Eight articles carried a publication date from before this domain was registered

The measurement

llm-tier.com was registered on August 19, 2026 at 10:10:53 UTC. That is a matter of public record and takes one lookup to verify.

Eight of the fourteen articles on this site published a date earlier than that. The earliest read August 3, 2026, sixteen days before the domain resolved. A ninth surface was wrong the same way: the last updated date, which we emit as dateModified in structured data, read August 19, 2026 at 09:30 UTC on all eight, forty minutes before registration.

Both values were crawler visible. Both were wrong.

Why it happened

The launch archive was seeded with dates spaced backwards to give the board an apparent history. No measurement, no run and no figure depended on those dates. They were presentation, and they were false.

What changed

All eight publication dates now sit on or after August 19, 2026, spread across August 19 and August 20 at distinct times rather than collapsed onto one timestamp. Four fall on each day. The last updated date on those pages now reads the timestamp of this correction pass, which is what it means.

Each of the eight carries a dated correction note at the foot of its own page.

One further stamp was corrected. Our own price correction article, published yesterday, was dated four hours before it was written. Backdating a correction notice understates how long the error it describes was live, which is the opposite of what a correction is for. It now reads the hour it actually went up.

What did not change

No index score, no axis subscore, no weight, no run duration and no roster entry moved. The runs data at /api/runs.json is byte for byte what it was yesterday.

The gate, stated so you can run it on us

For every published article, assert that its publication date and its last updated date are both at or after the registration date of the domain it is served from. Separately, flag any article whose publication date sits more than twenty four hours before the record was created.

We were already sweeping forward for articles dated in the future. That sweep has never found anything, because the error does not live in that direction. The backward comparison, against a fact that sits outside this site entirely, is the one that found this, and it is the one nobody was running.

Correction

September 9, 2026. The gate published above is correct and we still run it, but it is scoped to the articles table, and that scope is narrower than the problem it was written for. On September 9 we found the same defect in five further tables: 957 reader scores, open ratings, lab notes, comments and votes, every one of them dated before this domain resolved. The article dates corrected here were not the whole of it, and this gate returned zero throughout because it was never looking there. Those 957 rows are now corrected and the gate is restated wider in 957 reader rows carried a date from before this domain existed. Nothing in the article above has changed.