Methodology
BuilderProof editorial team5 min read18 views

Publication dates corrected on four June 2026 benchmark pages

A date audit found four benchmark pages carrying a publication timestamp earlier than the registration of builderproof.org itself. The dates are corrected, the method is unchanged, and the two lookups that verify the claim are written out in full.

Updated on August 24, 2026

Parchment lab card showing a struck-through date corrected beneath it in teal, beside a timeline whose earliest tick falls before a marked boundary line
Parchment lab card showing a struck-through date corrected beneath it in teal, beside a timeline whose earliest tick falls before a marked boundary line
On this page

Quick answer. A date audit run on August 24, 2026 found that four June 2026 benchmark pages on this site carried a stored publication timestamp that predated the registration of builderproof.org itself. Those timestamps could not be true, they were an artifact of the launch content import rather than real publication dates, and they have been corrected to the earliest dates consistent with the evidence that survives. No benchmark text was changed, and the August 21, 2026 withdrawal notices on those pages stand exactly as published.

What the audit found

Every article page on this site emits an Article node in JSON-LD carrying a datePublished value, and that value is also what the byline renders. For four pages the stored value fell before 2026-06-16T15:10:24Z, which is the registration timestamp of the builderproof.org domain recorded in the registry.

A page cannot have been published on a domain that did not yet resolve. So the claim was falsifiable against a fact held outside this site, and it failed. The mechanism was mundane: the four pages were among the records loaded during the initial content import, and the import carried a nominal date field rather than a real publication event. Nobody re-checked those fields against the domain's own age, because no gate existed that compared the two.

That gate exists now and runs weekly against every property this lab publishes.

The four corrections

Scroll to see more

PageStored date, beforeCorrected date
Output quality benchmarkJune 3, 2026June 17, 2026
Speed to first paint benchmarkJune 9, 2026June 18, 2026
Deploy quality benchmarkJune 12, 2026June 18, 2026
Agency suitability benchmarkJune 16, 2026June 19, 2026

Each corrected page carries a dated note recording the change. The dateModified value on each now reflects the real timestamp of that edit, August 24, 2026, which is the first time any of the four has carried a truthful modification date.

How to verify this yourself

The correction rests on two lookups, both public, both free, and neither of them ours to control. A reader who wants to check the arithmetic rather than take the claim on trust can run them in about a minute.

Lookup one, the domain's age. Query the registry's RDAP endpoint for builderproof.org and read the entry in events whose eventAction is registration. RDAP is the successor to WHOIS and its response format is specified in RFC 9083. The registration event for this domain reads 2026-06-16T15:10:24.944Z.

Lookup two, the page's claimed date. Open any benchmark page on this site, read the JSON-LD block of type Article, and take its datePublished field. That is the value search engines and language models consume, and it is the value that was wrong.

If lookup two is earlier than lookup one on any page here, the claim is false and we would want to know. The same two lookups work against any publication, which is the point of writing the recipe down rather than simply asserting that the dates are now correct.

What this correction does not claim

The corrected dates are a floor, not a receipt. No publication log survives from the import, so the true publication moment of these four pages is not recoverable. What the new timestamps assert is the narrowest defensible thing: each page was published no earlier than the date now shown, that date falls after the domain existed, and the relative order of the four pages is unchanged from the order originally published.

This is weaker than a measured date and it is stated as weaker deliberately. A lab that has just withdrawn four result sets for resting on figures it could not reproduce should not replace one unreproducible claim with another.

Scope

Only the published_at field and a dated note were changed on the four pages listed above. The briefs, the rubrics, the protocol descriptions, the caveats and the reference lists are byte-for-byte unchanged. The August 21, 2026 withdrawal notices are unchanged and remain in force: the scored tables those four pages once carried are still withdrawn, and scored results for those axes will return only from a harness whose runs a reader can reproduce.

No other page on this site was in scope. The remaining published pages carry stored dates that already fall after the domain registration timestamp, which the audit confirmed by checking all of them rather than a sample.

References

  1. IETF. (2021). RFC 9083: JSON Responses for the Registration Data Access Protocol (RDAP). Specification for the events array and the registration event action used in lookup one.
  2. ICANN. (2026). Registration Data Lookup Tool. Public interface for the same registration data, for readers who prefer not to query RDAP directly.
  3. Schema.org. (2026). datePublished. Definition of the property whose value was incorrect on the four pages.
  4. Google Search Central. (2026). Article structured data. Reference for how datePublished and dateModified are consumed from an Article node.
  5. BuilderProof. (2026). How we benchmark AI app builders, methodology v1. The published method, unchanged by this correction.
  6. BuilderProof editorial team. (2026). Date integrity gate: comparing stored publication timestamps against domain registration. BuilderProof Methodology.

Cite this benchmark

Plain text
BuilderProof editorial team. "Publication dates corrected on four June 2026 benchmark pages". BuilderProof, August 2026. https://www.builderproof.org/benchmarks/publication-dates-corrected-four-june-2026-benchmark-pages.
BibTeX
@misc{builderproof-publication-dates-corrected-four-june-2026-benchmark-pages,
  title  = {{Publication dates corrected on four June 2026 benchmark pages}},
  author = {{BuilderProof editorial team}},
  year   = {2026},
  month  = {aug},
  howpublished = {\url{https://www.builderproof.org/benchmarks/publication-dates-corrected-four-june-2026-benchmark-pages}},
  note   = {BuilderProof, builderproof.org}
}

Frequently asked questions

Why were the dates wrong in the first place?

The four pages were loaded during the initial content import, and the import carried a nominal date field rather than a real publication event. No gate compared those fields against the age of the domain, so nothing surfaced the contradiction until the audit of August 24, 2026.

How can a publication date be proved false?

By a fact held outside the site. The registry records when builderproof.org was registered, and that record is not ours to edit. A page claiming to have been published before its own domain resolved is making a claim the registry contradicts.

Are the new dates the real publication dates?

No, and they are not presented as such. No publication log survives from the import, so the true moment is unrecoverable. The new dates are the earliest values consistent with the surviving evidence, and they preserve the original relative order of the four pages.

Does this change any benchmark result?

No. Only the stored publication date and a dated note were changed. The briefs, rubrics, protocols, caveats and reference lists are unchanged, and the August 21, 2026 withdrawal notices on those four pages remain in force.

Were any other pages affected?

No. Every published page on this site was checked, not a sample, and the remaining pages carry stored dates that already fall after the domain registration timestamp.