Work Packages: Ratchet, baseline & census gate remediation

Inputs: Design documents from kitty-specs/ratchet-baseline-census-gate-remediation-01M3EW3Z/ (spec.md, plan.md rev 2, research.md, data-model.md, quickstart.md, research/*) Prerequisites: plan.md (rev 2, 13 WPs), spec.md (rev 2 + errata)

Tests: Required. Every WP lands a failing-first acceptance test as its first commit (C-001), including deletion-heavy WP05. The two exceptions are the deletion-only WP07 and WP09 (operator-approved charter exceptions, plan D-OP-4 / Complexity Tracking; WP07 by Decision Moment DM-01M3F3T1G2RYW7P0ZVWQS41GEZ). WP09's evidence substitute is the committed mutation script research/wp09_mutation_matrix.py; WP07's is tracer-recorded base evidence (audit.py exit 1, rekey_inventory.py --check STALE) with a reviewer reproduction block, plus the survivor 8/8 GREEN. C-001, C-005, C-006 and NFR-005 bind every WP.

Organization: Fine-grained subtasks (Txxx) roll up into work packages (WPxx). Subtasks are reference rows, not checkboxes: record completion with spec-kitty agent tasks mark-status <Txxx> --status done.

Prompt Files: tasks/WPxx-*.md (flat). Deep implementation detail lives in the prompt files.

Dependency overview

  • Lanes: WP01; WP02 → {WP03, WP04}; WP05 → WP06; WP07; WP08; WP09; WP10; WP11; WP12 (planning artifact). WP13 depends on all.
  • Critical path: WP02 → WP04 → WP13.
  • Sequenced out-of-map edits: test_ratchet_positional_anchor_ban.py (WP01 owns; WP13 edits), _baselines.yaml (WP06 owns; WP05 edits its inert leaves first), pyproject.toml format-exclude (WP07, WP09 remove their own lines; WP13 drains the rest).

Subtask Index

IDDescriptionWPParallel
T001Widen the Python arms of the positional-anchor banWP01
T002Add the text-file arm and its standing gateWP01
T003Invert the context-gate tests and add the fixture matrixWP01
T004Record the 94-finding RED; add 94 per-site exemption rows + exactness testWP01
T005Floors and per-arm self-mutation proofsWP01
T006Docstring rewrite, durations record, quality gatesWP01
T007Red-first join acceptance tests (stale + drift)WP02
T008Create _content_identity.py and test_content_identity.pyWP02[P]
T009Migrate the join allowlist to 4 content descriptorsWP02
T010Non-widening join proofs and docstring rewriteWP02
T011Validation, tracer, quality gatesWP02
T012Red-first kernel/os-detect drift and loader-rejection testsWP03
T013Migrate the kernel exemptionsWP03
T014Migrate the os-detect loader, .txt rows and gateWP03
T015Retarget the lock-ban and clock CALL: loadersWP03
T016Validation, tracer, quality gatesWP03
T017Red-first census drift tests and non-widening guardsWP04
T018Unify the qualname algorithm; add CensusKey and census_keysWP04
T019Delegate the census partitioner to _content_identityWP04
T020Re-key/equivalence script (--head-root, --self-test) and census-rekey-map.csv, committed on the planning branchWP04[P]
T021Re-key the destructive-op gate (22)WP04
T022Re-key the overwrite gate (2)WP04
T023Re-key the mutation gate (56)WP04
T024Non-widening, ordinal and diagnostics proofsWP04
T02580/80 equivalence proof plus self-test outputWP04
T026Full architectural run and quality gatesWP04
T027RED first: test_load_baseline_rejects_retired_keys; base evidence (0 callers, stale-row warning, #5117 reproduction)WP05[P]
T028Delete dead machinery in _inert_slots.pyWP05
T029Prune _inert_slots_baseline.yaml + parser (36 rows)WP05
T030Sequenced _baselines.yaml inert-leaf edit + test_reference_enum_ratchet.py proseWP05
T031Wire per-walk floors with self-mutation; token search; evidenceWP05
T032RED first: _SIZE_RATCHETS table, leaf helpers, leaf-enforcement testWP06
T033Rewire growth and shrink arms onto _SIZE_RATCHETSWP06
T034Derive hand-kept key lists; retire grandfather setWP06
T035Remove category_1 leaf + machinery; survivor proof; ADR bulletWP06
T036Remove skip_marker_blocks leaf + machinery; round-trip commentsWP06
T037Lower-below-live mutations, planted-leaf self-mutation, floorsWP06
T038Charter-exception evidence: tracer-record base audit exit 1 (8+8), --check STALE, token count; survivor baseline 8/8WP07[P]
T039git mv audit.py → _surface_resolution_scan.py, strip, survivor import switch, pyproject L954WP07
T040Delete remaining surface_resolution_audit/ files, pyproject L955WP07
T041Rewire survivor test_single_mission_surface_resolver.pyWP07
T042FR-009 reference edits, path-qualified token search 0, validationWP07
T043RED first: floor/missing/planted tests for both vacuous scansWP08[P]
T044Convert rich/typer ban onto live _internal_runtimeWP08
T045Retire duplicate ban + surface pins; relabel goldenWP08
T046Retarget test_next_no_unknown_state runtime scan + planted placeholderWP08
T047FR-016 residue in surface-resolution equivalence + planted divergenceWP08
T048FR-016 residue in three parity suites; validationWP08
T049Reproducible mutation matrix M1–M10 (research/wp09_mutation_matrix.py) and disposition evidence (tracer)WP09
T050Relocate unique determinism tests into test_reducer.py; retire duplicates (incl. L407/L660)WP09[P]
T051Transition-matrix dispositions (duplicates retired; lane↔enum relocated only if no survivor)WP09[P]
T052Delete tests/status/test_parity.py and its pyproject exclude lineWP09
T053NFR-006 tracer record and blast-radius validationWP09
T054Red-first src-coupling AST scan (exact 10-entry offender list on base) and self-mutation testWP10
T055Copied SPEC_KITTY_PACKS_ROOT mirror and on-disk fixture profileWP10
T056Convert marker tests; pack-root and control assertions; cache-order checkWP10
T057Rename to test_context_bootstrap_markers.py; update references (AST proof for src comment)WP10
T058Blast radius, tracer, deferral recordWP10
T059Base node-ID capture and red-first oracle-independence guards (expected-name set, --setup-plan check)WP11
T060Verbatim move of helper closure into _next_mission_scaffold.pyWP11
T061Verbatim move of 15 P0/fail-closed test functionsWP11
T062Format commit, then public-name rename commitWP11
T063Update test_finalized_task_routing.py and fixtures/bridge/README.md referencesWP11
T064Node-ID/oracle-identity/setup-plan/durations evidenceWP11
T065Red-first parity_catalog_check.py (45-module set equality)WP12
T066Author research/parity-verdicts.md (45 rows, #2620 format)WP12
T067Tracer pointer and WP13 reconciliation hand-offWP12
T068Red-first test_positional_anchor_exemptions_are_pinned_emptyWP13
T069Empty exemptions; warn→fail flip with self-mutationWP13
T070Retire resolution_gate_allowlist.yaml and update references (FR-012)WP13
T071Format-exclude drain for AST-changed excluded filesWP13
T072FR-019(d) and out-of-scope follow-up issues; verify #5116–#5118WP13[P]
T073Finalize the pre-seeded in-mission issue-matrix rows to terminal verdicts via spec-kitty agent issue-verdictWP13[P]
T074Final integration sweepWP13
T075SC-001..SC-006 verification and tracer assessWP13

