#2007 training-bug ticket mapping & sequencing impact — planner-priti
Date: 2026-06-16 · Author: planner-priti (read-only research op; no tracker mutation) ·
Source: gh issue view 2007 + per-candidate gh issue view (states/milestones verified live).
Decision context: docs/release-goals/3.2.x.md, the neutral panel
docs/plans/engineering-notes/3-2-x-goal-corroboration/SCORING-SYNTHESIS.md, and the in-flight naming rider
kitty-specs/naming-identity-routing-rider-01KV7SFD/ (branch feat/naming-rider-3-2-1).
#2007 itself: OPEN, labels bug + epic + 3.2.0, milestone = none (the 3.2.0 label is
a category tag, not the milestone — the epic is currently unmilestoned). Evidence = Robert's screenshots
2026-06-16 09:47–12:05 against a fresh spec-kitty-monorepo-prep workspace + Debugger Debbie read-only
investigations. No screenshots uploaded; the failure inventory in the body is the artifact.
1. The 16 bugs → existing-ticket-or-NEW map
Verified states/milestones (live gh, 2026-06-16). "Milestone" = GitHub milestone field, not the label.
| # | Bug (short) | Existing ticket | Ticket state | In 3.2.x milestone? | Disposition |
|---|---|---|---|---|---|
| 1 | doctrine list command drift; skills/prompts imply nonexistent surface |
NEW | — | — | net-new (command-surface class) |
| 2 | Charter stale/status loop; status --json traceback; sync noop-while-stale; entity_pages DRG warning |
NEW (related: #1336 charter --json on error, CLOSED) | #1336 CLOSED | no | net-new; #1336 is a prior partial in the same family, not a dup |
| 3 | Specify NO_BRIEF/NO_TICKET bootstrap ambiguity |
NEW | — | — | net-new (UX/typed-state) |
| 4 | setup-plan requires --mission while prompt/docs say no-flag |
NEW | — | — | net-new (resolver/contract) |
| 5 | agent context resolve requires --action (undocumented) |
NEW | — | — | net-new (resolver/contract + doc drift) |
| 6 | Submodule/.git-file root misresolution → SPEC_KITTY_REPO_NOT_INITIALIZED |
NEW | — | — | net-new (root detection) |
| 7 | spec_committed:false while committed — checks wrong branch authority; SpecifyCompleted gated wrong |
NEW | — | — | net-new (read-path authority) |
| 8 | decision open rejects valid mission handle (path would escape kitty-specs/) |
NEW | — | — | net-new (read-path authority) |
| 9 | Raw python -c "from specify_cli.core.templates" fails (out-of-venv + wrong import path) |
NEW | — | — | net-new (prompt-hygiene/guard) |
| 10 | finalize-tasks --validate-only exit 1 on zero-match globs; future-file vs glob semantics unclear |
NEW (adjacent: #1888 ownership existence-check, in rider) | #1888 OPEN | yes | net-new; distinct from #1888 (this is glob/warn-vs-error semantics, not phantom-path existence) |
| 11 | finalize-tasks from coord worktree reads planning inputs from wrong surface (meta.json/spec.md not found) |
NEW | — | — | net-new (read-path authority) |
| 12 | Explicit --mission error flattened into generic "pass --mission" remediation |
NEW | — | — | net-new (error-fidelity) |
| 13 | Broken coord worktree recovery; docs reference nonexistent agent worktree repair |
#1890 | OPEN | no | existing — exact match |
| 14 | STATUS_READ_PATH_NOT_FOUND for valid mission; coord topology exists, no primary fallback |
NEW (zone of #1832/#1716/#1619) | #1832 OPEN / #1716 OPEN | #1832 no / #1716 no | net-new symptom of the read-path/coord-topology class |
| 15 | next hides real failure — query mode discards preflight, reclassifies read-path miss as MISSION_NOT_FOUND |
NEW | — | — | net-new (read-path/error-fidelity) |
| 16 | implement command surface mismatch — --json on internal allocator, rejected by agent action implement |
#1891 (residual: "action implement rejects --json") | OPEN | no | existing — #1891 already names this exact residual |
Net-new vs existing count
- Existing tickets: 2 of 16 — #1890 (bug 13) and #1891 (bug 16, the documented residual).
- Adjacent-but-distinct existing tickets (same family, do NOT collapse): #1888 (bug 10 is a
different finalize defect than #1888's phantom-path existence check); #1832/#1716/#1619
(bug 14 is a symptom of that read-path/coord class, not a 1:1 dup); #1336 (CLOSED — prior
charter-
--json-on-error partial, related to bug 2). - Net-new (no ticket exists): 14 of 16 — bugs 1–12 (minus the #1890/#1891/#1888/#1832 overlaps noted), 14, 15. Concretely the genuinely-uncovered ones needing new sub-issues: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 15.
