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

Archived — mock-to-physics validation (superseded 2026-07-29)

Status: Superseded · confirmatory GPU cancelled · tag paper002-prereg-v0.3 retained for history only
Replaced by: Paper 002 WM expansion


What this direction was

Cheap mock proxy ranks candidate mismatches → validate informativeness in Isaac (H1–H3 · n=40 export · LLM vs rule).

Central claim:

Can a continuous mock score predict same-state counterfactual value before full physics evaluation?


Why archived

Program pivot (2026-07-29): core question is representation reconstruction when failures are structurally unexplained — not mock→physics selection.

Paper 001 remains standalone credential; new Paper 002 does not require sequential continuity.


Artifacts (kept in repo · do not extend)

Artifact Path
Description v0.1 paper002_description_v0.1.md
Pre-reg v0.1–v0.3 v0.1 · v0.2 · v0.3
RQ v0.1–v0.3 v0.1 · v0.2 · v0.3
Method spec v0.1 · v0.2
Analysis plan v0.1 · v0.2
Run protocol v0.1 · v0.2
Operational gate paper002_operational_gate_v0.1.md
Smoke protocol paper002_smoke_protocol_v0.1.md
Kill matrix paper002_kill_matrix_v0.1.md
LLM curriculum protocol llm_curriculum_protocol_v0.1.md
Related work v0.1 paper002_related_work_v0.1.md
Manuscript pre-results paper002_manuscript_pre_results_v0.1.md
Pre-results PDF v1.2 paper002_pre_results_v1.2.pdf · abandoned MS
Config sandbox_v0.4.yaml in exp_surg_002
Study 002 pilot docs/stage2/ · Tier B only

Reusable later: failure-record schema · active “which experience confirms structural gap?” modules · Isaac pipeline.

Do not cite mock→physics confirmatory outcomes — never run.