The claim narrative was drafted by someone else's AI. Verify it before it becomes the decision.
AI now drafts the claimant narratives, IME summaries, provider reports, and demand letters that arrive in a comp file. A fabricated injury date, a misreported wage, or an invented prior claim in these documents does not read as an error. It reads as the file, and the decision built on it inherits the exposure under bad-faith and unfair-claims-practices law.
The errors sit in the specifics a comp board examines: an injury timeline that does not hold, an impairment rating that does not match the cited guides, wage and benefit math that does not compute, a prior-claim history the record contradicts. Asking another AI to check the draft is not a fix. In our pre-registered study of twenty company profiles, three frontier checkers disagreed by as much as 9.6 to 1 on the same document, and the harshest one returned zero errors on documents containing sixteen verifiable ones.
The AI is the witness. The math is the judge. GauntletScore verifies each claim against primary sources, then computes the trust score with deterministic math. The agents gather the evidence; no model sits in the verdict. The same evidence always produces the same score.
GauntletScore runs a dedicated analysis of cause-and-effect claims, testing injury-to-causation chains, treatment timelines, and apportionment reasoning. A narrative whose causal story fails the test counts heavily against the document, before a board or opposing counsel finds it.
Upload the document at gauntletscore.com. Minutes later you are reading the Gauntlet Report: a trust score with a 95% interval, every flagged claim with its verdict and the primary source behind it, and a cryptographically signed, tamper-evident certificate. The narrative that cannot survive the board does not get to set the reserve, and the decision that issues carries the audit trail that the file was independently checked. In a live production run, the engine examined nineteen claims in one fluent, credible document and returned two debunked against the court record, each with the source that contradicts it. Every voting agent independently recommended against proceeding.
LIMITSIt does not decide the claim and it does not verify what no source can confirm, including medical records it is not given. A claim it cannot check is reported as unverifiable, not as false.
Run the narrative and the IME summary against the record before they set the reserve or the strategy.
Run the determination and the letter against the record while the decision can still change.
Verify every citation, date, and calculation in the brief or report before it is submitted.
There is a narrative or an IME summary in the file right now that the decision is about to rest on. Run it. Either it holds, or it does not, and you found it before the board did.
Validation study pre-registered on the Open Science Framework, March 2026; in progress, manuscript in preparation.