Task registry
TOWER evaluates six task families at two and three layers. Use canonical ids in new result reports. The machine-readable registry retains historical aliases.
| Task | Canonical id | Legacy id | Deadline | Success rule | Planned G/P | Pushes | Status |
|---|---|---|---|---|---|---|---|
| Task A (cross transfer), 2 layers | cross-transfer-2 |
task2-2 |
60 s | tower-sim-trajectory-v5 |
2/2 | — | benchmark |
| Task A (cross transfer), 3 layers | cross-transfer-3 |
task2-3 |
131 s | tower-sim-trajectory-v5 |
3/3 | — | benchmark |
| Task B (two-block transfer), 2 layers | two-block-transfer-2 |
task1-2 |
125 s | tower-task12-four-place-hold-v1 |
4/4 | — | benchmark |
| Task B (two-block transfer), 3 layers | two-block-transfer-3 |
task1-3 |
227 s | tower-sim-trajectory-v5 |
6/6 | — | benchmark |
| Task C (extract and stack), 2 layers | extract-and-stack-2 |
task6-2 |
40 s | tower-task6-roof-no-fall-v1 |
1/1 | — | benchmark |
| Task C (extract and stack), 3 layers | extract-and-stack-3 |
task6-3 |
77 s | tower-task6-roof-no-fall-v1 |
2/2 | — | benchmark |
| Task D (lateral insertion), 2 layers | lateral-insertion-2 |
task5-2 |
43 s | tower-sim-trajectory-v5 |
1/1 | — | benchmark |
| Task D (lateral insertion), 3 layers | lateral-insertion-3 |
task5-3 |
82 s | tower-sim-trajectory-v5 |
2/2 | — | benchmark |
| Task E (layerwise pushdown), 2 layers | layerwise-pushdown-2 |
task4-2 |
79 s | tower-task4-final-state-v3 |
3/3 | 6 | benchmark |
| Task E (layerwise pushdown), 3 layers | layerwise-pushdown-3 |
task4-3 |
113 s | tower-task4-final-state-v3 |
6/3 | 9 | benchmark |
| Task F (tool-assisted extraction), 2 layers | tool-assisted-extraction-2 |
task7-2 |
216 s | tower-sim-trajectory-v5 |
5/3 | — | benchmark |
| Task F (tool-assisted extraction), 3 layers | tool-assisted-extraction-3 |
task7-3 |
404 s | tower-sim-trajectory-v5 |
10/6 | — | benchmark |
| Aligned stack, 2 layers | aligned-stack-2 |
task3-2 |
— | — | — | — | experimental-legacy |
| Aligned stack, 3 layers | aligned-stack-3 |
task3-3 |
— | — | — | — | experimental-legacy |
Aligned stack is a compatibility identifier and is not a scored benchmark task. Task-specific contact geometry and scoring deadlines must remain unchanged. See metrics and the public benchmark.
Accepted identifiers are defined in the machine-readable task registry.
For example, lateral-insertion-2, D2 and task5-2 identify the same task;
see alias forms.