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.

Outcome architecture and shared-work research proposal

Status: Proposed / not authorized
Contract line: outcome-architecture-0.1.0-proposal
Official numeric scoring: Disabled
Credential issuance: Disabled
Real-person data and live administration: Prohibited in this package

This package develops two promising ideas without overstating their maturity:

  1. replace the ambiguous phrase “FDE Score” with three separately governed outputs—a Verified Production Record, a Capability Profile, and a future Governed Credential Tier; and
  2. create a Shared Work Challenge Battery that can study human-only, model-only, and human-with-AI work on a common enterprise scenario substrate without presuming that the resulting scores are interchangeable.

This package remains subordinate to SPEC.md and SCORING.md. Their v0.1 prohibition on an official global human number remains in force; no proposal record can change that boundary.

Controlling decisions

  1. Facts, measurement, and certification are different authorities. A verified production fact cannot silently become a capability conclusion; a capability profile cannot silently become a credential; and a credential format cannot make its claims true.
  2. “Psychometric instrument” is a future evidence claim. The Capability Profile is a proposed multidimensional measurement output until its intended interpretations and uses have adequate validity, reliability/precision, fairness, accommodation, and consequence evidence.
  3. A scalar is a possible research direction, not a configuration flag. A future scalar would create a new score interpretation and use. It requires a new benchmark release and the staged authorization gate in this package: evidence readiness, independent method review, and only then a separate governance release decision.
  4. A tier is a consequential classification. Standard setting, uncertainty rules, validity for the stated role and use, independent issuance, correction, appeal, expiry, suspension, and revocation precede issuance.
  5. A shared task does not create a shared construct. Human capability, AI system performance, and human-AI team performance remain condition-specific unless a linking or invariance study supports a narrower comparison.
  6. Missing evidence is not a low result. Every layer preserves Insufficient Evidence, unresolved, disputed, withheld, and not-applicable states.

Documents

Machine-readable schemas are under SCHEMAS/outcome-architecture/0.1.0-proposal. They are isolated from the v0.1 runtime. They require synthetic declarations and identifier shapes, but schema validation cannot prove data provenance. A separate application/environment ingestion guard is required before these contracts could safely be processed.

Evidence basis and limits

The design is informed by the occupational-assessment and benchmark-operation sources already qualified in the Source Ledger, including the Standards for Educational and Psychological Testing, SIOP personnel-selection principles, evidence-centered design, NIST AI 800-2, HELM, METR RE-Bench, ISO/IEC 17024, and W3C Verifiable Credentials.

Those sources supply design principles, not validation or endorsement. AI benchmark precedents do not validate occupational assessment. A cryptographic credential proves properties such as issuer authenticity and integrity; it does not prove that the encoded capability claim is true. No participant, result, field study, human baseline, credential, or scalar score exists in this proposal package.