How far should an embodied system change after unexpected experience? Research OS, Mismatch Lab, and Builder Lab by Seonhye Gu.
ARCHIVED · mock→physics direction · superseded 2026-07-29 · do not cite or extend Current Paper 002: WM expansion · archive index
Supersedes: v0.1 execution gates · Pre-reg: paper002_prereg_v0.3.md
Operational gate: paper002_operational_gate_v0.1.md
Smoke: paper002_smoke_protocol_v0.1.md
1. v0.3-amend commit + tag
2. seed-43 smoke (engineering · optional · separate manifest)
3. LLM curricula · 5 seeds
4. mock rule · seeds 42–46
5. mock LLM · seeds 42–46
6. consensus merge (median) · within-planner export · checksum freeze
7. rule Isaac engineering leg
8. operational go/no-go (NOT H1∧H2)
9. LLM Isaac leg (if go)
10. confirmatory H1 → H2 → H3 analysis
cd C:\projects\research-os
git add docs/paper002/ experiments/.../config/sandbox_v0.4.yaml scripts/
git commit -m "Paper 002 pre-reg v0.3-amend: proxy validation design"
git tag paper002-prereg-v0.3
See paper002_smoke_protocol_v0.1.md.
Exclude from mock_confirmatory_v0.1 and confirmatory export.
Consensus merge:
python scripts/merge_study2_mock_consensus.py \
--results-dir experiments/surgical_intelligence/exp_surg_002_dream_curriculum/results/mock_confirmatory_v0.1 \
--seeds 42,43,44,45,46 --planner rule \
--out experiments/surgical_intelligence/exp_surg_002_dream_curriculum/artifacts/consensus_rule_v0.3.json
Export (within-planner · 20 specs):
python scripts/export_study2_isaac_specs.py \
--records experiments/surgical_intelligence/exp_surg_002_dream_curriculum/artifacts/consensus_rule_v0.3.json \
--scope planner --planner rule --strategy top_bottom --top-k 10 \
--prereg docs/paper002/paper002_prereg_v0.3.md \
--out experiments/surgical_intelligence/exp_surg_002_dream_curriculum/artifacts/isaac_specs_rule_v0.3.json
Repeat for LLM consensus → isaac_specs_llm_v0.3.json.
Record SHA256 checksum on both manifests before Isaac.
VESSL commands as v0.1 · env STUDY2_RECORDS / specs from rule manifest only.
Promote → results/selection_ablation_rule_v0.3/isaac_aggregate.json
Compute from rule aggregate:
Log → results/operational_gate_v0.1.json
If NO-GO: skip Leg 7 · feasibility report · H3 N/A.
Same protocol · LLM manifest only · promote → selection_ablation_llm_v0.3/
After both legs (or rule-only if NO-GO):
# H1 + H2 @ rule export set
python scripts/compute_study2_h3_mock_isaac.py \
--specs experiments/.../artifacts/isaac_specs_rule_v0.3.json \
--isaac-aggregate experiments/.../results/selection_ablation_rule_v0.3/isaac_aggregate.json \
--continuous --top-k 10 --strategy top_bottom \
--prereg-version v0.3-amend \
--out experiments/.../results/h3_rule_v0.3/summary.json
# H3: repeat on LLM + compare Δρ (analysis notebook or second run)
Interpret H1 → H2 → H3 hierarchically per analysis plan v0.2.