Performance
Throughput, frame time, cold start, memory per pane, idle cost, and ConPTY round trip were measured on one machine using ReleaseFast builds. The methodology document records each figure with its date, run count, and the state the machine was in.
Two results miss a budget stated in PRODUCT.md: memory per additional pane, and the cold-start tail, where the median fits and the p95 does not. Both came from an older binary and have not been re-measured on the current code, so this page does not repeat them. Key-to-pixel latency is not measured at all; the suite has a software proxy that cannot stand in for it.
No comparison against another terminal is published, here or anywhere else in the project.
Benchmark methodology and figures