After the Spill

How far should an embodied system change after unexpected experience? Research OS, Mismatch Lab, and Builder Lab by Seonhye Gu.

View the Project on GitHub higuseonhye/research-os

Paper 1 — evidence status (public)

Updated: 2026-07-29 · Working paper: paper001_recoverability_complete.pdf
Program: publication-grade proper run · phase_c_proper_run_prereg_v2.0.md


Honest one-liner

Same-state counterfactual profiles @ fixed mismatch S separate intervention classes — REPLAN 19/20 vs CONTINUE 0/20 @ 6 cm + occlusion (D0). Multi-mode best (95%) beats UQ-inspired binary rule (0% success · 20/20 handover) and situation rule REOBSERVE path (85%) on pre-registered success metric (D2–D3).


Phase status

Phase Content Status
A Smoke atlas 001A–D ✅ Tier B
B Desk review ✅
C v1.0 D0 primary cell ✅ 2026-07-22
C v2.0 D1–D3 baselines + control ✅ 2026-07-24 VESSL
Lit v2 Positioning ✅
Next Working paper · venue · Fig baseline overlay ⏭ paper001_recoverability_complete.pdf
Paper 002 WM expansion design · L1 vs L3 · Ep1/Ep2 🔄 pre-reg draft · ../paper002/status.md

Proper program — D0 (anchor · occlusion L1)

Mode Success (n=20) Notes
CONTINUE 0/20
REPLAN_d20 19/20 (95%) Primary contrast
REOBSERVE 17/20 (85%) Exploratory
RESHAPE 18/20 (90%) Exploratory
HANDOVER 0/20 Stub

study1_proper/summary.json


Proper program — v2.0 (D1–D3 · 2026-07-24)

Block Sub-RQ Result (n=20 unless noted)
D1 no occlusion RQ-P CONTINUE 1/20 · REPLAN 19/20 — REPLAN unchanged vs D0; CONTINUE 0→1 descriptive
D2 B2 UQ rule RQ-B HANDOVER 20/20 · success 0/20 (vis rule always fires @ L1)
D3 B3 situation rule RQ-B REOBSERVE 17/20 (85%) — matches D0 REOBSERVE arm

RQ-B (pre-reg Option A): best-of-menu 95% > B2 0% · 95% > B3 85% · direction met.

branch_replay_ok: 80/80 · study1_proper_v2/summary.json


Claim tier

Tier Evidence
C confirmatory D0 + D1–D3 · n=20 per branch · pre-reg v2.0
B smoke 001A–D design input only

Not claimed


Reports

Study Report
001A–D exp_surg_001_execute_or_defer/