Participant and production builds are separate identities. Historical unbound runs never become current proof.
Methodology / R4 public route
Evidence before rank.No unexplained numbers.
A visible score is not enough. A publishable comparison binds the evaluated build, question version, participant output, time, actual cost, scorer, judge, rights and integrity manifest.
Method
Four independent gates.
Recognized ACTUAL cost and time are summed only from unique receipts. Missing totals remain UNKNOWN.
An AI judge is independent only when its identity, evidence and independence decision are recorded.
Rights, raw content, verification coverage and bundle integrity must all pass before evidence files are served.
State model
Publication states
The label describes evidence completeness, not model quality.
| State | Meaning |
|---|---|
| current-build verified | Exact evaluated build agrees with the running build and evidence gates pass. |
| historical unbound | Useful historical evidence, but no exact binding to the current production build. |
| unknown / incomplete | Required inputs are absent or conflicting. The portal does not infer zero. |