# Example checkpoint policy (sanitized demo)
#
# A checkpoint is the unit of memory. It is a recorded state change in what a
# lane knows, can do, is blocked by, or is allowed to write. A session reads
# recent checkpoints and resumes from there.

policy:
  interval: checkpoint
  not_every_turn: true
  rationale: >-
    Every chat turn is too granular and task completion is too coarse. The
    stable interval is a checkpoint: a real transition in lane state.

checkpoint_types:
  hydration:        first read after a session starts or rehydrates
  pre_write:        before any file-changing pass
  worker_report:    a worker reports a result, blocker, or evidence package
  gate_transition:  a gate opens, closes, fails, or is waived
  blocker:          progress is blocked by missing evidence, access, or conflict
  handoff:          ownership or next-action authority transfers
  commit_push:      a file-changing pass lands and is published
  authority_change: a schema, decision, or source-of-truth authority changes

minimum_standard: >-
  Log at every checkpoint. Do not log every turn. Do not wait for task
  completion if a gate, worker report, blocker, or authority state changes.