Verify a decision

{"actor":"judge","artifact_identity":"507bfe14a2db888927664572d0e544db","band_override":null,"decided_at":"2026-09-22T22:28:32.977Z","decision":"review_revise","delist_requested":false,"disputed_pct":0.3077,"investigation_id":"52937998-58d6-49ab-a8d8-eba48078fd33","new_score":17,"page_slug":"deepseek","prev_score":32,"reason":"Of the 13 individual claims the reviewer checked, 4 came back disputed, which crosses the 30% threshold for this tier. All of the page's central allegations -- the exposed database (claim_findings[0]), government bans (claim_findings[1]-[2]), the OpenAI/Microsoft distillation allegation itself (claim_findings[3]), the Singapore Nvidia chip probe (claim_findings[6]-[7]), the censorship research (claim_findings[10]), and the fake-token scams (claim_findings[9]) -- were confirmed against cited sources. The disputed items are all dating errors, not disputes of substance: three of the four (claim_findings[3], [4], [5]) are the same underlying mistake -- the page states OpenAI's memo to Congress was submitted February 12, 2025, when the page's own cited source and independent reporting date it to February 12, 2026, a one-year transcription error appearing in both prose and the timeline. The fourth (claim_findings[6]) is a one-day discrepancy where a Singapore fraud-charge date was recorded as the source article's publish date rather than the charge date. No link rot, no disputed core allegations, and the reviewer rated its own confidence at 0.8. Given the crossing of the 30% line is driven by a single repeated date error being logged three times rather than three independent factual disputes, this is placed at the low end of the penalty range for this tier.","score_delta":-15,"sequence_num":2,"submission_content_hash":null,"submission_id":null,"submission_kind":null,"submission_valence":null,"v":1}