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

Mismatch Lab — Product Overview v0.1 (public-safe)

Problem · product · evidence · vision — no fundraising or pricing material
For boundary rules see PUBLIC_BOUNDARY.md


Problem

Headline: Robot teams compare runs all day — but can't always tell when the model itself is wrong.


Product

Headline: Robot Diff — understand what changed. Adequacy when it matters.

Demo: Diff Explorer · Case 3 (repair plateau)

API sketch: api_schema_v0.1.json — mismatch.diff() · mismatch.analyze()


Why this layer

Headline: A decision interface between field experience and model change.

Field telemetry → Mismatch Lab → train / test / expand / escalate

Evidence (tier-labeled)

Preliminary (Research OS · Paper 002 · Tier B+):

In build:

Next research milestones:

Do not cite pilot percentages in external materials until confirmatory CI is frozen.


Vision (long-term · not current product scope)

Physical AI that keeps learning when reality breaks its assumptions.

Roadmap by trust level: Detect → Diagnose → Test → Revise → Validate → Deploy
2026–28 public scope: Detect + partial Diagnose/Test only.

See vision_narrative_v0.2.md for long-term research framing (not a company roadmap).


Closing line

Show us two robot runs. We'll show you what changed — and when your model may need more than retuning.


Updated 2026-07-29