SiftBenchmarks
Every number below is measured on a real Mac with the same harness you can run. No lab conditions, no cherry picking.
Cold start to a usable inbox, averaged over ten runs on Apple Silicon. Sift reads from a warm SQLite mirror, so the list is painted before the network is touched.
All interactions
Methodology
- Hardware: Apple Silicon Mac, macOS 13 or newer, warm OS caches.
- Measurement: each interaction is timed from the input event to the painted frame, averaged over ten runs, with the fastest and slowest discarded.
- Data: a local mirror of a real Gmail account with the full archive indexed in SQLite FTS5.
- What is not counted: network latency and remote image loads. Those belong to the network, not to the app. Sift never blocks an interaction on them.
About these numbers
Sift is measured on the same machine you will run it on, not in a lab. The cold-start figure is the first useful paint of the inbox with a warm OS cache. Interaction figures are input-to-paint on a local mirror, so network latency and remote image loads are not counted. Those belong to the network, and Sift never blocks an interaction on them.
Numbers vary with mailbox size and disk speed, so judge the shape of the result rather than the exact millisecond. The benchmark harness lives in tools/sift-bench in the repository; a published build is coming soon.