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 002 — Seed-43 smoke protocol v0.1

ARCHIVED · mock→physics direction · superseded 2026-07-29 · do not cite or extend Current Paper 002: WM expansion · archive index

Purpose: engineering pipeline validation only · not confirmatory
When: after v0.3 commit · before mock seeds 42–46 consensus export
Pre-reg: paper002_prereg_v0.3.md §9


Allowed uses


Forbidden uses

Forbidden Why
Include seed-43 mock in consensus export (Leg 4) Confirmatory ranking must use seeds 42–46 median only
Adjust thresholds, weights, mapping from smoke Pre-reg drift
Add/remove export candidates based on smoke Isaac outcomes Manifest must be frozen pre-confirmatory
Label smoke results as Tier A / confirmatory Engineering only

Procedure

CPU mock smoke (local)

python scripts/run_study2_dream_curriculum_mock.py \
  --config experiments/.../config/sandbox_v0.4.yaml \
  --compare --episodes 16 --seed 43 --agent rule \
  --promote-label smoke_seed43_mock_v0.1

Do not merge smoke_seed43_* into mock_confirmatory_v0.1.

Isaac smoke (VESSL · optional · 1–2 specs)


Required artifacts

Artifact Label prefix
Mock records smoke_seed43_mock_v0.1/
Export manifest isaac_specs_smoke_seed43.json
Isaac aggregate smoke_seed43_isaac_v0.1/
Run log smoke_seed43_run_log.json

Run log must state: "tier": "engineering_smoke", "excluded_from_confirmatory": true.


Relation to confirmatory export

smoke seed 43  ──► engineering only · separate manifest
confirmatory   ──► seeds 42–46 median · frozen checksum manifest

No shared spec_ids between smoke and confirmatory manifests.