⚡ Do This First: Load Agent Profile

Before reading anything else, load your assigned agent profile via /ad-hoc-profile-load (profile: python-pedro, role: implementer). Adopt its identity, boundaries, and discipline for the whole work package.

Objective

Fix #4897: spec-kitty doctor mission-state --fix (and spec-kitty upgrade, which routes into the same repair) quarantines the authoritative DecisionPoint* rows out of status.events.jsonl on a healthy mission — exit 0, errors=0 — and the advertised doctor decisions --repair then rebuilds index.json from the emptied log to zero entries. The decision record becomes unreadable by every CLI surface.

Structural root cause (parallel authority / inverted ownership)

Two subsystems disagree about who owns DecisionPoint* rows in the shared status.events.jsonl append log:

rebuilds index.json from "the authoritative event log"; status/store.py:597 is_non_lane_event returns True for ANY row with event_type (explicitly naming DecisionPoint*).

_is_preserved_non_lane_row (:1972) preserves only retrospective + LIFECYCLE_EVENT_TYPES`, and its docstring (:1940-1944) asserts DecisionPoint's "canonical store is elsewhere" — the inverse of the truth. It KNOWINGLY diverges from the reader (:1955-1956).

Decisions + durable reader treat the JSONL rows as authoritative. decisions/index_fold.py
The mission-state repair treats them as a prunable mirror. `migration/mission_state.py:1919

Operator-chosen fix (decision 01M3560PHPQM617TH5MZHWRF8J): one shared registry, not another hand-maintained preserved-set entry — closing the recurring whack-a-field class (#2376 retrospective, #3066 WPStatusChanged, #3541 review_result, #4897 DecisionPoint).

Guidance per subtask

T006 — Red-first regression test

Add tests/status/test_authoritative_non_lane_registry_4897.py with a @pytest.mark.regression test pinned to #4897: build a healthy mission with one resolved Decision Moment, run doctor mission-state --fix, assert the DecisionPoint* row count in status.events.jsonl is unchanged and doctor decisions stays clean: true and agent decision list count unchanged. RED before the fix. (Follow the QA repro script in the issue.)

T007 — Create the authoritative non-lane event-type registry

Introduce ONE registry — a frozenset[str] of authoritative non-lane event_types (at minimum DecisionPointOpened/Resolved/Deferred/Canceled/Widened) plus a predicate — placed with the existing status event-type authority (status/lifecycle_events.py, alongside LIFECYCLE_EVENT_TYPES, or a clearly-named sibling). It must express the same authority the durable reader already applies (any row bearing an authoritative event_type). Document that this is the single source both the reader and every repair/prune consult.

T008 — `is_non_lane_event` consults the registry

Refactor status/store.py:597 is_non_lane_event to derive its event_type decision from the new registry (behavior-preserving for currently-preserved rows).

T009 — `_is_preserved_non_lane_row` consults the registry

Make migration/mission_state.py:_is_preserved_non_lane_row DELEGATE to the shared registry/predicate rather than maintaining a narrower parallel list. DecisionPoint* rows must now be preserved; the already-preserved classes (annotation, retrospective, WPStatusChanged, review_result) must remain preserved (no regression). Update the now-inverted docstring.

T010 — Honest repair status

Ensure _repair_mission (:1671-1691) does not report errors=0/status="repaired" success when it would quarantine a canonical row. With T009 no canonical rows are quarantined; add a guard/assertion so a future divergence surfaces as an error rather than a silent success.

T011 — Single-authority test + no-regression

Add a focused test asserting BOTH consumers (is_non_lane_event and _is_preserved_non_lane_row) derive from the one registry (e.g., parametrize over the registry and assert agreement) — a non-vacuous guard against the whack-a-field class reopening. Extend/adjust tests/migration/test_mission_state_repair.py to cover DecisionPoint preservation and confirm the other preserved classes are unaffected.

Definition of Done

agent decision list count stable.

tests/integration/migration/test_lifecycle_events_preserved.py, decisions tests. ruff, ruff format --check, mypy clean. Touched functions ≤15 complexity.

T006 regression RED before, GREEN after.
doctor mission-state --fix preserves DecisionPoint* rows; doctor decisions stays clean;
One registry; both consumers consult it; single-authority test present.
Already-preserved classes unchanged (no regression).
Blast radius green: tests/status/, tests/migration/, tests/unit/migration/,

Reviewer guidance

Confirm the registry is a single authority (not a copied list); confirm is_non_lane_event and the repair both consult it; confirm DecisionPoint rows survive --fix; confirm no regression to retrospective/annotation/WPStatusChanged/review_result; confirm the regression test fails on merge-base.

Spec Kitty

Kanban Board