Claims audit · summary sheet

The June article, audited — and what CORRIDOR already fixed

A claims audit run in June 2026 on the Celaya article about the Meta agreement, plus a verification pass that resolved most of what it flagged. Shared in full because a proposal to collaborate on evidence standards should start by putting its own work on the table.

Subject Celaya on the Meta data center agreement Audit run 29 June 2026 Mode Autonomous — no author participation Instrument audit-claims v3 Claims assessed 25 after deduplication

Result, before and after the verification pass

56% → 76%Sourcing rate, after Meta's own statements were located

1 → 0High-risk flags remaining

80%Of verifiable claims checked out on first pass

0Defamation risk · meta-manipulation · fabrication

The finding was documentation, not accuracy. Where the article said Meta had made a commitment, Meta's own published statements confirmed it — in several cases verbatim. What was missing was the link, not the fact.

The single high-risk flag was the attribution "Meta has publicly said it now expects more" than fifty jobs — unsourced at the time of the audit, and therefore a legal exposure regardless of whether it was true. A follow-up pass against Meta's own documents found it was true: "we'll support more than 300 operational jobs at the data center once it is complete." The flag was correct and the claim was correct. Both things.

What held, and where the gaps were

Held up

Every public-record claim — the 5–3 vote, the named council positions, the intervention in the state review
All ten Meta commitments checked: 200% water restoration, closed-loop cooling, full cost of water and wastewater, the AguaCares grant, $10B, 1 GW, 100% renewable matching
The central argument. The 50-job contractual floor against the 300-job public promise is real, and the distinction between an enforceable cap and a public commitment is structurally sound
Technical competence throughout on the electrical and infrastructure material

Needed documentation

Contract characterizations asserted from a stated reading of the executed agreements, without quotes or links — the 80% abatement, the uncapped cash grant, the 10/15/5/35-year terms, the fee waivers, $9M in road reimbursement, 1,039 acres
Meta attributions given without date or link, though later confirmed
Self-reported credentials doing load-bearing work at the top of the piece
One absence claim — that no one had studied the three projects together — asserted without a stated search

The instrument does not distinguish a well-founded claim from a poorly founded one. It only asks whether a reader can check it without taking the author's word. That is the whole of the 44%.

What CORRIDOR appears to have already resolved

ClosedThe citation gap. CORRIDOR states that every load-bearing figure is verified against primary sources, and publishes each headline number with a copy-ready citation attached. That is a direct answer to the largest category in the audit.
ClosedOne-sided framing. The June piece was flagged for presupposition and for a "to be fair to Meta" construction that withdrew the credit it appeared to give. CORRIDOR states both the booster and the critic case at full strength, which is the structural fix, not a cosmetic one.
ClosedUndated claims. CORRIDOR carries an explicit currency date. Several June flags existed only because a status claim had no date attached.
OpenCity contract language. Meta's statements can source Meta's promises. They cannot source what the executed city agreements require — the abatement percentage, the cap language, the acreage, the reimbursement. Those still need direct quotes from the signed documents, and they are the claims carrying the remaining medium-risk flags.
ReconcileA water figure to align. The June article cited 2.5M gallons/day at full build-out, sourced to the city's data center page. CORRIDOR publishes 1.5M gallons/day permitted. Those are likely two different measures — permitted now versus maximum at full build-out — and probably both correct. Worth reconciling explicitly before either number appears in a jointly authored document.

Limits of this sheet

StatedThe audit was run in autonomous mode, without the author's participation, in June, before we had spoken. It was not commissioned and not adversarial — it was an instrument being exercised on a live local document.
StatedCORRIDOR itself has not been audited. The resolutions above are read from how CORRIDOR describes its own method, not from a claims audit of the field paper. That audit has not been run. It is available, and it would be more useful run reciprocally than in one direction.
StatedThe instrument was calibrated against its author's own byline first — 41 articles, 228 claims — before it was pointed at anyone else's work. That is the only reason this sheet is defensible to send.