Skip to main content
LatinFlash Research LatinFlashResearch
Vertical AI Governance Stack v1.1 · Human Validation Suite Pre-Registered Protocol External Evaluation Pending

Human Validation Experiment — Pre-Registration Manifest

Cryptographically pinned experimental protocol, frozen normative hashes, and pre-registered acceptance criteria for double-blind human inter-rater evaluation.

Human Validation Status: External Evaluation Pending

Distribution packages (evaluator-A.zip and evaluator-B.zip) have been cryptographically pinned and dispatched directly to independent external evaluators. To prevent experimental contamination, evaluation packages and raw case bundles are retained in private pre-registration storage until external evaluations complete.

1. Pre-Registered Cryptographic Hashes

Target Asset Normative Scope SHA-256 Checksum Status
v1.1 Draft Specification Normative Draft Document 329f851e1c5ae268fb5604dc8768308ee55fd60cbacbb88da2f8161c3f994804 FROZEN
v1.1 Requirements Matrix 22 Formal Requirements aa8feb98ae6ff6f67d22f23a703d0f0ad5542b0ac122ba30d9d1077c43deed60 FROZEN
Evaluator A Package 28 Level-2 Blind Cases 2d8d301101125dbcc9d49a065a86a3271ace920f4e8b13116da5fc054ca02ebc PRE-REGISTERED / PRIVATE
Evaluator B Package 28 Level-2 Blind Cases 4d45be74449d5223508cac41ab7f92d9a845bc2467036f23e53c25093aff891d PRE-REGISTERED / PRIVATE

2. Pre-Registration Protocol

  • Specification Frozen: v1.1 normative draft semantics frozen prior to evaluator exposure.
  • Test Corpus Frozen: 28 anonymized blind cases (6 Base, 10 Synthetic Boundary, 7 Mutation, 5 Path B Adversarial).
  • Oracles Sealed: Reference expected outcomes sealed in private pre-registration storage.
  • Acceptance Criteria Defined: Cohen's Kappa, inter-rater % agreement, and oracle alignment metrics pre-defined.
  • Public Reproducibility Claims: V1.1 CONFORMANCE ASSESSMENT REPRODUCIBILITY VALIDATED will be asserted only after completed evaluator forms are received and scored post-hoc.

For details on the formal specification and test suite, see Vertical AI Governance Stack v1.1 Draft Specification.