fix(heartbeat): scale pulsetime with poll_time to prevent missing time

[TimeToBuildBob]

Problem

At poll_time > 2s, OS scheduling jitter regularly exceeds the fixed 1-second
pulsetime margin (poll_time + 1), breaking heartbeat chains and causing
missing time in the activity timeline.

From a simulation (200 trials, 3-hour session, σ = 15% × poll_time) validated
against @TimoRenk-nebumind's measured data in ActivityWatch/activitywatch#1177:

poll_time	old pulsetime	time lost	new pulsetime	time lost
1s	2.0s	0.00%	2.0s	0.00%
2s	3.0s	0.07%	3.0s	0.07%
5s	6.0s	11.6%	7.5s	0.09%
10s	11.0s	30.0%	15.0s	0.12%

The fix explains why users with default poll_time=1s never see the problem
while users who increase polling to 5–10s lose significant session time.

Fix

# Before
pulsetime=poll_time + 1.0

# After
pulsetime=max(poll_time * 1.5, poll_time + 1.0)

max() (not min()) ensures:

• Backward compatible: at poll_time ≤ 2s, the formula returns poll_time + 1
  (same as before)
• Scales at higher intervals: at poll_time = 5s, pulsetime becomes 7.5s instead
  of 6.0s, absorbing the ~0.75s jitter typical at that interval

Note: @mrienstra's original min() proposal in ActivityWatch/activitywatch#1177
actually made things worse at high poll_time — max() is the correct operator.

Testing

Added parametrized regression test covering the four key boundary cases.

Closes ActivityWatch/activitywatch#1177

[greptile-apps]

Greptile Summary

This PR fixes missing time in the activity timeline when poll_time is larger than 2s, by replacing the hardcoded poll_time + 1.0 pulsetime with max(poll_time * 1.5, poll_time + 1.0) — extracting the logic into a dedicated compute_pulsetime helper that is also directly unit-tested.

• compute_pulsetime is extracted to main.py and called from heartbeat_loop, preserving the old behavior for poll_time ≤ 2s while scaling the jitter tolerance proportionally at higher polling intervals.
• Parametrized regression tests are added in tests/test_main.py that import the production helper directly, verifying all four boundary cases (1 s, 2 s, 5 s, 10 s).

Confidence Score: 5/5

Safe to merge — the change is a small, well-bounded formula update with no regressions at the default poll interval.

The formula max(poll_time * 1.5, poll_time + 1.0) is mathematically straightforward, its crossover point (poll_time = 2 s) is verified by the parametrized tests, and behavior at the default 1 s poll interval is identical to before. The helper is extracted rather than inlined, so both production code and tests operate on the same implementation. No risky side-effects, migrations, or interface changes are involved.

No files require special attention.

Important Files Changed

Filename	Overview
aw_watcher_window/main.py	Adds compute_pulsetime helper and wires it into heartbeat_loop; logic is correct and backward-compatible.
tests/test_main.py	Adds parametrized regression test that imports and exercises the production compute_pulsetime function directly, covering the four documented boundary values.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant HB as heartbeat_loop
    participant CP as compute_pulsetime(poll_time)
    participant C as aw-client

    HB->>CP: compute_pulsetime(poll_time)
    CP-->>HB: "max(poll_time*1.5, poll_time+1.0)"
    Note over CP: poll_time ≤ 2s → poll_time+1.0<br/>poll_time > 2s → poll_time*1.5
    HB->>C: "client.heartbeat(bucket_id, event, pulsetime=…, queued=True)"
    C-->>HB: ok
    HB->>HB: sleep(poll_time)

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant HB as heartbeat_loop
    participant CP as compute_pulsetime(poll_time)
    participant C as aw-client

    HB->>CP: compute_pulsetime(poll_time)
    CP-->>HB: "max(poll_time*1.5, poll_time+1.0)"
    Note over CP: poll_time ≤ 2s → poll_time+1.0<br/>poll_time > 2s → poll_time*1.5
    HB->>C: "client.heartbeat(bucket_id, event, pulsetime=…, queued=True)"
    C-->>HB: ok
    HB->>HB: sleep(poll_time)

Loading

_{Reviews (2): Last reviewed commit: "refactor(heartbeat): extract compute_pul..." | Re-trigger Greptile}

[TimeToBuildBob]

Fixed in a1bd4dc: extracted as a module-level function and updated the test to import and call it directly. The test now exercises the production code path, so a formula change in main.py will break the test as expected. No behavior change — the formula itself is unchanged.

[TimeToBuildBob]

@greptileai review

[TimeToBuildBob]

All gates clear and ready to merge:

• ✅ CI green (all platforms, tests, typecheck, CodeQL)
• ✅ Greptile reviewed (5/5 confidence, safe to merge)
• ✅ Review threads resolved
• ✅ Feedback addressed

Waiting for maintainer to merge.