2. The 16 → 6 suggested sub-issue clusters
The epic body proposes 6 sub-issues. Mapping each bug to its cluster:
| Cluster | Bugs | Net-new? | Existing-ticket touchpoints |
|---|---|---|---|
| C1 — Command-surface validation & docs/prompt drift | 1, 5, 9, 16 | mostly new | bug 16 = #1891 residual |
| C2 — Charter status/sync/preflight consistency | 2 | new | related CLOSED #1336 |
| C3 — Mission context / read-path resolver unification | 4, 7, 8, 11, 12, 14, 15 (+ bug 5 has a resolver tail) | new | zone of #1832 / #1716 / #1619 (unmilestoned) |
| C4 — Coordination worktree repair surface | 13 | existing | #1890 (exact) |
| C5 — Implement/review action JSON contract | 16 | existing | #1891 (exact residual) |
| C6 — Submodule / root detection hardening | 6 | new | — |
| (loose) Specify bootstrap UX | 3 | new | — fits C1 (prompt/typed-state) or stands alone |
| (loose) Ownership/finalize glob semantics | 10 | new | distinct from #1888; fits C3 or stands alone |
C3 is the centre of mass — 7 of 16 bugs (4, 7, 8, 11, 12, 14, 15) are the same "wrong mission-state
authority resolved, then typed failure flattened" defect the epic's Architectural Diagnosis #2 names. This
is precisely the read-path/single-resolution surface the neutral panel ranked first for 3.2.1
(#1832 → #1716), now corroborated by live training evidence.
Overlaps with the naming rider (feat/naming-rider-3-2-1)
The rider's spec (read directly) addresses #2000, #1971-tail, #1888 and explicitly defers the read-path/coord class to 3.2.2. Cross-referencing #2007:
- #1888 — bug 10 is adjacent to the rider's #1888 work but is a different finalize defect (zero-match-glob warn-vs-error semantics vs the rider's phantom-path existence check, FR-007). The rider's #1888 fix does not close bug 10. One overlap, partial.
- #1832 / #1716 / #1619 — the rider lists these out-of-scope (C-005), deferred to 3.2.2. Bug 14 (and the C3 cluster) lands squarely in that deferred zone. Overlap = by-deferral, not by-coverage.
- #1891 — the rider lists #1891 residual as out-of-scope (C-005). Bug 16 = that residual. Overlap by-deferral.
- #1890 — not in the rider at all. Net-new to the rider's scope.
Net: the naming rider touches #2007 only at the #1888 surface (and that only partially). Everything else in #2007 is either net-new or sits in the zone the rider deliberately pushed to 3.2.2. The rider does not close #2007; #2007 does not block the rider.
3. The decision question — does #2007 change the 3.2.1 lead?
What the prior decision was
The neutral panel (all three scorers) leaned impact-first: open the write-side/single-resolution surface via #1832 → #1716, demote naming to a parallel rider. The operator overrode to safety-first — naming rider as the 3.2.1 opener — recorded explicitly as a values choice, not a data verdict, trading "lowest-risk momentum + establish the ratchet/routing pattern" against "frees no P0/P1, parks the highest-blast-radius surface."
What #2007 adds
#2007 is live training evidence (a real operator workflow run) that the command-drift + read-path authority class — the exact surface the panel wanted to lead with — is causing observable workflow failures right now, including at least one launch-blocker-class symptom (see §4). 7 of 16 bugs are the C3 read-path/resolver-flattening defect; 4 more are command-surface drift. This is no longer a predicted ROI; it is reproduced field failure on a fresh workspace.
Critically, #2007 evidence arrived after the safety-first call. The panel's de-bias note already conceded "leading with naming forfeits impact, the safety is available inside the impact plan (because #1832 is the safest first WP)." #2007 strengthens the impact side of that exact trade-off with empirical weight the panel did not have.
