How AtoAxis calculates coding time

A heartbeat is an event from a connected editor. It can include a timestamp, file and project metadata, branch, language, editor and activity counts. It describes observed editor activity; it does not prove that every minute spent planning, reading or talking produced coding time.

• The timestamp places the event on the account timeline.
• Metadata lets the same interval be filtered by project, language, branch or editor.
• Counts and attribution fields remain separate from the duration calculation.

AtoAxis canonicalises overlapping evidence first, then orders the remaining intervals. The duration of a block is the union of its intervals, so two sources cannot count the same seconds twice. A single isolated event proves an instant but does not invent a long session after it.

09:00 interval ───── 09:05 interval ───── 09:10 interval
          one continuous duration block

The default human idle timeout is 15 minutes. When two proven intervals are separated by a positive gap no longer than that timeout, the gap is bridged to the earlier interval, matching the WakaTime-compatible session model. A gap longer than the configured timeout closes the block. The timeout is configurable per deployment, so the deployed setting is the source of truth.

• A short gap between two real events can be included in the earlier block.
• A gap longer than the timeout starts a new block instead of padding the old one.
• Time after the final event is never added just because the session was once active.
• Agent activity can use its own timeout when that deployment enables the split timeline.

Illustrative example, not customer data: suppose an interval ends at 09:00 and the next begins at 09:10. With the default 15-minute timeout, that 10-minute gap is bridged. If the next interval begins at 09:30, the 20-minute gap is too long, so the first block ends and a second block begins. Time after the final interval remains uncounted.

09:00 ──(10 min gap)── 09:10    bridged
09:10 ──(20 min gap)── 09:30    new block

The arithmetic explains the rule, not a promise about a particular dashboard total. Check the timeline and its interval evidence when a total looks surprising.

• Coding time is observed editor activity, not all time spent on a task.
• A duration total does not measure code quality, value, focus or productivity.
• Use project, language, branch and editor filters to find which evidence produced a total.
• Compare the duration with the activity timeline and coverage fields before drawing a conclusion.

Connect one editor, run wakaclock-cli status, make a small change and wait for the first heartbeat. The dashboard timeline should show the event and the Overview page should use the same duration model. If it does not, check the API key, server URL and local queue before adding another client.

Review owner: AtoAxis product team. Last reviewed: 20 September 2026.