filter

func Score ¶

func Score(query, name, description string, tags []string) int

Score computes a relevance score for name/description/tags against query.

The scoring pipeline:

1. Tokenize + stopword-filter query and all fields
2. Match each query token per field via exact → stem → prefix
3. Coverage bonus: +10% per additional matched query term (beyond first)
4. Co-occurrence bonus: +30% per field where 2+ query terms match together
5. Phrase adjacency bonus: +25%/+15%/+10% when adjacent query terms appear adjacent in name/tags/desc ("code review" → "code-review" +25%)
6. Short-tag bonus: raw tag == raw query word gives +8 (enables "go", "ci")
7. Name precision bonus: +20% when ≥75% of name tokens are covered by query (rewards "jwt-auth" over "jwt-auth-middleware-guide" for query "jwt auth")

Returns 0 if no query term matches (item excluded). Returns 1 if query is empty or reduces to all stopwords (pass-through).

func SortByScore ¶

func SortByScore[T any](items []T, scoreFn func(T) int) []T

SortByScore filters and sorts items by relevance to query.

• Items scoring 0 are excluded when query is non-empty.
• Items are sorted by score descending.
• Equal-score items keep their original relative order (stable, reproducible).
• If query is empty, all items are returned in original order unchanged.

func Tokenize ¶

func Tokenize(s string) []string

Tokenize splits s into lowercase alphanumeric tokens on non-alphanumeric boundaries. Single-character tokens and stopwords are discarded.

Examples:

"JWT Authentication"      → ["jwt", "authentication"]
"how to use mcp-skills"   → ["mcp", "skills"]    (stopwords filtered)
"mcp-instructions"        → ["mcp", "instructions"]
"**/*.go"                 → ["go"]