Inspect the recorded questions, answers, findings, and evaluation scope. Download the complete sample and check its signed files without a VeriTrooper account, installation, or network connection.
These reports come from the 26 August 2026 Scout run of gpt-5.6-sol against public SEC 10-K material: 88.08% with VeriTrooper versus a 72.89% vanilla-RAG baseline across 889 scored questions. This is a scoped result, not a general model rating.
The ZIP includes seven customer-facing reports, a minimized run record, three reviewer aids, a signed manifest, signer certificate, fingerprint, and standalone verification tools. It omits credentials, protected implementation material, full configuration, and internal logs.
Integrity verification proves that the listed files match the signed manifest. It does not prove substantive correctness or compliance. Confirm the published fingerprint through a separate trusted channel.
Our evaluation archive preserves months of results, failures, corrections, and retesting. Two ten-run studies use a 238-question panel drawn from the Uniform Code of Military Justice (UCMJ). The study traces each result to its saved run record and explains the scoring basis.
The same recorded pipeline verdict appears in all ten runs for 236 question IDs, or 99.16% of the panel.
The original study records ten runs, with five on each of two reported hosts.
These are internal evaluations of a fixed question panel under recorded conditions. The PDF separates the two studies and includes all twenty run references, the July correction history, and the limits of the findings. Repeatability alone does not establish correctness or performance on other workloads.
3 pages · Prepared 22 September 2026 · Supporting preserved records can be discussed during an evidence review.
The 10-business-day assessment uses your approved sources, boundary, and acceptance test.
