cycle-grade-eval

To eliminate self-evaluation inflation bias, independently re-grade closed ACE opportunity cycle closeout scorecards by anchoring scoring strictly to out-of-chain raw data sources like FLW delivery records, observation logs, and per-phase artifacts rather than relying on potentially inflated per-skill verdict files.

Evidence-anchored cycle evaluation independently validates performance claims by re-deriving at least one load-bearing cycle outcome directly from raw run evidence, such as observation logs and run state product blocks, structurally preventing the executing model from grading its own performance without external corroboration.

To validate closeout grades across all ACE opportunity archetypes including atomic visit, focus group, and multi-stage cycles, apply an independent re-grading process across five weighted dimensions to detect missing concrete learnings and non-specific recommendations, outputting a standardized verdict YAML.

Cycle evaluation explicitly handles incomplete Phase 10 closeouts by detecting missing closeout data and emitting an explicit incomplete verdict instead of forcing an inaccurate grade, ensuring edge cases are documented without corrupting the audit trail.

The best way to audit cycle closeout performance for vague recommendations is to apply structured verdict generation across weighted dimensions including recommendation specificity and trajectory framing, enforcing hard guards against uncorroborated claims to ensure actionable learnings for future cycles.

Auditing self-graded closeout scorecards requires out-of-chain raw run data sources including FLW delivery records, observation logs, per-phase artifacts, and run state product blocks to independently re-derive performance claims and eliminate structural self-evaluation bias.