# yams_any4d tasks

## Bootstrap (active 2026-07-08)

- [ ] Read `agents/yams_any4d/ROLE.md` + `agents/yams_any4d/inbox.md`
- [ ] Set status to `working`
- [ ] Clone TRI-ML/Any4D to `/data/cameron/repos/Any4D/` (if public; else confirm auth with Cameron)
- [ ] Study Any4D: architecture, dataset format expected, finetune examples, checkpoints
- [ ] Write `/data/cameron/vault/para/yam/any4d.md` — bootstrap doc capturing the format-mismatch delta between Any4D and YAM recordings
- [ ] Draft finetune plan in `outbox.md`
- [ ] Post open questions to Cameron
- [ ] Set status to `idle`

## Near-term (waiting on Cameron)

- [ ] Confirm target dataset (which YAM recordings to finetune on first)
- [ ] Confirm evaluation metric priority (task success? MSE? per-frame?)
- [ ] Confirm GPU host + memory budget
- [ ] First finetune run

## Standing

- [ ] Track wandb runs
- [ ] Iterate on hyperparams
- [ ] Feed results into paper via figure_maker