Agile process

A single turn of the loop:

1. Refine the product backlog — keep the next bucket honest as candidates change shape, and let new ideas land in deferred.
2. Plan a new sprint — see Sprint planning: define the sprint mission first, choose mission-fitting captures from next, promote each into a story, and decompose into one-PR tasks.
3. Execute — see Sprint execution: agents run BACKLOG tasks through the lifecycle one PR at a time; the heartbeat rule keeps Status fresh; discovered work is triaged as it appears.
4. Close the sprint — see Sprint closure: resolve every story, write the achievements, generate release notes and cut a release if applicable, run the retrospective, close the sprint.
5. Next sprint opens. New captures discovered during execution have already landed in the backlog and surface in the next refinement.

A version is the outer loop: a coherent group of sprints with a shared mission. It opens when its first sprint opens, and closes when the version's Definition of done is met.

Owned by System 3 (sprint coherence drives whether next is honest). Captures are filed by System 2 when discovered during execution. Identity-fit calls are referred to System 5; next↔deferred moves for the next version are driven by System 4.

Refinement is the cheap, ongoing work that keeps the next bucket usable. It is not a meeting — it is small touches whenever a capture is touched:

• New ideas land via compass add capture. Pick the bucket using the rules in product backlog.
• A capture's body may be tightened (description, tags) as a sprint approaches and it becomes a serious candidate.
• A capture moves git mv between buckets when its horizon changes (e.g. a "deferred" idea becomes "next" because its blocking dependency landed; a "next" idea drops to "deferred" because the next version's focus shifted).
• Captures that are no longer interesting are deleted, not archived. The product backlog is not an artefact museum.

A capture file carries no Status table — there is nothing to update during refinement except its body and bucket.

A sprint moves through three phases; each phase has its own page and the boundaries are sharp:

• Sprint planning — mission first, then stories, then tasks. Ends when the sprint is ready: mission filled, at least one story BACKLOG-ready.
• Sprint execution — tasks through the lifecycle, one PR each; heartbeat rule; discovered-work triage.
• Sprint closure — resolve stories, achievements, release notes, retrospective; ends when the record is complete.

A version closes when its Definition of done is met. The version file's Status moves STARTED → RELEASED, :CLOSED: is stamped, and the Sprints table on the version page is the historical record.

• Lifecycle — the generic TODO state machine on each document.
• Document types — per-type frontmatter and section contract.
• Product backlog — what the backlog is and how next/deferred buckets work.
• Sprint themes — the six themes and epic convention for sprint story tables.
• Sprint health charts — the four charts on every sprint page: what they measure and how to read them.
• Glossary — definitions of every term used here.
• How do I create a new doc? — the recipe used at every promotion step above.