[fartzy]

What If history requests could pile up, and the latest research, model/data snapshots, and broker fills were still local. This brings the complete November 11 stack into main, preserving the original logical commits on the published branch heads.

Changes:

• Fix bounded/cancelable What If loading and collapse the Jev detail display.
• Document the 57 active opening-pattern and 19 dip inputs; preserve the literature and intraday-context studies.
• Add a reproducible review of 93 excluded inputs and 15 pattern-evidence candidates. Audit all 190 prediction files against the frozen panel. Keep the live model unchanged: the Rank pattern-age candidate is promising but its interval includes no improvement, and the regression screen is not the deployed SVM-stacked Fit model.
• Preserve immutable news observations, article revisions, coverage failures, and classifier provenance; refresh saved morning candidates during the existing nightly job. The initial 39-candidate refresh preserved 518 article observations.
• Check in price/feature/weather snapshots, trained models and training history, broker ledger updates through the final November 11 CSQR fill, Jev/paper-book outcomes, and verified runtime archives.

Validation:

• Pattern-evidence and feature-review Bazel tests passed, including timing, live/replay parity, selection boundaries, and top-one return calculation.
• News store, ingestion, and existing news-feature Bazel tests passed. News collection and nightly targets built successfully.
• Earlier commits passed the intraday-context/collector/entry-model tests and TypeScript checks plus the affected What If/Jev/request tests; details are preserved in the local check-in audit.
• Every original changed Parquet file was read and validated; transaction-consistent database archives passed integrity/decompression checks. The trade ledger was reconciled with CSV and broker order IDs; repeated confirmations were deduplicated.
• No newly published blob exceeds GitHub's 100 MiB limit. The reproducible 810 MB research panel remains an ignored local cache; its input hashes, protocol, predictions, and result audit are committed.

The repository permits squash merges only; the 12 source branches retain the original logical commit history.

This repository is public. Publication of the included broker records, databases, models, news archives, and data snapshots was explicitly authorized by the repository owner.

References:

• docs/archive/local-checkin-2026-11-11/README.md
• docs/active-price-pattern-features-2026-11-11.md
• research/feature_revisit_2026_11_11/README.md
• docs/news-observations-2026-11-11.md