3.2.1 scoping — Debbie severity scoring (NEUTRAL panel)

Author: Debugger Debbie (NEUTRAL severity-scoring pass — no predetermined answer). Branch: design/naming-identity-ssot-alignment @ 3.2.0 (read-only; no commit/switch). Date: 2026-06-16. Directives applied (debugger-debbie.agent.yaml): D-001 (find the owning boundary / structural fork, not the symptom), D-003 (persist falsified/surviving hypotheses), D-030 (does a producer- conformance gate actually prevent recurrence?), D-032 (divergence-matrix lens for two-surface bugs).

Lens: bug-reality + severity. For each bug candidate I reproduced the logic path against the code and cite file:line. For epics/slices/naming I scored the severity of the problem addressed. Scores are 1–5 (5 = most severe / most-verified / highest impact). This is a fresh verification — I did not inherit the red-team's priority verdict; I re-derived each from code.

Verification notes (the evidence behind each score)

#1716 — coordination topology coherent from create→planning (P0, write-side) — VERIFIED REAL, OPEN

  • mission create writes the topology-activation signal immediately: src/specify_cli/core/mission_creation.py:411 (meta["coordination_branch"] = …).
  • The coord worktree is NOT materialised at create — the read resolver itself documents the gap: src/specify_cli/missions/_read_path_resolver.py:241-242 ("the transitional window between mission create and the first coord-worktree materialisation").
  • The recent topology-stabilization mission (01KTZVQ2, squash 991162c0a) closed #1164/#1883–#1888 + read-surface/ff-merge slices of #1878 — but its issue list does not include #1716 (kitty-specs/coordination-topology-stabilization-01KTZVQ2/spec.md:7). It added a graceful read-side fallback to primary during the window (mitigation) and FR-003/FR-010 fix two symptoms (setup-plan accepting a coord-committed spec; zero manual ff-merges). The architectural root cause #1716 names — write/topology split where the signal is set before authority materialises — is still present. GitHub state: OPEN (confirmed via API).
  • D-001 verdict: this is the owning boundary defect — the write side authors identity/topology while the authoritative surface does not yet exist. Highest blast radius of the candidate set.

#1832 — implement claim succeeds but "no workspace could be resolved" (P1, read-path) — VERIFIED REAL, OPEN

  • Logic path: src/specify_cli/cli/commands/agent/workflow.py:1336 (first resolve) → on not workspace.exists, top_level_implement(...) creates the worktree (:1357) → re-resolve :1372 → if still not workspace.exists, the exact reported error fires :1375.
  • The re-resolution depends on find_context_for_wp → build_feature_context_index (src/specify_cli/workspace/context.py:434), and only falls through to a derived path with mission_id=None (legacy grammar {slug}-{lane}, no mid8) at context.py:819-821. A read-after-write where the just-created modern-grammar ({slug}-{mid8}-lane-{id}) worktree is not re-indexed in-process resolves to a non-existent legacy path → the error.
  • D-032 verdict: classic two-surface divergence — the CREATE path and the RESOLVE path disagree on the workspace identity. The issue's own "single resolution path" suggested-direction is correct.
  • No fix commit references #1832; the brittle re-resolve at :1372 is unchanged. Still OPEN.
  • Severity tempering: worktree/branch/status are all actually correct; the failure is confined to the final report + prompt-regen skip. Workaround exists (use canonical kitty-specs/.../WPxx.md). The dangerous part is the skipped /tmp/spec-kitty-implement-WPxx.md regen colliding across missions — that elevates it above cosmetic.

#1827 — merge baseline validation before baseline_merge_commit written (P1) — OVERSTATED / LIKELY ALREADY FIXED

  • The issue claims a circular ordering: assert-before-write. Current code orders it correctly:
    1. _record_baseline_merge_commit writes the field into the target-checkout meta.json (src/specify_cli/cli/commands/merge.py:2574, writing to target_feature_dir).
    2. baseline_meta_path is appended to files_to_commit (:2694-2695).
    3. safe_commit lands it on the target branch (:2708, capability MERGE_BOOKKEEPING).
    4. Only then _assert_baseline_merge_commit_on_target validates the committed value (:2741).
  • The assert reads the recorded working-meta baseline, not a re-derived HEAD, expressly to make --resume convergent (:1727-1734). The "re-run re-merges and fails identically" symptom is guarded by resume reconciliation. Reported on rc40/rc41; the ordering + resume fix landed via the coordination/merge-stabilization line (78024bd48 "persist target bookkeeping", aa9ec6392 "stage merge bookkeeping from target checkout", squash 991162c0a / 3f2af08f0 #1879). GitHub: still OPEN, but the code no longer exhibits the described circular failure.
  • D-003 (persist): H-VERIFY — the exact failure in #1827's body is not reproducible against HEAD. Recommend: re-test on current build and close as fixed unless a residual resume edge is shown. Do NOT lead 3.2.1 with this.

#1891 — agent --json broken — PARTIALLY FIXED; one verified residual

  • Part 1 (map-requirements --json "CommitResult is not JSON serializable"): FIXED — 4c492aa85/bc927ef80 (present on this branch; committed is now a bool + commit_sha).
  • Part 2 (agent action implement --json rejected): STILL REAL — the command signature implement(...) at src/specify_cli/cli/commands/agent/workflow.py:1140-1162 has no --json option; an orchestrator cannot get workspace_path/prompt_file/lane as structured data. This is the residual, and it directly degrades the implement-review loop's machine surface.
  • Part 3 (preamble before JSON on finalize-tasks/setup-plan): not separately re-verified; bounded.
  • Severity tempering: affects orchestration/automation, not data integrity; human output works.

