Errata

The record of when we were wrong.

Every correction Oquendo & Company makes to a published benchmark or Reading is entered here and kept, with the original conclusion left legible beside it. A house that publishes only its conclusions is asking to be taken on faith. This page is the other half of the arrangement.

The RecordAs of January 1970

0 Errata published Corrections to published work, all time.

0 Falsifiers settled Named in advance; settle-by date reached; outcome recorded.

0 Caught by a falsifier Errata found by an observable we named before the answer was known.

0 Settlements overdue Settle-by date passed with nothing recorded. This figure must read zero.

The second figure is what gives the first its meaning. A list of corrections with no count of what was actually tested can be curated into any shape its author prefers; a correction count read against a settled count cannot. Both are published here, and both move on their own.

There are no errata today, and that is not a claim of accuracy.

No engine has yet completed a forward grading cycle. No published conclusion has reached the date on which the physical world settles it, so no pre-registered falsifier has come due, and nothing has yet been put in a position to be found wrong.

An empty record at this stage means the record is young. It does not mean the benchmark has been tested and held. Those are different statements, and only one of them is available to us today. If this page said nothing at all, it would imply the first; saying so plainly is the only honest use of an empty page.

The figure to watch is the denominator. When settled falsifiers accumulate and errata stay few, that ratio will be worth something. It is worth nothing yet, and we would rather say so here than let its absence do the work of a claim.

What an erratum records

Each entry names the publication and the date it spoke as of; what was wrong with it, stated against the original conclusion in its own terms; what changed in the evidence or in our reading of it; the corrected position as it now stands; and what reliance the correction bears on. Entries are never renumbered and never withdrawn — a record that can be tidied is not a record.

How it was found

Every entry also states how the error came to light, because a house that corrects only what it happened to notice is grading its own paper. Where a conclusion was put at risk in advance — an observable named at issue, before the answer was known — the entry says so and cites it. Where it was not, the entry says that instead. The distinction is the whole value of the field, so it is not softened.

Nothing settles quietly

A falsifier named in advance carries the date by which its observable would have to be visible. When that date passes, the outcome is recorded either way — upheld or broken — and the count above moves. A settlement that simply never happens would let this page be curated by omission, so a settle-by date passing with nothing recorded against it stops our own release. The overdue figure above is the one that would show it, and it is published for exactly that reason.

Its other half

This page records where a published conclusion turned out to be wrong. The prediction record holds the conclusions themselves — what was claimed, the base rate it had to beat, the date it settles and the body that settles it — entered at issue and fixed there. Read together they are one instrument: predictions that cannot be revised, and corrections that cannot be omitted.

What this page is not

It is not a monitoring service. A published reading states a calibrated opinion as of its date; Oquendo & Company does not undertake to watch the world on your behalf, and an entry here creates no duty owed to anyone who relied on the original. It is a record kept because a benchmark that cannot show its corrections has not earned the citation. Terms of access are set out under Access; the standing this record supports is described under About.