[Superseded]: Use ordinary GitHub issue state; do not build a triage workflow system

[morluto]

Disposition

Closed as not planned. Superseded by the direct-maintenance rule in #1148 and the closed #1501.

This issue correctly observed that fast-moving code leaves stale issue descriptions, duplicate symptoms, and unclear PR ownership.

Its proposed replacement—typed issue forms, a custom triage state machine, root-cause child tables, ownership checks, linked-PR enforcement, scheduled hygiene reports, project fields, and repository-wide backlog migration—would create another coordination system and source of truth.

Simpler practice

Use normal GitHub behavior:

• include an observed revision/reproduction in a defect when it matters;
• verify current main before implementing a stale report;
• use ordinary closing keywords for fully resolving PRs;
• link duplicates/superseding issues directly;
• revise or close an architecture issue when the target architecture changes;
• preserve unique evidence in a comment/body only when still relevant;
• use [Roadmap]: Reduce Jacobian to direct typed math tools, not a replacement framework #1148 as the current architectural direction, not a machine-enforced issue graph.

A maintainer may add or improve a small issue template when recurring low-quality reports demonstrate the need. Do not require several issue types, workflow labels, project fields, or a scheduled bot before ordinary work can proceed.

Do not introduce

• a custom issue-state machine or required triage label taxonomy;
• canonical root-cause owner/child registries;
• issue/PR ownership automation;
• source-line disappearance/revalidation bots;
• scheduled backlog hygiene services/reports as required architecture;
• project-management schemas/forms for every issue class;
• mass migration/classification of the backlog;
• a freeze until all open issues are dispositioned.

Canonical reference

• [Roadmap]: Reduce Jacobian to direct typed math tools, not a replacement framework #1148 — direct typed math-tool roadmap and PR deletion ledger.
• [Superseded]: Do not build a backlog-governance framework for the deletion program #1501 — closed backlog-governance framework proposal.

[morluto]

Audit root-cause index

Second-pass audit baseline: main at 50d663f65b45266fecadd2fe5421b662874fd25b.

Canonical architecture owners opened from this pass:

• [Architecture]: Give durable background work one runtime-owned lifecycle supervisor #1009 — durable background-worker supervision and terminalization (#679, #853, #887, #892)
• [Superseded]: Do not replace DomainBundle with provider inventories and component graphs #1010 — host provider inventory, producer/checker installation, authority, and tenant state (#856, #931; revalidate #692, #857)
• [Superseded] Redesign benchmark tasks before generating any shared Harbor protocol layer #1011 — manifest-driven Harbor verifier policy and deterministic hermetic generation (#631, #717, #728, #729, #850, #852, #854, #860, #861, #862, #885, #891)
• [Architecture]: Persist complete artifact contracts, not only JSON Schema shape #1019 — durable complete artifact contracts rather than JSON Schema shape alone (#848; revalidate #758)
• [Architecture]: Model fallible resource shutdown as an explicit terminal state machine #1021 — explicit fallible resource-shutdown state machine (#966, #968, #970, #971, #973, plus older lifecycle leaves)
• [Superseded]: Do not add verification protocol declarations/compilers; use concrete checker requests and results #1025 — first-class typed producer→verifier protocols (#886, #890, #922, #941)

Confirmed-current examples received focused comments. Reports whose source shape no longer supports the claim were marked for executable revalidation rather than closed by inspection: #851, #858, #859, #968, and #973.

No leaf issue was closed in this pass. The intended next tracker operation is to run each retained reproduction on a named revision, then classify it as confirmed-current, superseded/duplicate, or resolved by an owning PR.

[morluto]

Post-audit head check

After the static audit baseline, main advanced from 50d663f65b45266fecadd2fe5421b662874fd25b to 7d277e169c809455d1a171b3a96aaa0d915471b7 by one commit (Clarify Gaussian polynomial term ordering (#943)). The compare touches only:

• src/jacobian/contracts/probability.py
• tests/domain/probability/test_finite_probability.py

None of the code paths underlying #1009, #1010, #1011, #1019, #1021, or #1025 changed. The audit dispositions therefore remain current at this newer head; the original SHA remains recorded so every finding is reproducible.

[morluto]

Revision-aware triage pass — current main 7d277e169c809455d1a171b3a96aaa0d915471b7

Completed a root-cause pass over the open backlog and current source. I opened three missing canonical owners and adopted one existing owner rather than duplicating it:

