You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Bind Aave v4 benchmark ground truth to the ScFuzzBench subject #453
The read-only R60 preflight for #404 fails before any model work:
EVAL_GROUND_TRUTH_SUBJECT_MISSING
benchmarks/private-ground-truth/aave-v4-scfuzzbench/findings.yml still has a legacy top-level source and no schema_version or subject. #423 deliberately made the Modal/eval path require an exact subject repository and revision, so the file is no longer launchable as benchmark ground truth.
It would be incorrect to bind those 12 labels to upstream aave/aave-v4 at 6959e3219b5506bf2acae18551cbb2a68a5b8fba. The checked-in case study records that they belong to the injected ScFuzzBench fork. The verified exact subject is:
The revision is reachable and differs from upstream by the injected contracts/harness (tests/recon/** among 26 changed files), matching the semantics of the 12 labels.
Acceptance
Migrate the checked-in document to ultrafuzz.eval-ground-truth.v1 with the exact ScFuzzBench repository/revision as subject.
Remove or rename the legacy identity-bearing source field so it cannot conflict with subject; retain truthful non-identity provenance if useful.
Preserve all 12 scored bugs and the two canaries byte-for-byte in meaning.
readGroundTruthDocument(..., { requireSubject: true }) and assertGroundTruthSubject pass only for the exact fork/revision and fail for upstream Aave or another revision.
Regression covers the real checked-in file, not only an in-memory fixture.
Activity