# H07-E pick justification (one page)

**Date:** 2026-09-13T23:23:41Z UTC · **Hub job:** `77356229-193e-4693-a90c-cae8dcb0bc4a`  
**Ops:** H01-E KILL; H04/H05 MECHANICAL (inventory→STEO); H06-E KILL (HDD/CDD→STEO sales). James: keep going — DIFFERENT cheap exploratory on runnable public data.

## Chosen child
**H07-E** — EIA Weekly Petroleum Status **finished motor gasoline + distillate product supplied** (`WGFUPUS2` + `WDIUPUS2`) → STEO Table 4a **motor gasoline + distillate consumption** (`MGTCPUSX` + `DFTCPUS`) revision across adjacent STEO issues.

## Why runnable today
| Leg | Source | Status |
|-----|--------|--------|
| Physical | EIA dnav hist xls WGFUPUS2w + WDIUPUS2w | HTTP 200; gas sha `13cc828d02f57ac1…`; distillate sha `bb1ae189d7a56944…`; n_weeks=1857 (1991-02-08→2026-09-04); hist→vintage unverified |
| Outlook | STEO archives `*_base.xlsx` | **141/141** jan2015–sep2026 already on disk (shared H04/H05/H06) |

## Why not other lanes this cycle
- **H04/H05 inventory→STEO:** MECHANICAL — do not redo stocks.
- **H06 weather→sales:** KILL — do not redo DD.
- **H01 corn / FT900 / ESR / AAR:** blocked/killed / archive digs — James: do not dig.
- **Baker Hughes / DPR→STEO production:** STEO Table 10a already embeds RIGS* — closer to mechanical mediation; skip this cycle.
- **EIA-930→STEO generation:** gridmonitor 503 / first-print UNKNOWN — not cheaper than WPS today.

## Test shape (H01-E / H04-E discipline)
Frozen baseline vs +one feature; RMSE; ≥5% bar; EXPLORATORY / CAPITAL PROHIBITED / CLAIM_STRENGTH LOW / MODEL_CHANGE=NO.

## Mechanism (hypothesis only)
Realized weekly product-supplied intensity between STEO issues → next official gasoline+distillate consumption path may revise. Incremental predictiveness only — not capital.

## Mechanical risks (upfront)
1. STEO near-history may absorb WPS/PSM product-supplied actuals (same concept family).
2. Live hist xls revises — not first-print.
3. Release timing SCHEDULE_PROXY (Fri week-end + 5d).
4. COVID 2020 demand shock in holdout.