Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
File size: 1,217 Bytes
92c59c2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 | # P1 Measured Sweep Note
Artifact:
- `baselines/sweep_results/measured_sweep_20260308T050043Z.json`
Run shape:
- repaired 4-knob family
- low-fidelity verifier only
- `3 x 3 x 3 x 3 = 81` configurations
Summary:
- total configurations: `81`
- evaluated successfully: `63`
- crashed: `18`
- feasible: `0`
- crash rate: `22.2%`
Most useful result:
- the best coarse near-boundary point was:
- `aspect_ratio=3.8`
- `elongation=1.8`
- `rotational_transform=1.55`
- `triangularity_scale=0.55`
- `p1_feasibility=0.035208`
What the sweep tells us:
- the repaired family does have a usable near-boundary band
- coarse global resolution is still too blunt to capture the small low-fidelity feasible pocket seen in targeted local probes
- `rotational_transform=1.55` performed better than `1.2`, while `1.9` produced a concentrated crash zone
- `triangularity_scale=0.55` was the best coarse setting overall; `0.7` improved triangularity but also increased crash rate
Actionable conclusion:
- keep the current ranges and deltas for now
- do not freeze a new default reset seed from the coarse sweep alone
- use the coarse sweep plus local probes to drive the first tracked fixtures and manual playtesting
|