60ms was too tight in practice — rolled back to the 400ms ceiling. The fail-open accept tap in #382 remains as the swallow fix.

Greptile Summary

Reverts the hostPublishWaitCeilingMs constant from 60ms (introduced in #383) back to 400ms, and replaces the doc comment with an empirical rationale. The #382 fail-open accept-tap fix is unaffected.

• Restores the AX-publish poll ceiling to 400ms so Chrome's slower contenteditable publish on a busy page is reliably captured before generation fires.
• Updates the doc comment to drop the now-stale Fail-open accept tap: never consume bare printables, never consume without verified session #382/60ms justification and describe the value as empirically chosen for responsiveness vs. AX accuracy.

Confidence Score: 5/5

Safe to merge — single-constant revert with no logic changes.

The only change is restoring a timeout constant from 60ms to 400ms and refreshing the accompanying doc comment. The polling logic (pollForHostPublish) is untouched; the fail-open accept tap from #382 is untouched. No new code paths are introduced.

No files require special attention.

Important Files Changed

Sequence Diagram

sequenceDiagram
    participant CGEvent as CGEvent Tap
    participant SuggC as SuggestionCoordinator
    participant FocusM as FocusModel (AX)
    participant Predict as Prediction Pipeline

    CGEvent->>SuggC: keystroke captured
    SuggC->>SuggC: schedulePredictionAfterHostPublishDelay()
    SuggC->>FocusM: refreshNow() (snapshot baseline)

    loop Poll every 30ms, up to ~370ms (ceiling 400ms)
        SuggC->>FocusM: refreshNow()
        alt AX state changed (text / element / selection)
            SuggC->>Predict: schedulePrediction()
        else "Ceiling reached (nextElapsed >= 400ms)"
            SuggC->>Predict: schedulePrediction() [fallback]
        else Still waiting
            SuggC->>SuggC: asyncAfter(30ms) → pollForHostPublish
        end
    end

_{Reviews (1): Last reviewed commit: "Revert "Drop host-AX-publish ceiling to ..." | Re-trigger Greptile}