Historical Case Closure for Clinical Guidance: Scoring Complete Cutoff-Time Evidence

A reviewer reopening a dosing decision made before a later label revision can recover the right recommendation and still fail to close the case: the answer may point to the successor revision, omit co-active guidance, or cite a span that cannot be resolved in the claimed source. We model this handoff as an evidence-closure path from a dated request to its scoped claim, cutoff-valid source events, reconstructed state, and revision-bound witnesses. Historical case closure (HCC) scores the whole path, accepts adjudicated alternatives, requires coverage for absence claims, and records the first broken edge. On a common 359-request clinical-guidance manifest, GuidanceTrace closes 359/359 requests (95% Wilson CI: 98.9–100%), a bitemporal document retrieve–rerank–span system averages 90.3% \(\pm\) 0.7 across five seeds, and RAG+TimeFilter closes 42.1%. A matched keying intervention supplies the mechanism test: on separate 2,000-event and 1,200-case monitoring tasks it changes drift and contradiction F1 by at most 0.3 points, yet on the closure manifest it loses 52 cases (85.5% (307/359); paired deficit 14.5 points [10.9, 18.4]), including 37 first failures at effective-time version or bundle selection. Independent second-site replay preserves the advantage—92.2% (166/180) closure versus 81.7% (147/180)—and lowers median review time from 31.6 to 6.8 minutes. HCC exposes source-time addressability and evidence binding as concrete design targets for reviewer-complete historical answers.