What the measured evidence supports
The historical synthetic benchmark recorded a retrieval foundation at 10,000 notes: search p95 of 56.345 milliseconds, a 22.037-second full index, a 0.503-second incremental update and 24.558-second full validation.
The environment was macOS 26.6.1 arm64, Python 3.12.12 and SQLite 3.50.4. These component-level observations came from a preserved historical run. They do not establish current performance, relevance, end-to-end AI answer time, sustained team load or improvement over Sage. The summary lacks query-count diversity and repeated-run distribution.
Read the annotated benchmark, including the exact source revision and recorded acceptance thresholds. Those thresholds are targets, not previous baselines.
Lightbox preview
Presentation notes
10,000 · Synthetic notes in the historical benchmark
56.345 ms · Historical search p95, not end-to-end AI answer time
Historical synthetic retrieval measurements; no claim of current production performance or AI answer quality.