Phase 4: preregister prospective sector-forecasting confirmation

[gigichengnc]

Why this PR exists

v0.2 retired the old coordinate-forecast headline after a structural-null test showed that much of the apparent edge was mechanically induced by the target construction. The next research question therefore uses fully future sector returns relative to SPY and separates exploratory development from prospective confirmation.

This draft PR publicly timestamps the Phase 4 prospective protocol before the first eligible official decision week.

What is frozen

• candidate: price_macro_interactions
• baseline: price_sector_fe
• horizons: 1, 2, 4, 8 and 12 weeks
• preselected primary horizon: 4 weeks
• fully future target: sector log return from t to t+h minus SPY log return over the same future window
• fixed training boundary: discovery/development targets ending on or before 2025-08-15
• no Phase 4 refitting, retuning, feature selection, horizon switching, or model replacement during the confirmation window
• minimum confirmatory sample: 52 official weekly decisions, evaluated only after all 4-week outcomes mature

RRG-style variables are not credited as the Phase 4 predictive engine because Phase 3 did not establish robust incremental RRG value after the macro-interaction block.

First prospective decision

The first eligible official decision week is 2026-08-28.

The 2026-08-21 week is excluded because the contract and coefficient freeze occurred after that market week had closed. The earlier 2026-08-14 run is a pipeline smoke test only and is not prospective evidence.

Public freeze record

This PR adds a small public timestamp record under rebuild/2026/prospective/phase4/:

• prospective research contract;
• provenance and source metadata;
• training/model freeze summary;
• SHA-256 commitments for each fixed model file;
• Phase 3 numerical reproduction gate;
• empty prospective-ledger schema;
• hashes for the public freeze files.

The coefficient JSON files and the market-data archive are not published in this timestamp PR. Their SHA-256 commitments are recorded before the first official prospective decision. Any later coefficient file used for Phase 4 must match the committed hash exactly or it is a different model.

Interpretation boundary

This PR does not claim that a genuine market-prediction edge has been confirmed.

Before Phase 4 completes, the strongest allowed wording is:

Exploratory development evidence suggests that lagged macro regimes may improve cross-sectional sector ranking. A genuine forward-looking edge has not yet been confirmed.

A successful primary endpoint would support a narrower claim of prospective 4-week sector-ranking evidence. It would still not establish exact price prediction, guaranteed alpha, or an investable strategy.

Scope

This PR does not change repository licensing, visibility, releases/tags, PyPI publication, or the v0.2 correction. It contains no prospective prediction outcomes and should remain draft while the prospective record begins.