Snapshot v0.8.0
v0.8.0 is the first freeze of the expanded recursive-composition kernels. The benchmark contains 1,000 cases and all 82 bound reports were freshly rerun. The freeze has 4,036 outcomes:
| Outcome | Count |
|---|---|
not-reached | 1,388 |
reached | 1,239 |
inconclusive | 530 |
unsupported | 879 |
These counts describe coverage across independent scorecards. They do not pool analyzers, languages, score tiers, or model profiles into a leaderboard.
Population and measured changes
Section titled “Population and measured changes”The release adds 148 cases to the v0.7.1 corpus, producing 556 new
report cells. They cover balanced recursive composition pairs and the
interprocedural exceptional-exit persistence additions described by the
kernel contracts. Among the 3,480 cells shared with v0.7.1, 124 outcomes
changed under the refreshed analyzer pins: 104 CodeQL C/C++ cells became
inconclusive, while 20 Bifrost cells became decisive across C#, JavaScript,
TypeScript, and Ruby. These are retained measurements, not rewritten scores.
Timing qualification
Section titled “Timing qualification”This release retains 3,157 cold timings. The remaining 879 outcomes are unsupported and have no invented timing. Joern Java and Semgrep Java warm series were each repeated twice; nine invocation-overhead groups were each repeated three times in recorded quiet-host windows. Cache state remains uncontrolled, so these are desktop characterizations rather than causal cross-release speedup claims.
- Scope:
release - Tracks:
taint - Score tiers:
calibrationcorelanguage-extensionmodeling - Model profiles:
benchmark-controlledtool-native - Exclusions: none
Bound evidence
Section titled “Bound evidence”Freeze ID (manifest SHA-256): 582b2589648772926bc7da0a83844eed0450f015c3593fa5f2937c10669f4259
Freeze manifest: reports/freeze.json
Continue to analyzers, languages, semantic templates, case evidence, profiles, or latency.