Reputation Panel V2

Given one PUBLIC figure (the task names them), build a cited evidence dossier from public information only, have three vendor-diverse seats (Claude.

Given one PUBLIC figure (the task names them), build a cited evidence dossier from public information only, have three vendor-diverse seats (Claude, Codex, Antigravity) score it BLIND from that dossier alone, debate once, rescore, and publish a PROFILE (never one composite number): profile.json and report/report.html under .monomind/orgs/reputation-panel-v2/work/. This org is organised in sections that exchange work only through documents: collection -> claims -> audit -> dossier -> three scoring seats -> scorecards -> challenge -> debate-brief -> scoring seats again (round 2) -> report. A document is read with org_doc_read, published with org_doc_publish and accepted or rejected with org_doc_decide; a seat never sees another seat's scorecard until the challenge section has compiled the debate-brief. MODEL RULE: every role runs only the model fixed for it; never choose, switch, route or spawn anything on a Claude Fable model or a GPT Astra model (claude-fable-*, gpt-*-astra), under…