Autocomplete Search

src/search.ts implements the deterministic autocomplete ranking used by searchByName(). It is a pure function — no I/O, no external state. It can be tested and benchmarked in isolation.

---

Score tiers

Results are ranked by score descending. Each query character length unlocks additional tiers:

Score	Tier	Min query length
1000	Exact normalized name match	1
900	Full-name prefix (name starts with query)	1
800 − n	Word-prefix (a word in the name starts with query, minus 10 per word position)	2
700 − n	Substring (query appears anywhere, minus 2 per character offset)	3
200 − n	Fuzzy — Levenshtein edit distance (minus 50 per edit)	4
< 100	Excluded from results entirely	—

Tiebreak order within a score: match position (earlier is better) → name length (shorter is better) → alphabetical.

Example — query bar:

Card	Tier	Score
Bard	Full-name prefix (bard starts with bar)	900
Barrow Stinger	Full-name prefix (barrow stinger starts with bar)	900 → tiebreak: shorter name wins
Cannon Barrage	Word-prefix (barrage is word 2, starts with bar)	790
Singularity	No match	excluded

---

autocompleteSearch(cards, query, limit)

function autocompleteSearch(cards: Iterable<Card>, query: string, limit: number): Card[]

Parameters

Parameter	Type	Description
cards	Iterable<Card>	Full card list to search against (e.g. in-memory list for local ranking)
query	string	Raw query string — normalized internally via normalizeCardName
limit	number	Max results to return

Returns

Up to limit cards, ranked by score descending.

Behaviour

• Queries shorter than 3 characters only return prefix matches — Levenshtein fuzzy is never applied at that length.
• Returns [] when the normalized query is empty.

Server-side card search in SupabaseCardProvider uses Postgres FTS instead of this function; autocompleteSearch remains useful for pure in-memory ranking (tests, clients).

---

scoreCard(card, normQuery, queryLen)

Exported for testing. Scores a single card against a pre-normalized query string.

function scoreCard(card: Card, normQuery: string, queryLen: number): ScoredCard | null

Returns null when the card does not meet the minimum score threshold. The ScoredCard result carries { card, score, position } where position is the character offset of the match in the normalized name (used for tiebreaking).

---

Fuzzy edit distance

The fuzzy tier uses a single-row Levenshtein DP implementation. Maximum edit distance is:

• 1 edit for queries of length 4–5
• 2 edits for queries of length 6+

The algorithm short-circuits early when the length difference alone exceeds maxDist, and prunes rows where the entire row exceeds maxDist, keeping allocations small.

---

Name normalization

All comparisons use normalized names produced by normalizeCardName():

// src/normalize.ts
export function normalizeCardName(name: string): string {
  return name
    .toLowerCase()
    .replace(/['\u2019-]/g, "") // apostrophes, right-single-quote, hyphens
    .replace(/[^\w\s]/g, "")
    .replace(/\s+/g, " ")
    .trim();
}

Cards are stored with a pre-computed name_normalized field. The query is normalized at search time using the same function so comparisons are always apples-to-apples.