9 PHC — 9-Point Hallucination Calibration

9 PHC — 9-Point Hallucination Calibration A post-research audit program that runs every checkable claim in a finished piece of AI-assisted research through nine verification checks and returns each claim Verified, Corrected, or Held — unverified, with receipts. Paste in a finished draft or report. 9 PHC extracts every checkable claim and checks: source existence, source fidelity, alias sweep, right-level check, secondary sweep, own-files sweep, numbers recomputed three times in deterministic code, names and IDs letter-checked in deterministic code, and an adversarial pass in a fresh context whose job is to kill the claim. Sources are peer-reviewed or primary only; registries (Crossref, DataCite, Open Library, OpenAlex, Semantic Scholar) are read before any model judges anything. What was pre-existing: the idea only — conceived the week before the event. No code existed before the October 26 kickoff; everything shown was built during the event. Live record: scores 6/6 on the planted-defect sets across all three demonstration briefs; caught an invented citation and an invented tablet catalogue number cold on an unscripted brief; zero false corrections in its last five runs. Full record: docs/RUN-RECORD.md in the repo.