learnings-summary-eval

Grade closeout learnings summaries by evaluating coverage gaps, recommendation actionability, and tone calibration against actual delivery outcomes like UAT results and observation logs. This independent evaluation prevents biased closeout documentation authored by the same lifecycle LLM.

Closeout summaries often miss phase coverage and contain vague recommendations when self-reported by the same LLM running the cycle. An independent grading rubric enforces strict citation and actionable recommendation standards to detect these synthesis gaps.

Prevent inflated closeout tone by anchoring tone calibration to independent outcome evidence such as launch status and UAT results, rather than self-reported cycle grades. An inflation guard blocks celebratory documentation for opportunities that failed to launch or missed key metrics.

Validate Phase 10 learnings summaries by cross-referencing synthesis artifacts against per-phase run state, UAT results, observation logs, and original PDD baselines. A 5-dimension weighted grading rubric enforces forward-seeding clarity and evidence citation discipline.

Yes, the learnings summary evaluation applies specifically to ACE closeout workflows for Connect opportunities. It independently grades the full Phase 1–10 lifecycle artifacts to ensure concrete, actionable improvements align with what the opportunity actually delivered.

Avoid skipping independent closeout validation when a learnings summary is authored by the same LLM that ran the full opportunity lifecycle. Hard block rules for critical failures and an inflation guard are necessary to prevent missed phase coverage and tone misalignment with actual delivery outcomes.