{
  "version": 1,
  "as_of": "2026-08-26",
  "title": "USD Impact Score v2 Independent Replication Protocol",
  "status": "prepared_not_executed",
  "independent_review_completed": false,
  "independent_validation_claim_allowed": false,
  "first_eligible_release_week": "2026-08-28",
  "scope": "A protocol for an external reviewer to test whether one strict as-published USD Impact Score v2 release can be reproduced from frozen public artifacts and whether the public methodology is sufficiently clear to implement independently. It is not a predictive-performance audit and it does not itself constitute independent review.",
  "why_not_current_release": "Releases through 2026-08-21 predate the mandatory immutable reproduction-bundle contract. The first eligible release is 2026-08-28 or later after successful post-merge reproduction attestation.",
  "independence_criteria": [
    "The reviewer must not have designed USD Impact Score v2, selected its production variables, signs, weights or thresholds, or implemented its production pipeline.",
    "The reviewer must perform the primary recomputation in an environment they control and disclose any compensation, commercial relationship or other material conflict.",
    "USD Impact may answer clarification questions only if each material clarification is retained with the final review record; undisclosed private corrections are not permitted.",
    "The reviewer must be free to publish MATCH, MISMATCH, AMBIGUOUS or NOT_TESTABLE findings without USD Impact editing the substantive result.",
    "Running USD Impact's own CI, rehearsal or post-merge attestation does not satisfy the independence criterion."
  ],
  "release_eligibility": {
    "minimum_week_ending": "2026-08-28",
    "requires_strict_reproduction_bundle": true,
    "requires_latest_archive_bundle_identity": true,
    "requires_main_post_merge_reproduction_attestation": true,
    "requires_exact_release_commit": true,
    "requires_requirements_lock_hash_in_bundle": true,
    "requires_public_methodology_and_data_semantics": true
  },
  "review_materials": [
    {
      "id": "methodology",
      "path": "public/data/score_v2_methodology.json",
      "purpose": "Machine-readable production variables, signed weights, normalization, clipping, missing-data rules and regime thresholds."
    },
    {
      "id": "data_semantics",
      "path": "public/data/score_v2_data_semantics.json",
      "purpose": "Provider field, alignment, resampling, continuous-futures and timestamp/retention boundaries."
    },
    {
      "id": "strict_bundle_latest",
      "path": "public/data/score_repro_bundle_latest.json",
      "purpose": "Frozen weekly levels, normalization moments, z-scores, contributions, provenance, matrix fingerprints, pipeline SHA and dependency-lock hash for the strict release."
    },
    {
      "id": "strict_bundle_archive",
      "path_template": "public/archive/{week_ending}/score_repro_bundle.json",
      "purpose": "Immutable dated counterpart that must match the latest strict bundle for the reviewed release."
    },
    {
      "id": "score_json",
      "path": "public/data/usd_impact_score_v2.json",
      "purpose": "Published current-vintage score output and release metadata."
    },
    {
      "id": "bridge",
      "path": "public/data/weekly_input_latest.json",
      "purpose": "Published latest-week bridge used by the main website."
    },
    {
      "id": "dependency_lock",
      "path": "requirements.lock",
      "purpose": "Exact Python dependency environment whose SHA-256 is bound into strict reproduction bundles."
    },
    {
      "id": "reference_validator_secondary_only",
      "path": "scripts/validate_weekly_release.py",
      "purpose": "USD Impact reference validator. It may be used as a secondary cross-check but must not be the reviewer's only recomputation because that would test the implementation against itself."
    }
  ],
  "primary_replication_steps": [
    "Select one eligible strict release and record its week ending, exact merged production commit and public URLs before calculation begins.",
    "Hash the reviewed methodology, data-semantics, latest bundle, dated bundle, dependency lock, score JSON and bridge. Retain the hashes in the review report.",
    "Confirm the latest and dated strict reproduction bundles are byte-identical for the selected release and that the bundle's requirements.lock SHA-256 matches the reviewed lockfile.",
    "Without importing usd_impact_score_v2.py or using scripts/validate_weekly_release.py as the primary calculation, independently implement the published score arithmetic from the methodology and frozen bundle fields.",
    "For each of the eight drivers, recompute the unclipped z-score from frozen weekly level, mean and sample standard deviation; apply the published clipping rule, signed fixed weight and contribution.",
    "Sum the eight independently recomputed contributions and independently apply the published regime thresholds.",
    "Compare recomputed per-driver z-scores, contributions, total score and regime with the strict bundle and published latest-week output using the declared numerical tolerance.",
    "Review the public data-semantics contract and identify any field, resampling, futures-roll, timestamp, forward-fill or provider boundary that could prevent a new implementation from understanding what the production pipeline consumed.",
    "Optionally run USD Impact's reference validator only after the independent recomputation and report whether it agrees with the independent result.",
    "Publish the completed report using the required finding classes and disclose reviewer identity/qualification, independence statement, conflicts/compensation, environment, exact release commit, artifact hashes and all material clarifications."
  ],
  "required_finding_classes": [
    {
      "class": "MATCH",
      "definition": "The independently implemented check agrees with the frozen/public value within the declared tolerance and no material ambiguity prevents the check."
    },
    {
      "class": "MISMATCH",
      "definition": "The independent calculation or artifact comparison disagrees outside tolerance. The report must preserve both values and the smallest known point of divergence."
    },
    {
      "class": "AMBIGUOUS",
      "definition": "Two or more reasonable implementations are possible from the public specification and the ambiguity could materially change replication."
    },
    {
      "class": "NOT_TESTABLE",
      "definition": "A claimed or required check cannot be independently performed from the available public/frozen evidence. The missing evidence must be identified explicitly."
    }
  ],
  "minimum_report_checks": [
    "exact release commit identity",
    "reviewed artifact SHA-256 hashes",
    "latest/archive strict-bundle byte identity",
    "requirements.lock hash binding",
    "eight frozen weekly levels",
    "eight means and sample standard deviations",
    "eight unclipped z-scores",
    "eight clipped z-scores",
    "eight signed weights",
    "eight contributions",
    "total Score v2 value",
    "regime label",
    "source/provenance identity",
    "weekly matrix and provider-daily fingerprint presence and scope",
    "methodology clarity findings",
    "data-semantics clarity findings",
    "known raw-data-retention limitation",
    "reference-validator secondary cross-check status"
  ],
  "numerical_tolerance": {
    "absolute": 1e-9,
    "meaning": "Maximum absolute difference for frozen-bundle arithmetic checks unless a reviewed field is explicitly textual or exact-hash based."
  },
  "claim_policy": {
    "before_external_report": "Use only 'independent replication protocol prepared' or equivalent. Do not claim independent replication, validation, audit, endorsement or verified model performance.",
    "after_external_report": "Describe only what the report actually tested and found. A reproduction MATCH may support a narrow reproducibility claim for the reviewed release; it does not establish predictive power, economic usefulness, future performance or institutional endorsement.",
    "predictive_power": "Not tested by this protocol. Predictive claims remain governed exclusively by the separate preregistered prospective Score v2 study."
  },
  "known_boundary": "The strict bundle freezes the calculation inputs and normalization moments needed to reproduce the published score and includes hashes of broader provider-derived histories, but complete raw Yahoo/FRED responses and full provider-derived histories are not publicly redistributed. Therefore this protocol tests frozen-release arithmetic and public specification clarity; it does not claim a complete independent reconstruction from original provider transport bytes."
}