One start pose — pusher (255, 388), block 59 px from the goal, angle 2° off — pushed by the brain at five points in its development and on two engines. Every video is a real stored episode; nothing is re-rendered or staged. Press play and the five run in lockstep at 10 steps per second.
| # | episode | engine | steps | bouts | peak px/s | position | angle | coverage | cov ≥ .95 | SOLVED | store id |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | first brain (09:12) | pymunk-reference | 200 | 5 | 113 | 59 → 28 px | 2° → 4° | 0.447 | 66 | – | 68b3131bba3f |
| 2 | finish intent (10:34) | pymunk-reference | 200 | 3 | 112 | 59 → 1 px | 2° → 2° | 0.954 | 66 | 70 | 69a6d69527c3 |
| 3 | current baseline | pymunk-reference | 200 | 3 | 112 | 59 → 1 px | 2° → 2° | 0.954 | 66 | 70 | (worktree) |
| 4 | live tab (11:15) | pymunk-reference | 100 | 3 | 112 | 59 → 1 px | 2° → 2° | 0.954 | 66 | 70 | 0ba700a512c3 |
| 5 | SuperDex contact | superdex-reference | 200 | 4 | 114 | 59 → 2 px | 2° → 1° | 0.957 | 127 | 131 | (worktree) |
Read across: #1 → #2 is the effect of giving the brain a finish line (coverage 0.447 → 0.954 from the same start). #2 = #3 = #4 is reproducibility — same code, same seed, same pixels, whether launched from a script or the live tab. #4 → #5 is the engine: same plans, but under friction the block needs a fourth bout and solves 61 steps later. Videos 1 and 5 are the two that look different from the rest; everything in between is the same episode.