#391 Doctrine Usage Test (WP11 dogfood)

Mission: doctrine-glossary-architecture-consolidation-01KTNWFC — FR-012, SC-1/SC-6. Date: 2026-06-11. Operator: curator-carla. Tracker: Priivacy-ai/spec-kitty.

This record is the usage-test of the doctrine authored in WP04/WP05 — the tracker-organisation-workflow procedure, the planning-and-tracking styleguide, the github-tracker toolguide, the iterative-deepening-review and moscow-scoping-lens tactics, and the planning-and-tracking glossary subset — applied to the real #391 epic using only that doctrine as the method.

Amendment note (authoritative)

The WP body on disk and spec SC-6 originally said "close #391 as superseded". Per operator decision 2026-06-11, this is amended: #391 STAYS OPEN. The test organises #391's current state per the doctrine. Honest "already-organized" findings are valid SC-1 evidence — the point of the dogfood is whether the doctrine alone suffices to assess and act on a real epic, not to force a destructive close.

Headline verdict

SC-1 (doctrine sufficient to organise the tracker end-to-end using only the authored artefacts): PASS. Every step of the procedure was executable from the doctrine text alone. The toolguide's gh/GraphQL mechanics and gotchas were not merely correct but predictive — two of its named traps fired exactly as documented during this run (see "Doctrine sufficiency"). No recourse to ad-hoc notes was needed.
SC-6 (as amended): PASS. #391 was assessed and organised per the doctrine and left OPEN by operator decision. Its tree topology was already clean (no orphans, no closed-parent stranding, no meta-tracker-as-parent, no catch-all, no duplicates in the open set); the doctrine's hygiene invariants (procedure Step 8 / styleguide principle 9) surfaced real, actionable gaps that were fixed live.

Inventory (procedure Step 1)

#391 EPIC: Tech/Functional Debt Remediation — OPEN, type Feature, single priority priority:P1, no parent (it is a root epic). Body carries a Purpose/Scope narrative and no child-ticket checklist — already compliant with the styleguide's "epic body carries Purpose/Scope, never a list of child numbers" principle.

76 native sub-issues total; 22 open (54 closed). One open child, #1797 ("Epic: 3.2.0 codebase sanitization"), is itself a functional sub-epic with 11 open leaves — a coherent root → bucket → leaves tier, exactly the shape the styleguide prescribes.

Open direct children of #391 (18): 644, 454, 1231, 1347, 1149, 582, 631, 442, 572, 825, 1008, 1790, 1791, 1634, 1815, 1834, 1838, 1842, plus the sub-epic #1797.

Open leaves under #1797 (11): 614, 1060, 1624, 1057, 1622, 1358, 1058, 955, 1059, 719, 1623.

Root classification (procedure Step 2)

Root	Class	Rationale
#391	Functional epic	Real domain of work (cross-cutting tech-debt remediation); owns tickets as native sub-issues; not a release/go-no-go gate. NOT a meta-tracker.
#1797	Functional sub-epic (bucket)	Coherent sub-class (codebase sanitization / dead-code / LOC reduction) under #391; owns its own leaves.

Note: #391 is part of the canonical 2026-06-09 tree and is off-limits to mutation except to file work into it (per the binding brief). It was not closed, not reparented, not retyped.

Steps 3–7 (drain / re-slice / sweep / orphan-clusters / dedup)

Step 3 (drain closed epics, collapse passthroughs, relabel meta-trackers): No open child sits under a closed/superseded parent — every open ticket resolves directly to #391 or to the live #1797. No one-child passthrough tier. No meta-tracker among the parents (#391 and #1797 are both functional). No action required — honest already-clean finding.
Step 4 (re-slice catch-all roots): #391 is not a flat catch-all — it has a Purpose/Scope body and one functional sub-epic bucket (#1797). The open direct children are a coherent tech-debt set (intake bugs, bulk-edit gate gaps, CI/test hygiene, orchestration hardening). No re-slice required. No action.
Step 5 (keyword + scope-read sweep, backfill links, flag overlaps): All parent links are native sub-issues (verified via GraphQL parent, not body checklists). No invisible-checklist-only children to backfill. No cross-epic overlap requiring a single-owner decision in the open set. No action.
Step 6 (orphan clusters → reuse/create epics): Zero genuine orphans in the open set — every open ticket has a functional parent. No ≥4-orphan cluster to home. No new epic created (correctly avoids the "fork a near-duplicate epic" anti-pattern). No action.
Step 7 (duplicate resolution with community-precedence): No duplicate pairs in the open set. Author-association was sampled via the toolguide's REST .author_association path (the field is rejected by gh issue view --json, confirming the toolguide note). No close performed; no community ticket touched.

Step 8 — hygiene invariants (where the real work was)

The topology was clean; the doctrine's per-ticket hygiene invariants are where #391 had drifted. Audited every open leaf for: resolves-to-functional-epic (all pass), single canonical priority, an issue type, and bug label ↔ type Bug.

