GitHub
npm install -g @fuzit/cli
Fuzit publishes deterministic retrieval metrics plus real-repository timing and selectivity measurements, with reproducible methodology and hardware environment attached.
Deterministic rank quality across offline V1 retrieval fixture suites. No LLMs or external APIs involved.
Measured scan latencies, context selection latencies, and file-set reduction across 5 open-source repositories under a fixed 8,000 token budget.
| Repository | Language | Commit | Discovered | Selected | File-Set Reduction | Scan Median | Context Median | Context p95 | Budget Used |
|---|---|---|---|---|---|---|---|---|---|
| fuzit | TypeScript | e50f099e | 854 | 4 | 99.53% | 1879.3 ms | 3560.9 ms | 3660.0 ms | 7999/8000 |
| fastify | JavaScript | 2e81c38b | 390 | 5 | 98.72% | 1415.7 ms | 2327.9 ms | 2544.5 ms | 7995/8000 |
| flask | Python | 6a2f545b | 235 | 9 | 96.17% | 1112.2 ms | 1804.2 ms | 1894.1 ms | 8000/8000 |
| cobra | Go | adbc8813 | 66 | 3 | 95.45% | 726.7 ms | 973.7 ms | 1073.1 ms | 7998/8000 |
| spring-petclinic | Java | 88e37c15 | 131 | 4 | 96.95% | 1012.5 ms | 1408.3 ms | 1428.3 ms | 7996/8000 |
(1 - selectedFiles / discoveredFiles). It is strictly a file selectivity metric and is separate from token compression.Cold vs repeated CLI process startup measurements and peak memory allocation.
Explicit benchmark measuring canonical index reconciliation behavior on a synthetic 50k record set.
All 5 release validation check suites passed prior to publishing version 0.0.9 benchmark evidence.
Without complete disclosure of hardware environment, runtime version, and fixed budget constraints, numbers cannot be reproduced.