# Wolf-route knockout v1

Completed: `2026-07-15T04:50:47.502437+00:00`

## Split causal conclusions

- Endpoint behavioral knockout gate: **True**
- Early optimization privilege: **inconclusive_or_uncertain**
- Persistent privilege: **mixed_across_seeds**
- Endpoint fixed-budget route cost: **inconclusive_or_uncertain**

A later catch-up is not allowed to erase an early route tax. The local component curve and 32-update release curve are reported separately because one-step utility and dynamical takeover are different claims.

## Endpoint and integrated results

| seed | K(sham-null) | tau P | tau specific | endpoint P tax | endpoint specific |
|---:|---:|---:|---:|---:|---:|
| 56101 | 0.528998 | -0.00088525 | -0.00134566 | 0.00049318 | -0.00026213 |
| 56102 | 0.510097 | -0.00150984 | -0.00437525 | -0.00194247 | -0.00636951 |

## Early fixed-64 paired tax

- Seed 56101: meaningfully positive gated updates `[]`
- Seed 56102: meaningfully positive gated updates `[]`

## Interpretation boundary

The one-step curve isolates the actual wolfward component locally. The 32-update curve tests a released optimization policy/path, not isolated wolf causality. Neither identifies the accumulated alternative circuit.

Full preregistered output: `runs/wolf_route_knockout_v1.json`