per a filing dated 2026-10-16.”, now “KINS reports on 2026-11-06 at 8:30 AM ET, per a filing dated 2026-10-16.”
• Entry 80 (2026-11-06) corrects entry 50: was “LPG reports on 2026-11-05 at 10:00 AM ET, per a filing dated 2026-10-30.”, now “LPG reports on 2026-11-24 at 10:00 AM ET, per a filing dated 2026-10-30.”

The newest 93 entries

Every entry, with its digest, is in the JSON.

Check it yourself

Node 18 or later, no packages. It fetches the log, recomputes every digest in order and reports any entry that does not match. Save the JSON today and run it again later: an entry that was edited, dropped or reordered since will not recompute.

// verify-claim-log.mjs   run: node verify-claim-log.mjs
import { createHash } from 'node:crypto';

const log = await (await fetch('https://stocktools.ai/api/claim-log')).json();
let prev = log.genesis, bad = 0;
for (const e of log.entries) {
  const row = Object.fromEntries(log.committedFields.map((k) => [k, e[k]]));
  const d = createHash('sha256').update(`${prev}\n${JSON.stringify(row)}`).digest('hex');
  if (e.prevDigest !== prev || d !== e.digest) { bad++; console.log('MISMATCH at', e.seq); }
  prev = e.digest;
}
console.log(log.entries.length, 'entries,', bad, 'mismatched; head', prev === log.head ? 'matches' : 'DIFFERS');

What is logged and what is not

• Earnings dates confirmed from a company's own filing, one entry per filing; the same filing parsed to a different date later is a correction.
• CEO shares held, from the Form 4 or proxy the CEO page cites, one entry per filing. The dollar value moves with the close every day and is not a claim we correct, so it is not logged.
• AI move explainers, one entry per stored read, with the time it was written. Reads written before the log began carry a late flag.
• Not yet logged: projected earnings windows (estimates, labelled as such on the calendar) and figures recomputed on every page build. Our AI reads' own figures are checked on the AI figure check.

Keep going: the AI Answer Ledger's hash chain · the earnings calendar · CEO net worth · methodology