E4: Removal Test

Goal: test whether gains persist without the system.

Design

8 weeks on AgentSee, 4 weeks off (removal, passive logging only), compare vs passive logging control arm.

Outcomes

TTR and intention-execution coupling during removal, interaction density prior to removal (weaning effectiveness).

Falsifier triggers

Outcomes collapse during removal; user distress increases.

Tests

Invariant I3 (anti-capture), I4 (dependency minimization), and kill criterion 6.5.

Note on timing

The on-period duration (8 weeks) is a design parameter. How quickly the coherence-agency bootstrap builds measurable durable capacity is itself an empirical question. If E4 shows no gains at 8 weeks, one interpretation is the bootstrap is not operating; another is 8 weeks is insufficient. Distinguishing requires either a longer on-period or intermediate capacity probe trends.