Verify a decision

{"actor":"judge","decided_at":"2026-08-27T07:05:45.326Z","decision":"review_revise","investigation_id":"fc811077-cde8-4cc1-9c08-8b91657cbf92","new_score":43,"page_slug":"gmgn-ai","prev_score":58,"reason":"Independently computed disputed_pct is 18.5% (5 disputed of 27 claims), which falls in the minor-issues band, and the page's core factual scaffolding — founding date, team identity, fee structure, custody model, and general scam/impersonation pattern — held up under independent verification (claim_findings[0-5, 7-14, 17-23, 26]). The penalty sits at the top of that band because the disputed claims are not evenly minor: the page's own summary affirmatively states no breach of GMGN's own infrastructure or misappropriation of funds has been documented, and this is contradicted by well-corroborated October 2025 incidents — an exploit of GMGN's own anti-sandwich nodes and a separate ~107-user phishing wave, both compensated by the company (claim_findings[6]). Two coverage_gaps rated high priority reinforce that this is an omission of a checkable, sourced incident rather than a matter of interpretation. Separately, two timeline entries (claim_findings[15], [16], [24], [25]) date a fake-App-Store episode a full year off, based on the cited tweets' own Snowflake timestamps and a corroborating ChainCatcher article, which shifts an incident's apparent recency. These are date and omission problems rather than reversals of the page's core narrative, and the reviewer confirmed several of the page's more sensitive judgment calls were handled correctly (accurate AI-capability framing, correct non-speculative founder attribution, appropriately hedged unsourced regulatory-negative claim), so denial or a significant-issues escalation is not warranted.","score_delta":-15,"sequence_num":3,"submission_content_hash":null,"submission_id":null,"submission_kind":null,"submission_valence":null,"v":1}