Evidence record · August 11, 2026
On 2026-08-11, I ran the claims gate against a public-facing technical page I had written. It returned VIOLATIONS-FOUND on three of my claims because the authored reference did not support them.
Recorded verdict
VIOLATIONS-FOUND
0 contradicts · 3 ungrounded
It did not mean the claims were proved false. It meant the named reference did not support them strongly enough for the gate to return a clean verdict. The report kept the claim and the relevant reference language side by side, then withheld PASS.
Held claim 1
“Ch. 4 p. 131 (SSTable immutability as the substrate for snapshot isolation)”
The verified SSTable range was approximately pages 119–129. Page 131 was outside that range, and the authored reference did not connect SSTable immutability to snapshot isolation.
Held claim 2
visible(entry) ≡ creator_seqno ≤ snapshot.seqno AND (deleter_seqno is absent OR deleter_seqno > snapshot.seqno)
The authored reference described visibility at version-set level, but did not define creator_seqno or deleter_seqno fields. The formula was a refinement the available canon did not pin.
Held claim 3
“Phantom reads” as part of the cited chapter’s taxonomy
The authored Snapshot Isolation section named lost update and write skew, but not phantom reads. The unsupported attribution was the problem.
One dated event: the claims gate was run against its maker’s own public-facing technical writing, returned a non-pass verdict, and identified three claims the authored reference did not support.
The source report is preserved byte for byte in CurioChat’s launch evidence store. This publication removes private repository routing and unrelated implementation detail while preserving the date, verdict, all three held claims, their dispositions, and the limits of what the run can support.