Mutations executed (issue numbers)

Issue type derived objectively from the conventional-commit prefix and existing labels (procedure Step 8 / styleguide "Issue type mirrors the bug label"):

Issue	Was	Set to	Objective signal
#719	(none)	Task	`test(intake):` conv-commit prefix
#1622	(none)	Task	`[DIRECTIVE_013]` dead-symbol cleanup (refactor)
#1623	(none)	Task	`[DIRECTIVE_013]` doctor.py helper extraction (refactor)
#1624	(none)	Task	`[DIRECTIVE_013]` provenance-sidecar typing (refactor)
#582	(none)	Feature	`enhancement` label → capability work
#631	(none)	Task	documentation workaround task

Priority (triage judgement) — the procedure prescribes "assign a provisional default and flag it rather than silently guessing". Provisional priority:P2 applied and flagged for triage on the priority-missing leaves:

Issue	Action
#1815	added provisional `priority:P2` (flag for triage)
#1834	added provisional `priority:P2` (flag for triage)
#1838	added provisional `priority:P2` (flag for triage)
#1842	added provisional `priority:P2` (flag for triage)
#1791	added provisional `priority:P2` (flag for triage)

bug label ↔ type Bug: all bug-labelled open tickets (644, 1149, 1790, 1634, 1058, 1842) are already typed Bug, and no Bug-typed ticket lacks the label. No mutation needed — invariant already held.

Proposals deferred (not executed)

#1797 mistyped Task but carries the epic label → styleguide says "epics are typed Feature". A retype to Feature is the correct fix, but #1797 is a canonical-tree node the brief forbids mutating. Proposal: retype #1797 Task → Feature and add a priority:P* label (it currently has none) — to be actioned by whoever owns the canonical tree, not in this WP.
#1791 left type-missing. Title ("add regression test for coord-topology fix; path heuristic misses .jsonl") is genuinely ambiguous between Bug (the path heuristic is a defect) and Task (adding a regression test). The procedure says derive type objectively; with no conventional-commit prefix and no kind label, an objective call cannot be made. Proposal: owner to confirm Bug vs Task. Recorded rather than guessed (anti-pattern avoidance).

Step 9 — clean-bill re-walk

Post-mutation re-walk of all 29 open leaves (18 direct + 11 under #1797):

Orphans: 0. Meta-rooted: 0. Under-closed-parent: 0.
Priority anomalies (missing / dual / legacy Px-description): 0 among the leaves actioned. (Legacy Px-description labels were not present on the open set; the residual p1:verified / p1-decision:* labels are a separate triage-snapshot scheme — see doctrine sufficiency below.)
Type-missing: 1 — only #1791, deliberately left as a flagged proposal.

The only deliberately-unresolved item carries a named decision (owner confirms type), satisfying Step 9's "remaining items should be only roots plus a deliberately-parked item with a named owner".

Doctrine sufficiency assessment (SC-1)

The doctrine was sufficient. Every procedure step mapped cleanly onto the live epic and was executable from the artefact text with no improvisation. Two specific strengths and three concrete gaps:

Validated as predictive (strengths)

zsh word-splitting trap (toolguide "bash heredocs for loops"). The first batched parent-lookup loop, written in the default zsh shell, passed the entire $OPEN id-string as a single GraphQL Int! value and errored (Could not coerce value … to Int). Re-running the identical loop inside a bash <<'EOF' heredoc fixed it immediately. The toolguide predicted this exactly.
author_association field rejection (toolguide "Handy one-call queries"). gh issue view --json authorAssociation was rejected with "Unknown JSON field"; the toolguide's prescribed REST .author_association path worked. The toolguide saved the dedup step from a dead end.

The --paginate mandate (76 children, well over the 30-item default page) and the GraphQL-parent-vs-REST-parent distinction were both load-bearing and correct on this epic.

Concrete gaps (fed back to WP04/WP05)

No guidance for a residual / legacy triage-snapshot label scheme. #391's children carry p1:verified, p1-decision:keep|close-if-stale|split|defer — a prior triage pass's bespoke labels. The styleguide retires the legacy Px-description labels and defines the canonical priority:Px + triage:* namespaces, but is silent on what to do with a different legacy decision namespace like p1-decision:*. Are these to be migrated into triage:*, kept as an audit snapshot, or deleted? Suggestion: add a styleguide pattern "Reconcile or retire bespoke triage-snapshot labels" — either map p1-decision:close-if-stale → triage:stale, p1-decision:defer → future, and drop p1:verified; or explicitly bless them as an immutable audit snapshot. Without this, a reviewer cannot answer the styleguide's own quality-test cleanly for these tickets.
No "do not mutate protected / canonical-tree nodes" carve-out. The procedure's hygiene-invariant step (Step 8) would have me retype #1797 Task → Feature, but #1797 is a governed canonical-tree node that this WP is forbidden to touch. The doctrine has no concept of a node that is correctly classified-as-off-limits — the agent must import that boundary externally. Suggestion: add an entry/exit note to the procedure: "Some roots/buckets may be under change-control; record the prescribed fix as a proposal for the tree owner rather than executing it." (This is exactly what happened here.)
Provisional-priority default value is unspecified. Step 8 says "assign a provisional default and flag it" but never names the default. I chose priority:P2 (mid-scale) as the least-surprising default and flagged each. Suggestion: the styleguide should name the canonical provisional default (e.g. priority:P2 + a triage:* marker) so two operators converge instead of each picking their own mid-value.

None of these gaps blocked the run; all were absorbable by recording a proposal, which is itself the doctrine working as designed (procedure: "Where the doctrine is insufficient, that is a finding for the record, not something to silently patch around").

Summary of live mutations

Type set (objective): #719, #1622, #1623, #1624 → Task; #582 → Feature; #631 → Task.
Provisional priority priority:P2 (flagged): #1815, #1834, #1838, #1842, #1791.
#391 not closed, not reparented, not retyped (operator decision + canonical-tree protection).
Deferred proposals: #1797 retype/priority (canonical-tree owner); #1791 type (owner confirms Bug vs Task).