Match an error

// MATCH · ERROR → RECIPE

Paste an error, find a candidate.

Local scoring. No LLM — mechanical token overlap only.

// MATCH · ERROR → RECIPE

Paste an error or stack trace above — matches update as you type.

// 03 · WHERE IT WORKS

Three patterns the matcher catches.

1. Robust to synonyms

Paste "dtype mismatch" and recipes mentioning "datatype", "dtypes", or "numpy.dtype" surface as candidates too. The matcher normalises pairs like dtype⇄datatype, regex⇄regexp, and fork⇄spawn to a single key — you don't have to guess the project's preferred spelling.

2. Robust to typos (fuzzy match)

A single-character typo like "recursoin depth exceeded" is absorbed via Levenshtein distance 1 (limited to tokens of length ≥ 6 to avoid false positives on short tokens, which stay strict). Fewer "I know there's a recipe for this but it isn't matching" surprises.

3. Mixed-language input is fine (multi-language stopwords)

Pasting "pandas で empty DataFrame の dtype が変わる" — Japanese meta-text mixed with English technical terms — works because Japanese stopwords ("で", "の", "が") and English stopwords ("the", "a", "is") are both stripped. Stack-traced log lines paste in as-is.