Research — Coord-Read Residuals (#2185 + #2186)

Phase 0 — Code-state research (3 parallel agents, verified against main)

  • Kind corrections (debugger-verified against is_primary_artifact_kind): 6 of 10 Lane-A issue labels are wrong — merge/resolve.py:98, cli/commands/merge.py:269, lanes/worktree_allocator.py:360 read meta.json (PRIMARY_METADATA) not LANE_STATE; lanes/recovery.py:611 reads lanes.json (LANE_STATE) not tasks/. merge/executor.py/done_bookkeeping.py:237/recovery.py:356 are mixed PRIMARY+STATUS. Route by real kind.
  • Husk failure mode real: meta.json/lanes.json/tasks/ are PRIMARY-only post-#2106; coord-aware resolve lands on the empty -coord husk. next_cmd.py:187/253 swallow FileNotFoundError; :631 get_mission_type returns default software-dev (wrong-routing, not just telemetry).
  • Gate blindness: the dir-read scanner matches only resolver / "tasks"|"lanes.json"|"*.md" joins; identity reads (function-call shape) escape both arms → a net-new arm is required.
  • #2115 sequencing: implement.py:1389 is correct only via the :1018 fallback; guards must precede fallback removal.

Phase 0.5 — Post-plan brownfield check (2 lenses, 2026-06-26)

Foldable-issue search (planner-priti) — CLEAN CAMPSITE

No new issues to fold; #2185/#2186 are the mission. Adjacent open issues are correctly partitioned and must stay separate:

#RelationVerdict
#2115implement-loop twin (owned by sibling, C-009)NO-FOLD (hard boundary)
#2167repo-root scripts/tasks/ legacy readerNO-FOLD / reference
#2140, #2139, #2138strangler-residual cluster (parent #1878)NO-FOLD / reference
#2100route ~62 inline meta.json reads through load_meta (authority, not surface)NO-FOLD (orthogonal)
#2123, #2091worktree teardown / coord-branch construction (write-side)NO-FOLD
#2160epic parentreference

Watch-items (don't fold, don't regress):

  • #2139 (dual target_branch reader with silent main fallback in core/paths.py/core/git_ops.py) sits next to Lane A merge-reads — do not reintroduce or entrench the silent fallback while editing that neighborhood.
  • #2115 boundary — cli/commands/merge.py appears in BOTH missions at different lines (this mission :269; sibling _mark_wp_merged_done). The arch-gate pin set shrinks by the right issue's lines on each side.

Split-brain / LOC / deprecation (paula-patterns)

Split-brain pair — MUST converge (in scope, refines FR-002): merge/executor.py:887 already computes target_feature_dir PRIMARY, but _run_lane_based_merge recomputes feature_dir coord-aware at :976 for the same mission, then feeds identity (:981) and the review-artifact gate. Thread the existing :887 PRIMARY dir through to :976 rather than swap-and-recompute.

Missing shared seam — DEFERRED (NOT this mission): the two-call PRIMARY idiom primary_feature_dir_for_mission(x, _canonicalize_primary_read_handle(x, slug)) is hand-inlined at ~12 sites, with 3 duplicate _planning_read_dir wrappers (orchestrator_api/commands.py:324, agent/mission_feature_resolution.py:69, acceptance/__init__.py:787). A canonical primary_read_dir(repo, handle) would dedup them — but (a) it tension-conflicts with C-002 (consume-not-author the resolver), (b) it straddles the implement-loop sibling's owned workflow.py/implement.py legs (cross-mission collision), and (c) it is broader than #2185/#2186. Record as a separate follow-on dedup ticket (cousin of #2100); this mission consumes the existing resolve_planning_read_dir seam only.

Sizing (affects IC-03): lanes/recovery.py::scan_recovery_state already carries # noqa: C901 (already over the complexity ceiling). A per-leg PRIMARY/STATUS split inside it would worsen it — extract the PRIMARY-planning read and the status-events read into named helpers, drop the # noqa, add focused tests (do not add another branch). Other mixed functions (executor._run_lane_based_merge, done_bookkeeping._mark_wp_merged_done) have headroom — safe in place.

Deprecation / guardrail:

  • candidate_feature_dir_for_mission is NOT deprecated — it is the C-005 STATUS-partition primitive. The mission re-points PRIMARY reads off it; it must never remove or "converge away" the coord-aware primitive (a "one resolver" framing would be a regression and break C-001 status reads).
  • Distinguish in merge/resolve.py: :98 is the meta.json PRIMARY read (route) vs :63 is handle→dir-name canonicalization at the no-silent-fallback boundary (.name, exists-gated — leave on candidate_). Same caution for next_cmd.py:360 (canonicalization, leave) vs :187/:253/:631 (route).
  • merge.py:269 is coordination-teardown (--abort) — confirm PRIMARY_METADATA semantics but it is the coord-side; verify before routing.

Owned-file LOC (sizing reference): executor 1066, recovery 784, next_cmd 785, merge.py 582, done_bookkeeping 569, lanes/merge 508, worktree_allocator 470, worktree_topology 309, resolve 268, forecast 210. (workflow.py 2799 + implement.py 1459 are sibling-shared — touch only the owned identity legs.)