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 003 figures

python scripts/plot_paper003_negative.py
File Content
fig1_carrier_cannot_stop.png The mechanism: settling time against grip duration and duty-cycle period
fig2_arm_scores.png Every arm, injected coupling and three physical configurations
fig3_discordant_pairs.png Paired cells won by exactly one of arm D and SELF, CPU against physical
fig4_settling_sweep.png The criterion measured: settling swept under injected coupling

Where the numbers come from

fig4 and fig2's CPU column are computed from results/settling_sweep_v1.0/burst_off_4.json, produced by scripts/paper003_settling_sweep.py in this repository. Re-running the sweep regenerates them.

Everything physical is transcribed from results/physical_h2_v1.0/RESULTS.md. The Isaac records were written on a rented pod and are not in this repository — their summaries are. Because transcribed constants go stale:

python scripts/plot_paper003_negative.py --check

re-reads that file and reports whether every transcribed number still matches the source line it came from.

Two things these figures deliberately do not do

No smooth densities. Distributions are drawn as the order statistics that were actually measured — median, p10, p90, max. A kernel through four numbers would be a curve we never observed.

fig2's CPU column is not assembled by hand. An earlier version of it took arm B and C from the by-offset arm sweep and arm D and SELF from the preregistered SELF comparison — different seeds, different commit bands — and printed them as one column. It now comes from the settling sweep's settling 0 row, which scores all four arms on one set of cells.

What fig4 shows that the manuscript's first draft got wrong

The predicted crossing is marked on it, at the commanded pause of 4, labelled REFUTED — because it was. The observed crossing is at the duty-cycle period of 14. The figure keeps both so a reader can see which claim was tested and which one survived.

Arm D's curve is the other half: it does not collapse. Settling explains the mode operator's return and not the relational arm's physical failure.