Verified Reports
Performance Evidence
These pages publish benchmark loops, retained changes, raw data, and validation commands used for recent ivygrep performance work.
Reports
- Evidence and claim dashboard - immutable quality, latency, footprint, release-size, portability, variance, and comparison-status history generated from retained artifacts.
- Daemon hot-query cache benchmark - warm daemon query p95 improved from 455.37 ms to 4.90 ms while preserving daemon/local result equivalence.
- Linux kernel cold-index benchmark - fresh-index primary score improved by 10.62% on a 93,493-file Linux checkout.
- Public code-retrieval benchmark - pinned CoIR quality, latency, indexing, memory, size, and variance evidence with raw machine-readable results.
- Public million-chunk benchmark - deterministic generated-corpus indexing, interleaved warm-query p95, full-system paths, quality, and CI regression evidence.
- Portable embedding model bake-off - public quality, latency, memory, and model-selection evidence for the default neural profile.
- Bounded learned reranker - held-out public quality, per-task regression, latency, model identity, and deterministic-fallback evidence.
- CI benchmark dashboard - continuous Criterion benchmark history from GitHub Actions.