big-static-data-frontend

Split the index into many small JSON shards loaded with Promise.all and merged in memory, keeping heavy detail records in separate files fetched lazily per id. Never ship the full catalog as one multi-MB file or inline script constant, since parse cost blocks time to interactive.

Precompute one lowercase haystack string per row combining name, tags, and facets, then filter with token-AND matching against that string. Add a small LRU memo on the filter signature to handle backspace-heavy typing without rebuilding strings on every keystroke.

A routing index is a small sidecar file containing per-shard token blooms or normalized word lists. The browser picks candidate shards by query overlap and fetches only those, falling back to all shards if the index is missing or yields no candidates.

Yes, the patterns rely only on fetch, JSON.parse, and plain in-memory objects or Maps, with no bundled search library. A true inverted index is listed only as an optional later step if profiling shows filtering is still slow.

A single multi-MB JSON file forces synchronous parsing of a huge string on the critical path, blocking time to interactive. Sharding the index and lazy-loading detail records spreads parse cost and lets the list render from small index files first.