v1.8.0

Agent Vitals v1.8.0

This release finalizes av-29 detection credibility and release integrity work.

Highlights

• Confabulation is now a first-class backtest detector metric (confabulation_at labels + P/R/F1 reporting).
• Real confabulation corpus expanded with manually labeled traces (AV29.C01, AV29.C05) and relabeled AV26.F02.
• Backtest CI gate added (scripts/ci_backtest.py) with blocking composite gate (vitals.any) and JSON artifact output.
• CI workflow now runs backtest on Python 3.11 and uploads backtest-results.json.

Validation Snapshot

• Combined three-way corpus size: 49 traces (33 synthetic + 16 real).
• Composite vitals.any: P/R/F1 = 1.000 / 1.000 / 1.000.
• Confabulation (first-class): P/R/F1 = 1.000 / 1.000 / 1.000.

Notes

• Retro tags for v1.6.0 and v1.7.0 were intentionally skipped by operator decision for this release cycle.
• Detailed backtest report: docs/vitals/av29-backtest-report.md.