#1619 / #1666 — execution-context / domain-boundary epics (P0 epics)

  • The problem (coord/main/lane split-brain authoring identity in ≥2 surfaces) is exactly what #1716/#1832 are concrete instances of — high severity at the class level. But the synthesis + red-team (corroborated here) refute "thread the ExecutionContext everywhere" as the fix: the composite is a mutable @dataclass with an internal branch_name≠branch_ref.target_branch inconsistency, and threading converges on ~2 sites. A shippable 3.2.1 slice is small — fix the internal inconsistency at the 2 context-holding sites after the write-side authority is consistent. Score reflects the problem severity, not the (refuted) full-epic framing.

#1878 write-side slice (P2 umbrella)

  • The coordination-stabilization mission already consumed the read-surface/ff-merge portions of #1878. The remaining bounded write/entry slice is essentially the #1716 root cause re-framed. Severity inherits from #1716; as a standalone P2 umbrella its incremental value is the structural guard, not a distinct new bug.

Naming routing rider (#2000/#1971/#1993/#1888/#1900)

  • The authority exists (mid8/resolve_mid8 at src/specify_cli/lanes/branch_naming.py:122,169); 20 bare mission_id[:8]/[0:8] sites remain in src/specify_cli/ (grep-verified). Routing them is cheap and real (I conceded C1 in the red-team pass). But: the ratchet (tests/architectural/test_no_worktree_name_guess.py, \bmid8\b) passes today with those 20 sites live → it is a syntax tripwire, not a completeness oracle (D-030). None of the recurring high-severity bugs (#1716/#1832/#1827) is a bare-[:8] slice — they are surface-selection / read-after-write / ordering bugs. Lowest problem-severity of the set.

Scored table

Candidate Severity Evidence (verified-real?) User/launch-impact One-line justification + file:line
#1716 topology create→planning (P0) 5 5 5 Launch-blocker root cause still live; signal set before authority materialises — mission_creation.py:411 vs _read_path_resolver.py:241-242; OPEN, not in 01KTZVQ2's issue list (coordination-topology-stabilization-01KTZVQ2/spec.md:7).
#1832 no-workspace-resolved (P1) 4 5 4 Reproduced create→re-resolve divergence workflow.py:1336→1357→1372→1375; legacy-grammar fallback context.py:819-821; dangerous prompt-regen skip; workaround exists. OPEN.
#1827 merge baseline ordering (P1) 2 2 2 Overstated/likely-fixed: record→commit→assert is correctly ordered merge.py:2574→2695→2708→2741; resume-convergent by design :1727-1734. Body not reproducible on HEAD.
#1891 agent --json (P1) 3 4 3 Part 1 FIXED (4c492aa85); residual real: action implement has no --json workflow.py:1140-1162. Automation-surface, not data-integrity. OPEN.
#1619/#1666 execution-context epics (P0) 4 4 3 Problem (split-brain authoring) is real & high-class severity (#1716/#1832 are instances); the fix framing (thread-everywhere) refuted; shippable slice is small (~2 sites).
#1878 write-side slice (P2) 4 3 3 Read-surface/ff-merge parts already shipped (01KTZVQ2); remaining slice ≈ #1716 root cause; value is the structural guard.
Naming routing rider 2 4 2 Cheap, real consolidation (20 [:8] sites, branch_naming.py:122,169); ratchet is a syntax tripwire passing today with 20 live sites (test_no_worktree_name_guess.py); not the recurring high-severity class.

Top-3 by severity (Debbie / verified)

  1. #1716 (P0, write-side topology). The highest-blast-radius, verified-still-open defect: the topology signal is authored at mission create before the authoritative coord worktree exists, and the recent stabilization mission closed adjacent issues but not this root cause. This is the owning-boundary fix (D-001). Lead 3.2.1 here.
  2. #1832 (P1, read-path). Verified create-vs-resolve divergence (D-032) hitting every claim in at least one dogfood mission; the prompt-regen skip makes it more than cosmetic. Cheap, contained, high-frequency. Strong second.
  3. #1619/#1666 — a bounded execution-context slice. Score the problem class, not the refuted "thread everywhere." The smallest defensible slice: make the composite's internal branch_name/branch_ref consistent + immutable at the ~2 context-holding sites, sequenced after the write side is fixed. (#1891-residual is a worthy cheap rider but lower-severity.)

Demote: #1827 (overstated — re-test then close), and the naming rider (lowest problem- severity; keep as a cheap opportunistic rider + honest tripwire, never the lead).

Newly-surfaced item (not in the candidate list)

  • No new HIGH-severity bug outside the set. The open-P0 list is #1716/#1666/#1619 plus meta-trackers (#1599/#1601) and known-debt (#1766 ownership leeway). The synthesis-cited god-modules (merge.py / agent/mission.py high CC) are a maintainability risk that amplifies recurrence cost (#1716/#1827 both live in merge.py's blast zone) but are not a standalone launch bug — fold extraction into whichever of #1716/#1827 touches those files.

Persisted hypotheses (D-003 — so they are not re-litigated)

  • H-VERIFY-1: #1827's circular-ordering body does not reproduce on HEAD (record→commit→assert is ordered correctly, resume-convergent). Re-test on current build; close if green.
  • H-SURVIVES-1: #1716's write-side root cause survives the 01KTZVQ2 mission (which fixed neighbours + a read-side fallback, not the materialisation-timing split).
  • H-SURVIVES-2: #1832 is a read-after-write/two-surface divergence, not a naming-slice bug; the naming rider would not fix it.
  • H-FALSIFIED-1: "All four P1 bugs are live" — falsified; #1891 part 1 is fixed and #1827 is (very likely) fixed.
Edit this page