Skip to content

VALIDATION_REPORT.md

Latest commit

 

History

History
17 lines (12 loc) · 1.03 KB

VALIDATION_REPORT.md

File metadata and controls

17 lines (12 loc) · 1.03 KB

Validation Report

CompText V7 final hardening validates deterministic compression, golden corpus replay, token telemetry, semantic forensic audit, and dashboard export. Evidence is produced by python scripts/validate.py all, python -m pytest, and python dashboard/industrial_dashboard.py --once.

Evidence summary

  • Golden corpus hashes are fixed in GOLDEN_CORPUS.md.
  • Semantic retention across golden datasets: 1.0.
  • Anomaly survivability across golden datasets: 1.0.
  • Anchor retention across golden datasets: 1.0.
  • Safety-critical retention across golden datasets: 1.0.
  • Replay hash stability: stable across repeated seeded passes.
  • Token drift fingerprint: 4ebdf430f4233805f7303d7d459f84cdc872d82de8d7d548ba192dced2422e65.

Known limitations

The current environment reports tokenizer version fallback-regex if tiktoken is not installed. The validation API supports cl100k_base and o200k_base; installing tiktoken upgrades counts to model tokenizer counts without changing the deterministic validation interface.