What ripgrep says, vs. what it scored.
Line-oriented regex search CLI -- no AST or symbol awareness, the fastest possible answer to a text-match question.
“It can replace many use cases served by other search tools because it contains most of their features and is generally faster.”
| Tool | MRR | p95 | p100 |
|---|---|---|---|
| ★ Atelier +semantic (BGE) | 0.727 | 390ms | 1057ms |
| ★ Atelier lexical (default) | 0.676 | 134ms | 319ms |
| ripgrep | 0.376 | 66ms | 522ms |
The one tool here that's honest about benchmarking: ripgrep's README and its creator's own 25-scenario benchmark post ("ripgrep is faster than {grep, ag, git grep, ucg, pt, sift}") are real, reproducible, and about raw text-search speed. What's never published is an accuracy number -- ripgrep has no model of code structure, so on our matched 7,213-query set it scores 0.376 MRR: a plain grep, used correctly, finds the right code less than 4 times out of 10.
The true story
Every tool in this comparison, ripgrep included, has been through the exact same 14 repositories and 7,213 query/gold pairs that score Atelier — no cherry-picked queries, no separate corpus. Full methodology, every raw number, and the other 9 tools →