Evidence Watch
Our source-aware daily AI evidence monitoring and versioning safeguards.
Evidence Watch follows the original question.
Ten tracked claims and twelve original-source URLs remain in the dated baseline watchlist. On 31 October 2026 we separately verified the integrity of all ten published snapshots and their public Hugging Face export. This technical check does not establish new external outcomes or independently re-test the underlying science.
All ten original claim questions were revisited through targeted source research. No new independently verified outcome or claim-status revision was justified. This was not a complete live HTTP check of every external link; the original source-linked baseline remains dated 25 October 2026.
Read the dated ten-case review, checks and limitations ↗Review known original sources and search for relevant new reports; preserve factual source dates and URLs.
Match a source to a precise tracked claim, eliminate duplicates and distinguish company assertions from external tests.
Record internal QA evidence; only material findings become append-only v2+ snapshots after checks.
No automatic certification.
An inaccessible source, missing independent test or unclear evidence stays pending. No system should silently mark a company promise fulfilled simply because related news appeared.
Public source watchlist →Monitoring baseline →Priority-driven checks without new infrastructure
The existing daily routine now tracks 20 source-backed claims. Two substantial primary-source checks per day give a nominal ten-day rotation, with upcoming commitment deadlines and oldest last-verified source checks first. Access blocks roll forward and are never recorded as successful verification. New baseline source readings on 5 November do not count as independent replications or historic scheduled QA checks.