Rework dispatch capacity architecture beyond TTL mitigation #1192

Merged

coilysiren merged 1 commit from issue-1191 into main

2026-07-13 21:03:29 +00:00

coilyco-ops commented

2026-07-13 20:22:48 +00:00

Owner

Why: the old reservation shape treated a launch intent as active capacity, which could block repo slots before a visible engineer existed. The new split keeps intent, visibility, and terminal release separate so list, dispatch-health, and reap report the real state model.

list now reports running engineers and launch intents separately.
dispatch-health reports launch intents independently of running workers.
Repo working capacity ignores non-visible launch intents, and visibility failures clean up immediately.
Docs and the feature inventory were updated to match the split model.

Tests: ward exec test, ward exec lint, and pre-commit run --all-files all passed locally.
closes #1191
ward.workflow: pull-request-and-merge

This reworks dispatch capacity around visible workers instead of prelaunch intents. The launch broker still tracks intent and release state, but repo/global capacity now counts only running engineers, and failed visibility clears the launch intent immediately instead of waiting for TTL-based recovery. Why: the old reservation shape treated a launch intent as active capacity, which could block repo slots before a visible engineer existed. The new split keeps intent, visibility, and terminal release separate so list, dispatch-health, and reap report the real state model. - `list` now reports running engineers and launch intents separately. - `dispatch-health` reports launch intents independently of running workers. - Repo working capacity ignores non-visible launch intents, and visibility failures clean up immediately. - Docs and the feature inventory were updated to match the split model. Tests: `ward exec test`, `ward exec lint`, and `pre-commit run --all-files` all passed locally. closes #1191 ward.workflow: pull-request-and-merge

coilyco-ops added 1 commit

2026-07-13 20:22:48 +00:00