sprint-cadence

Sprint Calendar (Established March 13, 2026)

Sprint	Start	End	Status
Micro Sprint (Pipeline Fix)	2026-03-13	2026-03-15	ACTIVE
Sprint 1	2026-03-16	2026-03-31	Planned
Sprint 2	2026-04-01	2026-04-15	Planned
Sprint 3	2026-04-16	2026-04-30	Planned

Why: Pipeline was broken — 67 zombie processes on QC2, 45+ failed tasks, no completions. Micro sprint to stabilize before Sprint 1 starts.

How to apply:

• Sprint calendar lives at tasks/sprint-calendar.json
• Sprints end 15th and last day of month (bi-weekly)
• Release meetings on last day with Amen Ra (required)
• Daily standups at 1:30 PM ET (automated Slack post)
• WIP limit: 3 agents per farm, 8 total system
• phase0-workqueue-loop.sh is DISABLED on all farms — haiku-dispatcher.js (every 5 min) is the ONLY dispatcher

Key Fixes (March 13, 2026)

1. Killed 67 zombie agentic-loop processes on QC2
2. Disabled phase0-workqueue-loop.sh cron (zombie factory) on QC2 and QC3
3. Archived 45 failed + 4 stale dispatched tasks to tasks/archive/
4. Hard-reset both farm repos to clean remote state
5. Removed duplicate pending tasks (REQ-S1-04, REQ-S1-05)
6. Dry run successful — feedback-integration-test.md dispatched to QC2

Pipeline (Active)

haiku-dispatcher.js (every 5 min) → classifies task → dispatches to available worker → agent in worktree
sonnet-planner.js (every 10 min) → polls requirements/ → writes pending/
pr-reviewer.js (every 10 min) → reviews open PRs
merge-watcher.js (every 5 min) → merges approved PRs
worktree-checker.js (every 5 min) → monitors worktrees, opens PRs

Known Issues

• dispatch.sh fails when dispatching farm-1 TO farm-1 (self-dispatch) — Canopy insurance task failed
• QC3 still has old dispatched/failed tasks from pre-cleanup agent commits (11 dispatched, 8 failed)
• Sonnet planner pulled from backlog and burned through BL-001 to BL-007 — all went to failed/