# H05_E_MECHANICAL_CONFOUND_RESULT_v0_1

**Confound freeze:** `H05_E_MECHANICAL_CONFOUND_FREEZE_v0_1` sha `4a9d2e501d375a5d9c0a9296da12bc9f571dba9b3fa3f850307fbbd36864da2a`  
**Result JSON sha256:** `35203f16769cc5bffd7cb5312c4151404f011e92f7aa04d12713e7ca313d7209`  
**Parent H05-E result sha:** `27e7af66549e57661fc7bad4a7af06f67a2c74a1a804f165c25bbb0bbbd8597d`  
**Hub job:** `77356229-193e-4693-a90c-cae8dcb0bc4a`  
**Created:** 2026-09-13T23:16:32Z

## STATUS
EXPLORATORY · VINTAGE UNVERIFIED/INCOMPLETE · CLAIM_STRENGTH LOW · CAPITAL_USE PROHIBITED · MODEL_CHANGE NO

## Primary check C1_NESTED_MECH_RAW_DELTA
Same sample as H05-E: Development N=59 · Holdout N=81

| Model | RMSE | MAE | DirAcc |
|-------|------|-----|--------|
| A0 (H05-E baseline) | 110.537017 | 78.860523 | 0.4815 |
| A_mech (A0 + raw mean Δ) | 76.735116 | 64.071031 | 0.6914 |
| B_vs_mech (A_mech + surprise) | 79.313873 | 65.654568 | 0.6173 |

### Primary metric (decision)
- Incremental % RMSE improvement B_vs_mech over A_mech: **-3.3606%**
- Threshold: **5.0%**
- Surprise coef | mechanical: **3.429757**
- Mech coef in A_mech: **3.293136**

### Diagnostics (not decision)
- A0 → A_mech % improvement: **30.5797%** (raw NW2 delta alone)
- A0 → B_original % (H05-E replay): **28.4883%**

## Disposition
# **MECHANICAL**

Claim framing: **DOWNGRADED_MECHANICAL**

H05-E mechanical confound C1: incremental holdout RMSE improvement of surprise over A0+raw_mean_delta = -3.3606% (bar 5%). A_mech RMSE=76.7351, B_vs_mech RMSE=79.3139. Diagnostic: raw delta alone vs A0 improves 30.58%. Disposition MECHANICAL. CLAIM_STRENGTH LOW; capital prohibited; not a validated edge.

## Caveats
- Still not first-print vintage; STEO dates mixed PARSED/SCHEDULE_PROXY
- Coverage: L48 physical vs US-total NGWGPUS (AK small)
- Collinearity between raw Δ and surprise is expected; that is the point of the screen
- CAPITAL PROHIBITED · NOT a validated edge · Explore the click only