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Paper 002 — Research questions v0.1 (superseded)

ARCHIVED · mock→physics direction · superseded 2026-07-29 · do not cite or extend Current Paper 002: WM expansion · archive index

Superseded by paper002_rq_v0.2.md

Paper: CF-validated generative curriculum for informative failure scenarios
Experiment: EXP-SURG-003 · Codebase: exp_surg_002_dream_curriculum
Pilot: Study 002 / EXP-SURG-002 (Tier B · archived)
Ruler: Paper 001 same-state CF @ fixed S · study1d occlusion contract
Date: 2026-07-28


Position in program

Layer Artifact Status
Measure @ S Paper 001 ✅ complete
Generate probe Study 002 ✅ pilot · informs design
Generate confirmatory Paper 002 🔄 this document

Paper 003 (deferred): latent mismatch + response selection — not in scope here.


Central question

Can a mock–Isaac aligned generative curriculum—using Gaussian and diffusion dreamers under an agentic planner (rule or LLM)—produce informative failure scenarios whose mock rank predicts Isaac same-state counterfactual outcomes @ S?

Informative @ spec (frozen): CONTINUE unsuccessful ∧ REPLAN successful (Paper 001 definition).


Sub-questions

ID Question Role
RQ-B1 Does mock per-spec informative rank correlate with Isaac rank at confirmatory scale (40-spec pack · study1d)? Primary
RQ-B2 Does tier separation (top vs bottom mock rank) hold on Isaac with stricter floors? Primary
RQ-B3 Does an LLM JSON curriculum validate on Isaac—not only mock—vs the rule planner? Primary confirmatory extension
RQ-B4 Gaussian vs diffusion: yield vs param diversity tradeoff? Exploratory
RQ-B5 Hybrid dreamer (mixed proposals) improve tier gap or ρ? Exploratory

What Study 002 established (pilot · not re-tested as primary)

Pilot finding Paper 002 use
Diffusion > Gaussian on yield Not primary — closed at Phase 1
Occlusion misalignment broke H3 Mandatory study1d + visibility map
H3′ ρ=0.899 @ n=20 Power basis for n=40 confirmatory
LLM JSON mock-only RQ-B3 — Isaac leg required

Success definition (paper-level)

Tier A candidate (primary story):

  1. H1 + H2 PASS @ rule agent (replication at scale), and
  2. H3 PASS @ LLM agent (Isaac validation), or honest Tier B: “mock filter works · LLM = rule on Isaac”

Tier B (still publishable):

Stop (no GPU spiral):


Non-claims