Corrections, handled in the open.

How errors get found

Three ways, in rough order of frequency: our own scheduled re-verification passes (volatile claims — pricing, limits, legal statuses — are re-checked against official pages before any piece is republished); the automated checks that run on every build (every internal link must resolve; thin and duplicate pages fail the build); and readers, who remain the best error-finding system ever invented. If you find one, that email is the most valuable one we get.

What happens next

Substantive corrections — a wrong number, a broken method, a claim that no longer holds — are fixed in the text and noted on the page with the date and what changed. If a claim can no longer be verified at all, it is removed rather than softened into vagueness. Re-verifications (the claim still holds, re-checked on a later date) are noted the same way, because a date that silently refreshes is worth nothing. Wording fixes — typos, unclear phrasing — are fixed directly; they change nothing you'd rely on, so they don't get a stamp.

The receipts habit

Behind the published pages, every batch of work keeps a written record: which sources were checked live, which claims were gated and how they resolved, what was deliberately left out because the evidence didn't support it. When this desk's own numbers have changed under honest re-audit — and in the QuantLab project they have, dramatically — the change is the story, not an embarrassment: a backtest result that didn't survive its own entry-bar audit became the most useful article on the shelf.

The standard

No silent re-dating, no retro-fitting claims to what turned out true, no deleting the evidence of what we wrote before. The record is the product.