Knowledge Ranking Algorithm

Canonical source: docs/14-ai-native-development/knowledge-ranking-algorithm.md — this page is rendered from it on every publish; edit the source.

Knowledge ranking algorithm

Deterministic lexical/structural retrieval:

• Task terms: lowercased tokens of length > 2, de-duplicated.
• Score: count of task terms found in a candidate's text (name, models, patterns, doc body).
• Boosts: explicit --service/--event/--route pin the result; the matching task pack adds a boost to its dimension.
• Ties: broken by stable cmp (code-unit order) for reproducibility.

No external AI provider is required. Optional semantic retrieval may be added behind a feature flag later, but deterministic retrieval stays canonical.