[adenhertog]

Summary

Make two timing-dependent bus-core tests deterministic so they stop failing on loaded CI runners.

Background

The dispatcher spec "when it is stopped after a send timed out" failed on CircleCI build 1359 (stop() took 28 ms, expected at least 30 ms). workflow/started.spec.ts sleeps 2 s per step inside a 5 s beforeAll timeout.

Problem

• The dispatcher test measured how long stop() took after polling for the timeout warning every 5 ms, so less time was left on a busy runner.
• started.spec.ts slept a fixed 2 s after each message, which can exceed the test timeout in a loaded full run with coverage.

Approach

• The slow send now waits on a promise the test controls. The test asserts stop() hasn't resolved while that send is pending, releases it, and asserts stop() then resolves.

• started.spec.ts uses an InMemoryQueue and await queue.idle() instead of sleep.

• The dispatcher's "started with a periodic task" test (failed on build 1371) and its "told a message was scheduled" test use Jest fake timers and a fixed jitter, asserting what ran at exact points in time instead of measuring real elapsed time.

• "when sending is paused and there is nothing to probe" and "when probes keep failing" also use fake timers, so their pause gaps are exact (including the wake timer's 1 ms margin) instead of real-time upper bounds.

• "stopped after a send timed out" now samples whether stop() resolved right after the send finished, so it fails if stop() waits the full sendTimeoutMs.

• Tests only. The dispatcher spec passed 20 runs in a loop; unit suite, lint and format check pass.

• This is original work under the clean-room policy, not ported, translated or copied from another messaging framework

• No changeset needed (tests only, no user-facing change)

• Docs: none needed (tests only)

🤖 Generated with Claude Code