Round 1 results + combined exp-1a2a config for Round 2

Document Round 1 findings: exp-1a (state_age embed) = +7.2pp
change_acc, exp-2a (press events) = +1.9pp. Exp 1a beat its
prediction ceiling; exp 2a fell short. Both beat baseline.

Add combined config stacking both flags for Round 2.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: ScavieFae