The suite under Linux, 2026-08-15

D-022 makes Windows and Linux both first-class, and everything built on 2026-08-15 (the reader layer, D-035, and the clean run's interpreter rule) had only ever run on Windows. This is the Linux measurement, taken the same day.

What was run

The repository checkout on the Windows disk, driven from the default WSL distribution, against the same Postgres the Windows host uses:

wsl -e bash -lc 'cd /mnt/c/Users/<user>/Documents/Github/laneward &&
  DATABASE_URL=postgres://laneward:laneward@<podman-machine-ip>:5433/laneward_test bun test'

bun is installed inside WSL at /home/<user>/.bun/bin/bun, and the container published on 5433 is reachable from there.

Result

275 pass, 0 fail, 927 expect() calls, across 32 files, in 160 s.

On Windows the same commit is 274 pass, 1 skip, 0 fail. The difference is one test, not a coincidence: tests/conductor.signals.test.ts skips "SIGINT kills the children and hands the lanes back to HUB" on win32, because Windows terminates the process rather than raising a catchable event, which the roadmap records as a real gap. Under Linux that test runs and passes, so the conductor's signal cleanup is verified on the platform where it can work.

tests/git-guard.test.ts also passed, which is the first Linux evidence for the Git shim that D-023 relies on and that the README still lists as unverified.

What this does and does not establish

It establishes that the code merged on 2026-08-15 is not Windows-only: the spawn-based layers, the reader, the clean run's interpreter resolution and the per-lane database machinery all behave under Linux.

It does not establish that the system has been operated on Linux. Nothing here started a hub, ran a conductor pass, built a candidate or spawned a real codex from Linux, and the WSL distribution has no podman, so a Linux host would reach a database some other way. A green suite is evidence about the code, not about the deployment, and D-013 asks for the second before calling a platform done.