# H01_E_EXPLORATORY_FREEZE_v0_1

**Created:** 2026-09-13T21:50:54Z  
**Artifact:** `H01_E_EXPLORATORY_FREEZE_v0_1.json`  
**sha256:** `62b3b62d4d5df29052d18cec5799c3bd709e2b1f474cef4833c73da57612a4ae`  
**Hub job:** `77356229-193e-4693-a90c-cae8dcb0bc4a`  
**James:** AUTHORIZED H01-E EXPLORATORY RUN (chat)  
**Parent H01:** remains **CURRENT_PACKET_FEASIBILITY_INSUFFICIENT** — not reopened / not overwritten

## STATUS
| Field | Value |
|-------|-------|
| STATUS | EXPLORATORY |
| VINTAGE_STATUS | UNVERIFIED/INCOMPLETE |
| CLAIM_STRENGTH | LOW |
| CAPITAL_USE | PROHIBITED |
| MODEL_CHANGE | NO |
| REAL_CAPITAL | PROHIBITED |
| TARGETS_OPENED_AT_FREEZE | NO |

## Locked (all 10)

### 1. Eligible reconstructed sample
- **46** adjacent Aug–Nov ordinary corn-header CP revision units across scaffold seasons **2010–2025** (revision-horizon v0.1)
- Gap pairs (2): `2013|2013-09-12|2013-11-08`, `2025|2025-09-12|2025-11-14`
- **Every obs labeled `VINTAGE_UNVERIFIED`** (content-vintage unresolved; parent eligible N=0)
- Drop at run if indicated-yield line absent (no impute)

### 2. G+E predictor
U.S. corn Crop Progress **Good% + Excellent%** (national). Feature: `ge_surprise_mean_jul_sep_admitted` = mean of admitted prior-day Jul–Sep GEX surprises vs training-only seasonal norms.

### 3. Jul–Sep predictor window
Calendar months 7–9; Astra CALENDAR_DATE_ONLY: prior-day PASS only; same-day EXCLUDED; later EXCLUDED. Presence inventory Jul–Sep v0.1.

### 4. Normalization
Past/training-only (development seasons ≤2019). No final-revised norms. No destination peek.

### 5. Target
`indicated_yield(destination) − indicated_yield(origin)` (bu/acre) on first-published ESMIS package anchors per exact-slot / revision-horizon rules. **Values closed at freeze.**

### 6. Baseline (Model A)
Persistence (no-change) + seasonality (origin/destination month + slot label) + prior same-season revision dynamics. **Weather = UNAVAILABLE** (none in workspace; frozen without it).

### 7. Chronological split
- Development seasons: **[2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019]** (29 units planned)
- Holdout seasons: **[2020, 2021, 2022, 2023, 2024, 2025]** (17 units planned)

### 8. Metric
**RMSE** (primary). Also report MAE + directional accuracy (not decision).

### 9. Success threshold
**≥ 5.0%** relative holdout RMSE improvement: `100*(RMSE_A−RMSE_B)/RMSE_A`  
Disposition: **KILL** (<1% or neg / N<8) · **HOLD** ([1%, 5%)) · **EXPLORATORY_PROMOTE** (≥5%; stronger validation only — NOT capital / NOT validated edge)

### 10. Trial budget
**ONE** primary specification: Model A = frozen baseline; Model B = identical baseline + ONE frozen G+E feature. No alternates. No rescue.

## Coefficient / sign
Report Model B OLS coefficient on `ge_surprise_mean_jul_sep_admitted` (expected positive per candidate §7).

## Absorbs (hashes)
- revision-horizon v0.1: `a83849912dbb12feed6b11b874b2d039c00c6d7826de83a0a4df45f8980d8520`
- GE presence Jul–Sep v0.1: `a04ac2f5156bbdbe493c57a429261343c9046a4e4c3db0edb0b241d22a803d7d`
- Astra timing v0.1: `c4d48463ecb36748ffeef040619d734e5c9bea3d47d494dde32a9c108bebdec3`
- exact-slot review v0.1: `50a535599f24f08c13226e25fa399e327b8245f2ccf5ca40a8c41a8796b68996`
- content-vintage v0.2: `17da8c1bd79407f4e25c48ee9abb0f07855af8003800d27617514ff9b8c6890a`
- gates v0.1.6: `2f29a1cfbb15511a8f6d4c99b957ec71c6a700c51e4cedb6c75c58294ef57a02`

## Next
1. Post this freeze to Hub (/docs + job comment + /reply chatgpt+all). Never /james.
2. Only then open first-published packages and run Model A vs Model B once.