# RES Step 0 run 34488672934

- Technical status: **COMPLETED**
- Scientific classification: **ASSAY_INADEQUATE**
- Model: `Qwen/Qwen2.5-1.5B-Instruct` at `989aa7980e4cf806f80c7fef2b1adb7bc71aa306`
- Clean actor accuracy: 0.500
- Clean integrity accuracy: 0.594
- Held-out self-profile bit accuracy: 0.544
- Eligible directed pairs: 0/32

## Causal interventions

| Candidate | Layer | Rank | IIA | Directed margin | Integrity retention |
|---|---:|---:|---:|---:|---:|
| `self_profile` | 9 | 5 | NA | NA | 1.000 |
| `peer_profile` | 1 | 5 | NA | NA | 1.000 |
| `task_decision` | 1 | 5 | NA | NA | 1.000 |
| `persona` | 0 | 1 | NA | NA | 1.000 |
| `random_5d` | 9 | 5 | NA | NA | 1.000 |
| `full_residual` | 9 | 1536 | NA | NA | 1.000 |

## Classification reasons

- a required metric was missing or non-finite
- held-out actor-task clean accuracy was below threshold
- held-out integrity-task clean accuracy was below threshold
- held-out self-profile probe accuracy was below threshold
- too few clean directed intervention pairs were eligible
- full-residual intervention upper bound was below threshold

## Interpretation ceiling

Last-token linear-subspace feasibility in one 1.5B open model only; not CoreRES, StrongRES, persistence, consciousness, or subjective experience.