JMCP: Papers - jmcp/extract-recommendation-scoring-weights-i-AnLIKEry

[landigf]

Review: Extract recommendation scoring weights into config

Change summary: Moves hard-coded numeric weights from scoreFeedPaper and scoreTrendingPaper in packages/db into two named as const config objects (feedScoringWeights, trendingScoringWeights) exported from packages/contracts. The scoring functions now reference w.<field> instead of magic numbers. No logic changes — all values are identical to the originals.

Validation confidence: High

• Pure mechanical extraction — every literal maps 1:1 to the config field with the same value (5, 4, 3, 2, 2, 24 for feed; 4, 3, 3, 16 for trending).
• as const satisfies Record<string, number> gives both literal types and structural safety.
• No new runtime code paths; scoring formulas are structurally identical.

Risks that still remain:

1. ScoringWeights union type is loose — it's a union of two structurally different objects (feedScoringWeights has interestMatch/topicOverlap that trending lacks). Consumers accepting ScoringWeights won't get compile-time safety on field access without narrowing. If no consumer uses this type today it's harmless, but consider making it a discriminated union or splitting into two explicit types if you plan to pass weights generically.
2. No tests exercise the weights indirectly — if the scoring functions already have test coverage, this is fine. If not, a future typo in the config (e.g. changing a weight thinking it's one scorer but it's the other) would go undetected.
3. recencyMaxDays is a ceiling, not a weight — naming it alongside multiplicative weights could mislead a future contributor. A comment or grouping would help, though low priority.

Overall: clean, low-risk refactor. Good to merge after npm run test && npm run check passes.

[landigf]

Closing: will integrate valuable features directly on main to avoid cascade merge conflicts.