Options assessed
| Option | What it means | Trade-off |
|---|---|---|
| (i) Rider as-is, #2007 → 3.2.2 | No change; #2007 becomes the 3.2.2 headline | Cleanest, preserves the in-flight rider's momentum. Cost: ships a stabilization patch (3.2.1) that does not touch the class actively breaking the operator's own training run; the launch-blocker-class bug 14/13 waits a full patch. |
| (ii) Rider stays, bounded #2007 slice joins 3.2.1 | Rider proceeds; carve a small, safe #2007 sub-slice (e.g. C4 #1890 repair-command + C1 command-surface CI guard + bug-12 error-fidelity) into the same patch | Keeps the safe opener AND lands the cheapest, highest-confidence #2007 wins. Cost: modest scope-creep on 3.2.1; needs a clean WP boundary so it doesn't entangle with the rider's byte-parity work. C4/#1890 and the command-snippet CI guard are genuinely independent of the rider's seam routing. |
| (iii) #2007 becomes the 3.2.1 lead; rider yields/parallels | Promote the C3 read-path resolver unification (= the panel's #1832→#1716 lead) to the headline; rider runs parallel or slips | Aligns 3.2.1 with both the neutral panel's original lean AND the new field evidence. Cost: higher characterization-test cost + semantics risk (exactly what safety-first was avoiding); discards in-flight rider momentum; re-opens a decision the operator already made deliberately. |
Recommendation — (ii), trending toward (iii) for 3.2.2
Keep the naming rider as the 3.2.1 opener (honor the operator's deliberate safety-first call — it is already in flight, severable, and the panel confirmed its safety is real), BUT fold a bounded, low-risk #2007 slice into 3.2.1: #1890 (C4 worktree-repair command — net-new, independent, fixes a recovery dead end Robert hit) and the C1 command-surface CI guard (bug 1/5/9 — a regression tripwire, the same pattern-establishing move the rider's ratchet already embodies, so it fits the safety-first thesis perfectly). Then make #2007's C3 read-path resolver unification the explicit 3.2.2 headline — which is also the panel's original #1832 → #1716 lead, now field-proven.
Rationale, tracker-grounded:
- The C3 resolver class (7/16 bugs) is the deferred 3.2.2 zone the rider already names (#1832/#1716/ #1619, C-005). #2007 does not create new sequencing — it raises the priority and evidence weight of the work already slated for 3.2.2, and supplies the characterization corpus (Robert's reproduced paths) that 3.2.2 needed anyway.
- #2007 does not change the 3.2.1 lead (naming rider stays) because reversing an in-flight, deliberate, severable values-call on fresh-but-not-emergency evidence costs more momentum than it buys — and #2007's safe subset (C4 #1890 + the C1 guard) can ride along without reversing anything.
- It does change 3.2.2: #2007 should become the named headline of 3.2.2, absorbing the panel's #1832→#1716 lead, with the 6 sub-issues opened as its children and milestoned onto 3.2.x.
Tracker hygiene this surfaces (do regardless)
Both the panel and this analysis confirm the milestone inversion persists: #1716, #1832, #1827, #1891, #1619/#1666, and #1890 are all unmilestoned, while only #2000/#1971/#1993/#1900/#1888 sit on 3.2.x. #2007 itself is unmilestoned. Milestone #2007 + its read-path children onto 3.2.x so the burndown reflects the real cycle. (Recommendation only — this op is read-only.)
4. Launch-blocker flags
The epic body does not carry the launch-blocker label itself, but the framing ("3.2.0 training
bugs that Robert witnessed", fresh-workspace evidence, "Proposed Epic Acceptance Criteria" gating a
training run) is launch-readiness evidence. Bug-level blocker assessment:
- Bug 6 (submodule/root misresolution →
SPEC_KITTY_REPO_NOT_INITIALIZED) — launch-blocker-class. In a submodule-style checkout (a real consumer topology, e.g.econcept-nextmonorepo prep) Spec Kitty is unusable — it resolves to the parent repo and refuses to initialize. Hard stop, no workaround short of restructuring the consumer repo. - Bug 14 (
STATUS_READ_PATH_NOT_FOUND+agent action implementsame miss) — launch-blocker-class by association. It is the read-path symptom of #1716, which carries the explicitlaunch-blockerlabel (verified). The implement path failing closed with no fallback blocks the core implement loop on coord topologies. - Bug 13 (#1890 broken coord-worktree recovery, points at nonexistent command) — high-severity, not
strictly blocking: there is a real (if undocumented) recovery via
doctor workspaces --fix; the defect is that agents are steered into manual recursive deletion. Recovery-surface trap, not a hard stop. - #1716 itself carries
launch-blocker(verified label). The C3 cluster is its read-side manifestation, so the whole C3 read-path class inherits launch-blocker weight — reinforcing the §3 recommendation to headline it in 3.2.2 rather than let it drift.
Everything else (charter loops, doc drift, error-fidelity, JSON contract) is workflow-friction / confidence-eroding but has agent-survivable workarounds — important for a clean training run, not a hard launch gate.