feat(features): implement promotion compute method (#109)

[w7-mgfcode]

Summary

Lands PRP-3.1D — the promotion compute slice for time-safe featuresets. Rebased onto dev after PR #111 (lifecycle) landed.

• Adds FeatureEngineeringService._compute_promotion_features(df, promotion_rows_df).
• Wires it at step 7 of compute_features, gated on self.config.promotion_config.
• Sidecar pattern via getattr(self, "_promotion_rows_df", None) — no new public API. Empty-DataFrame fallback when unset (matches the additive-contract invariant).
• Per-kind features: promo_{kind}_active_lag{N} (Int64 nullable) + promo_{kind}_intensity_lag{N} (float64).
• Chain-wide promotions (store_id IS NULL) handled via two-pass match — never via NaN-key merge.
• Overlapping promos on same kind/day reduce via max(discount_pct) for intensity; active stays 0/1.
• Sorted-kind, active-before-intensity column ordering (deterministic).

Time-safety

• groupby(entity_cols).shift(lag_days) is the leakage gate — same-day promo MUST NOT appear in active_lag{N} at day D.
• TestPromotionLeakage (4 cases) asserts this invariant: day-of-start = 0, day-after = 1, and chain-wide does not bleed across products.
• test_leakage.py remains the canonical spec; no existing invariants weakened.

Scope

• Files touched: service.py, tests/test_leakage.py, tests/test_service.py (3 files, +544/-0 vs dev).
• No schemas.py / routes.py / conftest.py / migration / docs edits.
• Additive-contract snapshot (config_hash_unchanged_when_phase2_omitted) still passes.
• The end-to-end loader extension (joining promotion rows from the DB) lands in PRP-3.1E.

Rebase notes

• Rebased from 09c9bc6 → 037b62f on top of new dev (which contains B's lifecycle additions).
• Conflict surface was the step-6 wiring slot in compute_features + the method-list slot just before class FeatureDataLoader: + identical class-list slots in the two test files.
• Resolution: kept B's lifecycle additions from dev, appended D's promotion additions next to them (step-7 wiring + new method + new test classes at the end of the existing class chain). One intentional non-pickup: did not propagate D's silent deletion of assert result.feature_columns == ["lag_1"] in TestComputeFeatures::test_empty_dataframe_handling — that looked like an unintended edit artifact.

Test plan

• ruff check + ruff format --check clean
• pytest -v on app/features/featuresets/ — 95 tests pass (4 new leakage + 9 new service cases on top of the existing slice)
• CI green on PR

Refs

Part of #109 (Phase 2 feature wiring). PRP-3.1E will land the loader extension that populates _promotion_rows_df from the DB join.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 69424eda-9a55-41ce-83aa-6f07a1c407df

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/featuresets-promotion

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[sourcery-ai]

Sorry @w7-mgfcode, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery