Not certified. An application for certification as an independent dispute resolution entity under 45 CFR 149.510 is in preparation. Not accepting disputes. This is the calibration passport kept for the federal independent dispute resolution process at 45 CFR 149.510 (the No Surprises Act). Every determination listed in this program's operating record is independently verifiable at `/verify/{determination_id}`. Aggregates below n=11 are suppressed; no party names appear anywhere on this page.
Cells are published once they clear the k-anonymity floor of 11; counters below read zero because the underlying count is below that floor.
13 record(s) held in the stores are excluded from every figure on this page because they are not production disputes: they were created by end-to-end tests of this system, are identified as such in the records themselves, and involve no real party. They remain on the append-only chain exactly as written and are excluded by an exact list of ids, never by a pattern.
Case-team personnel (attorneys and senior reviewers holding the required arbitration credentials) are engaged under contingent, per-case agreements, consistent with the independence requirements of 45 CFR 149.510(a)(2)(iv), (e)(2)(iv) and (e)(3). The roster below reflects the record of every determination signed: personnel are named, by name and credential, on the record of every determination they sign, and cells clear the k-anonymity floor of 11 before publication.
How this entity came to decide each dispute -- initiator-named (no objection), payer-named (the non-initiating party objected and proposed this entity, initiator agreed), joint (both parties agreed), or random (Departments' assignment).
Share of issued determinations whose signed rationale differs from the software's draft. A high share means reviewers are changing drafts; a low share is not evidence the drafts were right. The draft is a rule-based starting point computed from the record; reviewers are expected to depart from it whenever the evidence warrants, so this share measures reviewer engagement with the record, not the accuracy of the draft.
Of issued determinations where a reviewer recorded a blind pick before seeing any draft, how often that independent pick agreed with the software's draft -- an anchoring measurement distinct from 'reviewer edits to the draft' above, which only ever measures edits made after the draft was already seen.
A fixed one-in-N sample of issued determinations, chosen by hash so it cannot be steered, is re-read by a second credentialed reviewer who did not sign it. Agreement is whether the second reviewer would have selected the same offer; disagreement is recorded, not corrected, because an issued determination is final.
How predictable this entity's selections are from the offers and case context alone -- a fixed, pre-registered model (never fit on this entity's own data) scores each of this entity's own determinations and is compared against what this entity actually selected. Higher = more rule-like. This is consistency, not quality: a formula-driven entity that always sides with the smaller ask would score HIGH here.
Every record store is an append-only hash chain, verified end to end when this document was built. Per-store names, record counts and head hashes are deliberately NOT published: on a page rebuilt continuously they are a live event counter, and diffing them across builds reconstructs one dispute's timeline even though every aggregate on this page is suppressed. The single attestation below commits to every store's head at once.