Initial Commit

openspec/archive/2026-04-19-enterprise-dedup/specs/dedup/spec.md

@@ -0,0 +1,27 @@
+# Capability Spec: dedup
+
+**Package**: `@suaveplan/dedup`
+**Version**: 0.0.0 (genesis)
+**Status**: ADDED
+
+---
+
+## ADDED Requirements
+
+### DD-01 — Match keys
+`MatchKey` MUST be `{ field, extract?, normaliser?, weight }`. The engine MUST support multiple match keys per config and combine their contributions by weighted sum, normalised to `[0, 1]`.
+
+### DD-02 — Similarity primitives
+`exactMatch(a, b)` MUST be `a === b`. `levenshteinDistance` MUST compute edit distance in O(|a|·|b|). `jaroWinklerSimilarity` MUST return a value in `[0, 1]`. `computeSimilarity(a, b, rules)` MUST apply the configured `ScoringRule[]` and return a composite score in `[0, 1]`.
+
+### DD-03 — Detect contract
+`DedupEngine.detect(entities, config)` MUST return `DuplicateGroup[]` where each group has `{ representative, members, scores }`. Every entity MUST belong to at most one group. Entities below `config.threshold` MUST NOT be grouped.
+
+### DD-04 — Merge strategies
+`latestWinsStrategy` MUST pick the entity with the most recent `updatedAt`. `manualStrategy` MUST defer to the caller via a resolver callback. `scoredMergeStrategy` MUST merge fields per-field using the highest-scored source. All strategies MUST return `MergeResult { merged, sources }`.
+
+### DD-05 — Observability
+Every `detect` call MUST record `dedup.detect.duration_ms` histogram and `dedup.groups.total` counter. Every merge MUST record `dedup.merge.total{strategy}`.
+
+### DD-06 — Configuration validation
+`DedupConfig` MUST be validated via Zod. Invalid configs MUST surface as `DEDUP.INVALID_CONFIG`.