fix(m6-synthesis): exhaustive stale-claim sweep — reconcile EVAL-COVERAGE.md and SYNTHESIS.md against canonical scorecard

(A) EVAL-COVERAGE.md: mark scientific-method robust-ELEVATE (+9.3pp p=0.002) claims [SUPERSEDED] — canonical verdict is DIRECTIONAL-NOT-REPLICATED (M5 fresh primary +5.3pp p=0.061 fails gate; replication +8.0pp p=0.001 cannot rescue); mission outcome = ZERO robust-ELEVATE.

(B) SYNTHESIS.md: add complete adjacent provenance (pre/post-edit + directional/significant/null + replicated/not) to all bare numeric lift/p-value claims in Sections 1.5 and 1.7.

(C) Exhaustive sweep complete across all analysis/*.md + README.md — every ELEVATE/robust/replicate hit reconciled against canonical scorecard. No stale claims remain unmarked. skills/**, reviews/**, and untracked user docs left untouched.