# Route 2: PAct + OT Variant Evaluation 20260510

Generated: `2026-05-10T14:11:05+00:00`

## Mean Metrics

| metric | baseline | OT variant | delta |
|---|---:|---:|---:|
| `joint_count_accuracy` | 0.4000 | 0.6000 | +0.2000 |
| `joint_count_abs_error` | 1.4000 | 1.2000 | -0.2000 |
| `joint_type_accuracy` | 0.6000 | 0.6000 | +0.0000 |
| `axis_mean_error_deg` | 54.1784 | 54.2100 | +0.0316 |
| `origin_mean_error` | 0.3116 | 0.3372 | +0.0256 |
| `axis_finite_rate` | 0.6000 | 0.6000 | +0.0000 |
| `limit_sanity_rate` | 0.6000 | 0.6000 | +0.0000 |
| `urdf_parse_success` | 1.0000 | 1.0000 | +0.0000 |

## Samples

| sample | category | GT | baseline axis | OT axis | prior |
|---|---|---:|---:|---:|---|
| `Door_8867` | `Door` | 2 | 0.00 | 0.00 | `1363` |
| `TrashCan_10357` | `TrashCan` | 1 | 89.84 | 90.00 | `1363` |
| `Safe_1363` | `Safe` | 2 | 1.05 | 1.05 | `8867` |
| `Bottle_3517` | `Bottle` | 1 | 90.00 | 90.00 | `10357` |
| `Display_3386` | `Display` | 5 | 90.00 | 90.00 | `10357` |

## Interpretation

This run is useful as a concrete OT insertion probe: it demonstrates how a transport plan can regularize PAct's joint-slot assignment without changing the generator.  It should not be treated as a final model improvement until the prior is moved into the denoising/attention stage and evaluated on more examples.