Cold start is process launch to first interactive frame. Time to render is input to painted frame, reported as p50 and p95 over samples. Both are measured on named machines by a pinned harness command; both are compared only against the previous run on the same machine, built the same way, on a machine that had nothing else running.
Cold start
Process launch to first interactive frame: spawn the binary, wait for the TUI to accept input and paint a frame that responds to it.
Best measured cold start (p50)
2.63 s
firstInteractiveFrameMs on linux-x64-amd-ryzen-5-7600x-6-core-processor-12 · 2026-10-03
source run
linux-x64-amd-ryzen-5-7600x-6-core-processor-12 · source run
Harness 2 · source run · bun scripts/bench/cold-start.ts --runs 20 --json · bun 1.4.2 · run cold-start-20261003T085127Z-9adsa · head at finish fe3e6ee99922c75c128c02a50469d06f12dec26d
Time to render
Input to painted frame: submit an input event in a live TUI session and time until the resulting frame is painted. Reported as p50/p95 over samples.
Best measured time to render (p50)
32.8 ms
inputToPaintMs on linux-x64-amd-ryzen-5-7600x-6-core-processor-12 · 2026-10-03
source run
linux-x64-amd-ryzen-5-7600x-6-core-processor-12 · source run
Harness 2 · source run · bun scripts/bench/time-to-render.ts --runs 200 --json · bun 1.4.2 · run time-to-render-20261003T085656Z-eg5oa · head at finish fe3e6ee99922c75c128c02a50469d06f12dec26d
Recorded, not published
These runs recorded more than one same-benchmark run active on the machine (machine.concurrentRuns > 1), so they were measured on a contended machine. The contract records a noisy run and does not publish it: no row, no chart, no delta, and no baseline for the run after it.