AgentSeeResearch Notebook
version 1.0.0 · created 2026-04-08 · updated 2026-04-08

E2: Micro-Randomized Stabilization Trial

experiment
ClaimProtocol for testing whether bounded actions reduce TTR without autonomy loss via micro-randomized trial.

Goal: test whether bounded actions reduce TTR without autonomy loss.

Design

Micro-randomized trial (Klasnja et al. 2015), decision points triggered by high confidence of leaving R, randomize among: silence, state legibility, choice compression, micro-controllability task, opt-in physiology prompt.

Outcomes

TTR distribution shift, autonomy measures, monitoring anxiety, user-rated coercion, interaction density trends.

Falsifier triggers

No TTR improvement; autonomy decline; increased monitoring anxiety; increased dependency markers.

Tests

P1, P3, and kill criterion 6.5.