Context

PR #3202 introduced absolute, drift-corrected scheduling in CwxLocalKeyer to fix macOS sidetone stutter under GUI load (#2980). The implementation is correct and was manually verified by @jensenpat on macOS with the canonical "30 S characters + aggressive panadapter drag" methodology.

What's missing is an automated regression test for the drift correction itself. The existing cw_sidetone_test and cwx_panel_test were written against the relative-scheduling path and pass against the new code, but they don't specifically exercise the late-tick-recovery behaviour that the new m_elapsed + m_nextEdgeMs machinery exists to provide.

Proposed test shape

tests/cwx_local_keyer_drift_test.cpp:

1. Construct a CwxLocalKeyer and start a deterministic macro (e.g. 10 × S — pure dit pattern, easy timing).
2. Drive scheduling synchronously by intercepting the timer callback rather than letting Qt's event loop fire — call onTick() manually with a simulated QElapsedTimer offset that lags behind the intended timeline.
3. After each onTick, assert that m_nextEdgeMs continues to track the intended cumulative time (sum of unit durations) and that the requested next-wait shortens when the simulated clock is ahead of the target.
4. Add a "sustained overload" case where simulated elapsed time exceeds m_nextEdgeMs — verify wait clamps to 1 ms and the keyer fires elements back-to-back to catch up.
5. Cross-macro continuity case: enqueue two start() calls back-to-back, drain the first, verify epoch survives into the second and timing continuity holds.
6. Symmetric reset cases: stop() and final drain both invalidate m_elapsed and zero m_nextEdgeMs so the next run starts fresh.

Why this matters

CwxLocalKeyer::scheduleNext is the hot path for local sidetone cadence and the drift logic is subtle (absolute target time vs relative wait, partial-drain keeping epoch alive, recursive scheduleNext re-entry on empty-but-queued state). A future refactor — especially one that touches the m_currentlyDown + CharGap insertion at line 117 or the recursive call at line 122-124 — could quietly regress the drift behaviour and only surface as "macOS contest CWX feels stuttery again."

Scope

Single new test file + CMakeLists.txt entry. No production-code changes. The QElapsedTimer injection point may need a tiny seam in CwxLocalKeyer (or a friend-class hook in the test) — author's call.

References

• PR [cw] drift-correct local CWX sidetone keyer #3202 — the drift-correction landing
• Issue CWX: Saved Messages stuttering and timing spacing incorrect on macOS (CW timing issues) #2980 — original macOS stutter bug report
• PR HOLD FOR TESTING - fix(cwx): drift-correct local sidetone keyer to stop macOS stutter (#2980). Principle I. #3109 — earlier HOLD-FOR-TESTING attempt (closed, superseded by [cw] drift-correct local CWX sidetone keyer #3202)
• CwxLocalKeyer.cpp:143-159 — the scheduling code under test

73, Jeremy KK7GWY & Claude (AI dev partner)