Research transparency

Independent Score Replication

The protocol is prepared. An independent external replication has not yet been completed.

Current status

Prepared — not executed

No independent result yet

USD Impact has published a reviewer protocol that fixes the independence requirements, eligible release boundary, frozen artifacts, recomputation steps, finding classes, numerical tolerance and claim limits before a reviewer result exists.

First eligible release: week ending August 28, 2026 or later, after the strict reproduction bundle and successful post-merge reproduction attestation exist.

Open protocol JSONRead reviewer protocolOpen execution tracker
Why the review waits

Use a strict frozen release, not a moving reconstruction

Releases through August 21, 2026 predate the mandatory immutable reproduction-bundle contract. Starting with the August 28 boundary, a valid strict release must preserve the weekly calculation inputs and normalization moments needed for offline reproduction, bind the dependency lock, retain dated/latest bundle identity, and pass post-merge attestation.

Waiting for that boundary prevents a reviewer from being asked to reproduce a historical value against market-provider histories that may have changed after publication.

Independence gate

What counts as independent

  • The reviewer did not design Score v2 or select its variables, signs, weights or regime thresholds.
  • The reviewer did not implement the production pipeline.
  • The primary recomputation runs in an environment controlled by the reviewer.
  • Compensation and material conflicts are disclosed.
  • The reviewer remains free to publish negative, ambiguous or not-testable findings.
Not independent evidence

First-party controls stay first-party

  • USD Impact CI and security checks.
  • The non-publishing reproduction rehearsal.
  • The strict release validator.
  • Pre-merge and post-merge reproduction attestations.
  • Any result produced only by USD Impact production code.
Reviewer test

What the external reviewer must recompute

1Freeze the exact production commit and hash the reviewed public artifacts.
2Confirm latest and dated strict bundles are identical and the dependency-lock hash matches.
3Implement the disclosed arithmetic independently rather than importing the production Score code.
4Recompute all eight z-scores, clipping, signed weights and contributions.
5Recompute the total Score and regime against the fixed numerical tolerance.
6Report methodology/data-semantics ambiguity and any evidence that cannot be tested.
Required findings

No cosmetic pass/fail

MATCHIndependent result agrees within tolerance and no material ambiguity blocks the check.
MISMATCHIndependent result diverges outside tolerance; both values and the smallest known divergence remain visible.
AMBIGUOUSThe public specification permits materially different reasonable implementations.
NOT TESTABLEThe required public/frozen evidence is insufficient; the missing evidence must be named.
Claim boundary

Replication is not predictive validation

Until a completed external report exists, USD Impact claims only that an independent replication protocol is prepared. It does not claim independent validation, audit, endorsement or verified model performance.

A future MATCH can support a narrow reproducibility statement for the release actually reviewed. It would not establish predictive power, future returns or trading value. Predictive evidence remains governed by the separate preregistered prospective Score v2 study.

The strict bundle freezes the calculation fields required for release arithmetic and hashes broader provider-derived histories. Complete original Yahoo/FRED response payloads and full provider-derived histories are not publicly redistributed, so the planned review is not described as a complete reconstruction from original transport bytes.

Academic evidence mapScore methodologyAbout & transparency