Flamehaven Verification Ledger

Verification Ledger

Endpoints

GET api/v1/runs.json

Index of all verification runs — EQA, BAV, BSC lanes. Returns id, verdict, brief, detail_url for each run.

GET api/v1/runs/{id}.json

Per-run detail — full summary, key_metrics, findings[], and evidence links.

Example: runs/bav-exp-034.json · runs/toe-test-0056.json · runs/bsc-yorkeccak-bio.json

GET api/v1/metrics/bav.json

Aggregated BAV pipeline metrics table across all 6 experiments.

cross_domain_consistency calibration_brier_score balanced_accuracy end_to_end_reliability

GET api/v1/schema.json

Vocabulary — 3 lane definitions, 13 verdict codes, 6 metric definitions with provenance labels.

Architecture. All files are pre-generated from each experiment’s manifest.json — no live computation. The build script (scripts/build_api.py) is deterministic and manifest-driven; a CI gate (--check) fails the build if committed files drift from source manifests.

Provenance. Internal pipeline metrics (SR9, DI2, p‍_e2e) carry provenance: advisory_internal in schema.json. Verification weight rests on external anchors: public MIT repositories, Zenodo DOI 10.5281/zenodo.15487327, and standard third-party metrics (AlphaFold pLDDT, PAE).

Public API v1.0.1