@AetherClaude changes applied.

Reviewed the diff — implementation is clean and the wiring is correct. A few notes:

Feedback-loop avoidance looks right. The editingFinished handler sets the edit text first, then calls m_*Slider->setValue(v). The slider's valueChanged callback re-checks hasFocus() on the edit before setting text, so:

• Enter pressed (focus stays on edit) → slider's setText is skipped, edit already shows the clamped value.
• Focus lost (Tab/click) → slider's setText runs, but it's a no-op write of the same value.

Either way the model setter fires once via the slider path, so the existing command-emit path is preserved. Good call routing through the slider rather than calling m_model->setCwDelay(v) directly from the edit handler — keeps a single source of truth.

syncCwFromModel is safe. The explicit if (!hasFocus()) setText(...) after each slider->setValue is technically redundant (the slider's own valueChanged lambda would set the text), but it's defensive and harmless — and makes the sync read clearly when scanning the function. Worth keeping.

Pitch path is consistent. Since pitch has no slider, editingFinished calls setCwPitch directly, mirroring what the up/down buttons already do. Note editingFinished can fire twice on the same edit (once on Enter, again on focus-loss); the clamp + setText is idempotent so this is fine.

Empty input edge case. text().toInt() returns 0 on empty input, and qBound snaps it to each field's minimum (0/5/0/100). Reasonable behavior — users get a sane floor instead of an error.

Minor accessibility nit (optional). The new line edits set setAccessibleName but no setAccessibleDescription (matching the old labels). The corresponding sliders have descriptions like "CW break-in delay in milliseconds" — could mirror those on the edits for screen readers, e.g. m_delayEdit->setAccessibleDescription("CW break-in delay in milliseconds, 0 to 2000"). Not blocking.

Net: looks good to me. The hasFocus() guards in both directions are the key correctness piece and you've got them right.

@AetherClaude A11Y changes added.

A11Y additions look good — all four edits now carry descriptions and they consistently extend their corresponding slider descriptions with the valid range:

• Delay edit: "CW break-in delay in milliseconds, 0 to 2000" ↔ slider "CW break-in delay in milliseconds"
• Speed edit: "CW keying speed in words per minute, 5 to 100" ↔ slider "CW keying speed in words per minute"
• Sidetone edit: "CW sidetone monitor volume, 0 to 100" ↔ slider "CW sidetone monitor volume"
• Pitch edit: "CW sidetone pitch in Hz, 100 to 6000" (no slider counterpart, but matches the same <purpose>, <min> to <max> shape)

Including the range on the editable widget is the right call — for a typed input, screen reader users benefit from knowing the bounds before they start typing, where on a slider it's implicit from the track. Pattern is consistent with how m_micLevelSlider already documents its range elsewhere in this file, so nothing feels out of place.

Nothing else to flag from me — implementation, feedback-loop guards, and accessibility are all in good shape. LGTM.

Claude here — clean SmartSDR-parity feature, CJ. The hasFocus() guards on the slider valueChanged paths AND on syncCwFromModel are exactly the right shape — mid-edit input is preserved against radio echoes, and Enter-then-loss-of-focus still routes through the slider so the existing model command path fires unchanged. Validator + qBound belt-and-braces is also good (validator blocks bad chars, qBound clamps out-of-range partial input). Pairs nicely with #2428 — that fix was the prerequisite for these edits to actually persist. Merged.

73, Jeremy KK7GWY & Claude (AI dev partner)