Techniques · Evasions
The lean back
Taking the head back out of range without moving the feet.
Where it stands
| Technique | Teacher kept | In the student | Own runs |
|---|---|---|---|
| lean back | none yet | not yet | 2 |
The lean back is from my motion capture, Muay Thai 02, take_01, 23.05–24.35 s: the head goes back out of range and the feet stay where they are. It has no target. It counts as done when the head passes within 5 cm of where my head is at the apex. Its teachers were cloned from the student distill4, alongside the slip and uppercut and the two rolls (work log).
The first, leanback1-231.8M, never leaned. It stood almost still, the head 7 cm down at most, and scored 0 of 8, 38.9 cm from the apex. The second, leanback2-234.4M, trained with a reward for the head following its path along my reference (which, unlike the others, wasn’t widened). It leans now, but never reaches the apex: 0 of 8, 19.3 cm off. Both are recorded as failed models.
The evasions shared a cause: the punch clock, made for punches, let the teachers rush my moves, and the gate checks neither timing nor shape. No lean-back teacher is kept. Next it starts again from my original capture, with the clock held to the capture’s pace and a gate that measures timing and shape.
Every attempt
Every run that trained it on its own, in order.
| Run | Outcome | Machines | Hours | Cost |
|---|---|---|---|---|
leanback1-2026102524 October · lean back · Teachers: slips, rolls and the lean back | Failed | 1 | 2.1 | $1.14 |
leanback2-2026102525 October · lean back · Teachers: slips, rolls and the lean back | Failed | 1 | 2.1 | $1.28 |
| 2 training runs | 2 | 4.2 | $2.42 |
How the models descend
Its own runs, each with an arrow from the run it started from; faded boxes are models from elsewhere it started from or learnt from. Tap a run for its details. The whole history is on the lineage page.