solicitation-review-eval

To validate solicitation recommendations against actual human award decisions, compare ACE's top-ranked solicitation recommendation to the recorded human award outcome, scoring alignment, rationale quality, specificity, and edge case coverage to prevent unvalidated recommendations from passing quality checks.

Well-crafted solicitation recommendations pass quality checks but fail to match actual awards when evaluations lack an award-existence gate. Without requiring a recorded human award decision as immutable ground truth, craft-only pass inflation allows misaligned recommendations to pass.

Ground-truth validation for LLM-as-judge solicitation review is the process of anchoring evaluation scores to an out-of-chain human award decision. This mechanism ensures recommendation alignment scoring reflects real-world accuracy rather than writing quality alone.

You should run a solicitation recommendation evaluation for ACE Connect opportunities immediately after a human award is finalized and a lead organization is selected. This ensures the mandatory award-existence gate is met before generating dimension scores and pass verdicts.

The scoring dimensions used to evaluate solicitation recommendations are recommendation alignment weighted at 40%, rationale quality at 30%, recommendation specificity at 20%, and edge case coverage at 10%, with hard-deduct blockers enforced for contract violations.

A solicitation recommendation cannot pass quality assurance without a recorded human award. The award-existence gate mandates a successful human award with a selected lead organization before any pass verdict can be issued, eliminating craft-only pass inflation.