Garry Tan
|
c530c5a5be
|
test(eval): BrainBench unit suites — validator seal, metric formulas, governance, adapters, write-back, continuity, generator determinism
61 tests pinning: the sealed-gold validation classes, micro-averaged formula
edges, canonical-baseline byte determinism + every gate verdict path
(count-aware, corpus-bless, justification, allow-regression, isolation at
zero), seam budget/suppression contracts, production-pipeline write-back
provenance, cross-adapter continuity, byte-identical regeneration, holdout
pair discipline, and ledger drift.
|
2026-06-12 12:06:21 -07:00 |
|