Skip to content

feat(engine): calibrate evidence and reduce compact scan memory - #4

Merged
danielcadev merged 9 commits into
mainfrom
codex/trust-composition-execution-boundaries
Aug 30, 2026
Merged

danielcadev merged 9 commits into
mainfrom
codex/trust-composition-execution-boundaries

Conversation

@danielcadev

Copy link
Copy Markdown
Contributor

Summary

  • improve semantic precision for trust composition, authorization state, filesystem identity, outbound requests, process execution, caches, secrets, and policy boundaries
  • keep uncertain evidence as explicit deterministic abstentions instead of findings
  • add repository-independent vulnerable/control matrices and metamorphic regression coverage
  • preserve compact and full secure-json-v1 contracts while retaining causal evidence and bounded reporting
  • add a direct compact scan path that counts fact-only global graph evidence without materializing evidence removed by the compact projection
  • document local npm CLI resource and equivalence evidence with explicit limitations

Measured local evidence

On the clean, authorized npm CLI checkout at b888cc9a9ff34a8b023ff47b784692396635397b, two release samples reduced mean compact peak RSS from 987,470 KiB to 737,420 KiB, approximately 25.3%.

The optimized report preserved:

  • 4,847 scanned files
  • 57,748 normalized facts
  • 208,157 conceptual graph nodes
  • 368,627 conceptual graph edges
  • zero findings and three explicit abstentions
  • compact fingerprint 0c76afa4aa89f99d7c48bc08edc37f8e7e375a46ebb00aa43149f78bd3635b82

Wall time is not presented as a speedup. The measurements are traceable single-host evidence, not an independently reproduced or universal resource guarantee.

Verification

  • cargo fmt --all -- --check
  • cargo clippy --workspace --all-targets --all-features -- -D warnings
  • cargo test --workspace --all-features — 236 passed
  • cargo audit --no-fetch — 426 dependencies, 1,226 advisories, no reported vulnerabilities
  • compact/full projection equivalence and truncation regressions passed
  • release binary SHA-256: c4b2dbe6c06d7b020c483bbb38b3a26959155cc118e81e2e8c777b7e83d77d4a

Limits

Zero findings is not a clean verdict. The npm abstentions still require human review. Full-graph mode remains explicitly opt-in and retains its original memory cost. A release or version change is not part of this PR.

@danielcadev
danielcadev merged commit 16fd9c5 into main Aug 30, 2026
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant