Tooling to measure two things we used to guess at: how long a cold launch takes, and how many files we make the OS (and Defender) scan. These numbers are the baseline that Phase 6 (slimming) and Phase 7 (Squirrel vs NSIS) are judged against.

Boot timing

electron/main.js emits four ordered milestones to backend.log, one line each:

[perf] app-launch t=<ms>
[perf] first-paint t=<ms>
[perf] backend-http-ready t=<ms>
[perf] first-agent-response t=<ms>

t is milliseconds since process start. first-agent-response is fired by the renderer (WebSocketManager.dispatchDelta) on the first streamed agent token.

Read and assert them after launching a packaged build:

node scripts/perf/parse-timing.js          # human table, exits 1 if missing/unordered
node scripts/perf/parse-timing.js --json    # machine-readable
node scripts/perf/parse-timing.js --log <path-to-backend.log>

backend.log lives next to auth.token:

macOS: ~/Library/Application Support/OpenSwarm/data/backend.log
Windows: %APPDATA%\OpenSwarm\data\backend.log
Linux: ~/.local/share/OpenSwarm/data/backend.log

File count / size

node scripts/perf/file-count.js                       # auto-detects packaged tree
node scripts/perf/file-count.js --root <resources-dir> # explicit
node scripts/perf/file-count.js --json

Recorded baseline (Windows, win-unpacked/resources)

dir	files	size
python-env	8421	373.6 MB
router	2651	35.9 MB
backend	86	8.2 MB
frontend	11	232.3 MB
debugger	75	1.2 MB
other	3	621.3 MB
TOTAL	11247	1.2 GB

python-env is by far the largest file count, so it dominates Defender's per-file scan cost on first launch. That is the Phase 6a target.

Tests

node scripts/perf/test-perf.js builds throwaway fixtures with known counts and known timing logs, runs both tools against them, and asserts exact numbers plus that the failure paths fail. Hermetic; needs no packaged build. 17 assertions.