• [Test infrastructure]: Make runtime fixtures profile-driven and cost-explicit #1032 — profile-driven, cost-explicit test runtime construction; owns [Feature]: Prefer CheckerAuthorityMode.NONE in producer tests unless asserting checker behavior #991 and test(composition): test_graph_isomorphism.py uses authorized_complete_runtime for tests that only compute #995–test(composition): consolidate parametrized tests that rebuild equivalent runtimes per case #1000. test(composition): test_graph_capabilities.py parametrized cases rebuild runtime 4x #1001 closed as duplicate of test(composition): consolidate parametrized tests that rebuild equivalent runtimes per case #1000.
• [Superseded]: Do not replace VerificationService with plan/outcome/transition frameworks #1035 — typed verification plans, one checker executor, fail-closed transition kernels; owns VerificationService is a 1488-line god class with 5 entangled verification protocols #886 and 82 fail-closed verification fixes suggest reactive rather than proactive testing #890.
• [Reliability][Deletion]: Add small operation-local bounds for proven aggregate-growth bugs; do not build a cost framework #1036 — aggregate execution/output/checker replay cost as a capability contract; owns [Bug]: IntegerSequenceRequest has no per-item digit bound #866 and Bound aggregate growth in exact mutual-information certificates #1013.
• [Architecture]: Model fallible resource shutdown as an explicit terminal state machine #1021 — existing canonical close-state architecture; owns the fallible shutdown cluster including [Bug]: TenantRuntimeRouter close can leave _closing set on BaseException #966 and [Bug]: StateDatabase.close marks database closed on teardown failure #971.

Closed after current-main verification

Issue	Disposition
#649	completed — StorageLimits now rejects max_parents < 4
#650	completed — search explicitly preflights 4 ordinary / 6 witness parent capacity before persistence
#851	completed — Windows drive-qualified/relative paths rejected via PureWindowsPath(...).drive
#855	completed — lifespan registry retains blocking task and request/tenant lease until late completion
#858	completed — URL normalization runs in mode="before", then normalized length is field-validated
#859	completed — exact domain-product cardinality checked before Cartesian materialization
#894	completed/stale — current BaseException cleanup paths re-raise rather than swallow interruption
#953	completed — math.find nested output projections are closed typed Pydantic models
#954	completed — private SDK tool-manager/model mutation removed; current seam uses public SDK APIs
#968	completed/stale — LazyLoader holds the shared RLock through load/publication, preventing the claimed race
#1017	completed — root-union math.find schema is object-rooted and covered by serialized tools/list conformance tests
#758	completed — committed and pending model bindings are both checked before transaction assignment
#1001	duplicate — same parametrized runtime-copy case as #1000

Closed as not planned / clarified contract

Issue	Disposition
#692	identity and first-use readiness are intentionally separate; required attributes are checked during readiness; broader architecture is #1010
#970	registry close is terminal for one MCP server lifespan; reuse should construct a new owner
#973	reported bare-raise failure is invalid because raise executes inside the active cleanup except block

Confirmed live or narrowed

• [Design]: Topology bundle advertises operations that fail at invocation without SymPy #857 remains live: topology's bundle declares jacobian.topology, while prime-field _domain_matrix() still imports SymPy at invocation. It is a concrete undeclared component dependency under [Superseded]: Do not replace DomainBundle with provider inventories and component graphs #1010. This corrects the earlier source-inspection assumption that all runtime SymPy use had disappeared.
• [Bug]: Encode math.find browse, search, and inspect argument modes in its input schema #952 remains live but narrowed: current runtime rejects invalid browse/search/inspect combinations, but the advertised input schema remains one flat optional-field object rather than mutually exclusive typed modes.
• [Bug]: TenantRuntimeRouter close can leave _closing set on BaseException #966 and [Bug]: StateDatabase.close marks database closed on teardown failure #971 remain live as state-model problems under [Architecture]: Model fallible resource shutdown as an explicit terminal state machine #1021; their suggested boolean fixes are not accepted verbatim because reopening partially torn-down owners is unsafe.
• [Bug]: IntegerSequenceRequest has no per-item digit bound #866 and Bound aggregate growth in exact mutual-information certificates #1013 remain live under [Reliability][Deletion]: Add small operation-local bounds for proven aggregate-growth bugs; do not build a cost framework #1036.
• mcp: define semantic pagination and transport limits for math.find #932 remains the semantic pagination/transport-limit owner; it should not be conflated with [Bug]: Encode math.find browse, search, and inspect argument modes in its input schema #952's input-mode schema.

This pass deliberately did not mass-close reports whose current invariant could not be verified directly. The new roots contain maintained child tables, migration order, acceptance criteria, non-goals, and best-practice references so implementation planning occurs at the cause while leaf issues retain their concrete reproductions.

[morluto]

Root-owner correction

The latest disposition table listed #1035 as the verification owner. During final read-back I found that #1025 already owns the same architecture—typed verification protocol declarations, explicit producer→verifier mappings, decomposition of VerificationService, closed outcomes, one record-commit kernel, and generated negative-path conformance.

• [Superseded]: Do not replace VerificationService with plan/outcome/transition frameworks #1035 is now closed as a duplicate of [Superseded]: Do not add verification protocol declarations/compilers; use concrete checker requests and results #1025.
• Its useful immutable validated execution-plan detail was added to [Superseded]: Do not add verification protocol declarations/compilers; use concrete checker requests and results #1025.
• The comments on VerificationService is a 1488-line god class with 5 entangled verification protocols #886 and 82 fail-closed verification fixes suggest reactive rather than proactive testing #890 were updated to point to [Superseded]: Do not add verification protocol declarations/compilers; use concrete checker requests and results #1025.

Final new canonical owners from this pass are therefore #1032 and #1036; existing canonical owners adopted are #1021 and #1025.