FDE BenchmarkPublic document reader
Available in sourcev0.1 · live public draft · official release pending

Canonical source view. This renders the bundled repository document as readable HTML. Source metadata remains in the repository, and relative document links remain inside this reader.

Future scalar-score authorization gate

Status: Research proposal / not authorized
v0.1 global human score: Prohibited
This document contains a score formula or score: No

1. Position

The current prohibition on a single global human score is both a governance rule and a scientifically appropriate boundary for the evidence available today. Governance could choose to study a scalar in a later release, but it cannot make the scalar valid by voting to remove the prohibition.

A future scalar would be a new score interpretation and use. It changes what the benchmark claims about a person, how evidence is aggregated, which weaknesses can be hidden, how uncertainty is communicated, and how downstream decisions may be made. It therefore requires a new benchmark release and a fresh validity and consequence argument.

2. Non-negotiable separation

Even if a scalar is later supported, it must not replace:

  • the Verified Production Record;
  • the multidimensional Capability Profile;
  • visible confidence, coverage, maturity, verification, recency, contradiction, and critical-gate states;
  • the Insufficient Evidence state; or
  • the independent credential-tier decision.

A scalar would be an optional, purpose-bound projection. It would not be an intrinsic property of a human.

3. Staged gate receipts

The machine contract separates three sequential stages so the final decision cannot become a circular prerequisite for review readiness:

  1. thirteen evidence-readiness receipts must be satisfied before the candidate can become eligible for independent method review;
  2. an independent-method-review receipt and outcome determine whether the candidate may proceed to a governance release decision; and
  3. a separate governance-decision receipt records the final decision only after method review.

Each property is bound to a named receipt identifier and evidence-artifact type. A satisfied state also requires an assigned review-authority class, the artifact digest, review time, and a matched SHA-256 verification record. A document title or an unverified digest is not sufficient. These machine fields are necessary but not sufficient: an operational gate would still have to verify the artifact against its external store and the reviewer against an authorized, versioned authority registry. The proposal JSON cannot attest its own trustworthiness.

Stage Gate Minimum evidence question
Evidence readiness Construct and intended use What exact interpretation and decision use is proposed, for whom, in what context, and with what prohibited uses?
Evidence readiness Job analysis and content blueprint Does the assessed work represent important FDE activities and link tasks, observations, and rubrics to the construct?
Evidence readiness Dimensionality and aggregation Do the data support the proposed aggregation? What information and critical weaknesses does it conceal? Are weights empirically and substantively defensible?
Evidence readiness Calibration and sample adequacy Are samples large, relevant, diverse, and independent enough for each claimed interpretation? Was the analysis plan prespecified?
Evidence readiness Reliability, precision, and uncertainty Are results sufficiently stable across tasks, raters, occasions, and forms for the stated use? Are conditional uncertainty and decision consistency reported?
Evidence readiness Standard setting for any tier Were cut points set through a documented method by qualified, non-conflicted experts, with sensitivity and classification-error analysis?
Evidence readiness Fairness, accessibility, and accommodations Does the process measure the intended construct across relevant groups and accessible paths without avoidable construct-irrelevant barriers?
Evidence readiness Criterion and outcome linkage Does the proposed interpretation relate to independently defined field outcomes in the claimed context, without confusing association with causation?
Evidence readiness Consequences, adverse impact, and alternatives What beneficial and harmful consequences occur, what decision errors matter, and are less adverse valid alternatives available?
Evidence readiness Rater and evaluator quality Are raters qualified and calibrated, drift controlled, model assistance bounded, and adjudication quality measured?
Evidence readiness Security, contamination, and reproducibility Are controlled cases, leakage, identity, artifact integrity, replay, and deterministic recomputation adequately protected and tested?
Evidence readiness Privacy and legal-use review Are purpose, consent, disclosure, retention, data rights, jurisdiction, and proposed employment uses reviewed by qualified owners?
Evidence readiness Correction, appeal, revocation, and monitoring Can factual and measurement errors be corrected, decisions appealed, credentials changed, and drift or harm detected after release?
Independent review Independent external method review Have reviewers with relevant measurement, domain, accessibility, security, and affected-party expertise evaluated the complete evidence package?
Final decision Governance release decision Has the authorized, non-conflicted body approved the exact immutable release, interpretation, use, limits, and monitoring plan?

4. Evidence standards for aggregation

Before studying a scalar, the research program must predeclare:

  • whether the proposed construct is reflective, formative, hierarchical, or a decision index;
  • why aggregation is more faithful to the intended interpretation than a profile alone;
  • which dimensions are non-compensatory critical floors;
  • how weights are derived, reviewed, and stress-tested;
  • how missingness, not-applicable scope, recency, contradictions, and low coverage are handled;
  • the estimand and uncertainty method;
  • the target population, role profile, and administration conditions;
  • the minimum precision and decision-consistency criteria; and
  • the material changes that force recalibration or a new release.

Good fit statistics cannot substitute for a defensible construct. Expert weights cannot substitute for empirical study. Predictive association cannot by itself justify a consequential employment use.

5. Research sequence

  1. Profile first. Demonstrate that the domain and competency measurements are interpretable, sufficiently reliable, and useful without a scalar.
  2. Pre-register scalar candidates. Specify alternative aggregation models, estimands, weights, critical floors, exclusions, and falsification criteria.
  3. Use independent calibration data. Do not choose a model and report its apparent performance on the same sample.
  4. Compare against profile-only decisions. Measure information loss, decision consistency, subgroup behavior, and whether the scalar changes decisions in defensible ways.
  5. Run sensitivity and consequence studies. Vary weights, missingness, uncertainty, cut points, and role contexts; record who is helped or harmed and why.
  6. Conduct external adversarial review. Publish the interpretation-and-use argument, evidence map, known rebuttals, limitations, and independent review disposition.
  7. Record a separate governance decision. Independent review readiness and completion must not depend on a decision that has not yet occurred.
  8. Authorize through a new release only. A proposal, feature flag, schema pass, or software deployment cannot activate a score.

6. Tier relationship

A scalar and a tier are separate decisions. A scalar may exist without a tier; a tier may be based on noncompensatory profile rules rather than a scalar. Whenever a tier is proposed, its standard-setting, classification-error, fairness, consequence, and lifecycle gates remain mandatory.

7. Proposal contract behavior

scalar-score-authorization-gate.schema.json records 13 property-bound evidence-readiness receipts, an independent review receipt and outcome, a later governance-decision receipt and outcome, and a staged research-review disposition. The relational gate verifies that a satisfied receipt's expected, observed, and bound artifact digests match and that its review state is consistent with the stage. The proposal always sets:

  • official_scalar_score_authorized: false;
  • scalar_score_present: false;
  • proposal_can_self_activate: false; and
  • v0_1_global_score_prohibition_preserved: true.

Satisfying the 13 readiness receipts may reach only eligible_for_independent_method_review; it does not require or imply a final governance decision. A passing independent review may then reach eligible_for_governance_release_decision. Even a recorded governance decision cannot activate a scalar through this proposal because the authority fields remain false and no normative release artifact can be present. A later normative release—not this proposal—would be required to carry authorization.