Episode-cluster sensitivity census

Because two seats share an episode, seat-level intervals are anti-conservative under positive dependence — the Round 2 dependence critique. This census reruns the P5-1a, P5-2, and clause-(b) classifications with complete episodes as the unit. The frozen rule is treated as historical; the interpretation rule is explicit: any disagreement among defensible interval constructions is treated as method sensitivity, not resolved by choosing the favorable method. Clause (b) survives everywhere (minimum familywise lower bound 0.462); P5-1a flips under the Jeffreys construction only.