WP01 – Widen positional-anchor ban + interim per-site exemptions

Summary

  • Goal: close the positional-anchor ban's three blind spots (substrate-import gate, Path(...) path elements, path:line[:op] keys) plus keyword records, class-body seeds and _exemptions/*.txt lines, then land GREEN through 94 exact per-site interim exemption rows (FR-001, FR-002, FR-003 interim, NFR-002, NFR-004).
  • Priority: P1 (US1; unblocks the SC-001 pin in WP13).
  • Independent test: with an empty exemption set the widened standing gates report 94 findings (6 / 2 / 22 / 56 / 2 / 6); with the rows they pass; a planted (Path("a.py"), 12), {"x.py:9:op": "r"}, Entry(path=..., lineno=3), class-attribute tuple or CALL:x.py:12 text line is flagged, and each arm-disable self-mutation turns its plant green.

Subtasks T001 Widen the Python arms: drop the import gate, add path-ish element, file-line tuple, embedded-key, keyword-record and class-body arms; carry symbol + site on findings (WP01) T002 Add the text-file arm and test_no_positional_anchor_in_architectural_text_files (WP01) T003 Invert the four context-gate tests and add the positive/negative fixture matrix (WP01) T004 Record the 94-finding RED, then add the 94 one-line per-site exemption rows and test_positional_anchor_exemptions_are_exact (WP01) T005 Floors (≥ 262 py files, ≥ 20 txt files, ≥ 6 entry lines) and per-arm self-mutation proofs (WP01) T006 Rewrite the module docstring, record --durations=0, run quality gates (WP01)

Implementation sketch: commit 1 = widened arms + inverted tests + empty exemption set (RED, 94); commit 2 = generated rows (GREEN); commit 3 = floors, self-mutations, docstring. All predicates stay in the ban module (C-005: is_file_line_anchor untouched). File is format-excluded (pyproject L1019): no ruff format.

Dependencies: none.

Risks: false positives from widened arms (fix predicate, never exempt); rows over 164 chars (module constants); escape-hatch # diagnostic-locator abuse (forbidden); exact-set interplay with parallel lanes (row-without-finding only warns until WP13).

Estimated prompt size: ~265 lines.

Prompt file: tasks/WP01-widen-positional-anchor-ban.md

WP02 – Content-identity matching authority + join allowlist

Summary

  • Goal: create tests/architectural/_content_identity.py, the single multiset matcher and descriptor-line serialiser for content-keyed allowlists (D-OP-9), and migrate _KNOWN_JOIN_ALLOWLIST (6 line pins) to 4 ContentDescriptors, deleting the 2 dead entries (FR-004, FR-007, NFR-001, NFR-003).
  • Priority: P1 (critical path WP02 → WP04 → WP13; WP03 also depends on it).
  • Independent test: test_join_allowlist_entries_each_suppress_a_live_join is RED on base naming src/kernel/paths.py:88 and src/specify_cli/runtime/home.py:79 and GREEN after; test_join_allowlist_survives_line_drift (3 derived files, both mutations) keeps (unexpected, suppressed) identical; a new join planted at old kernel/paths.py:88 is reported.

Subtasks T007 Red-first join acceptance tests (stale entries + line drift) through an extracted gate seam (WP02) T008 Create _content_identity.py (resolve, multiset partition, drift mutators, descriptor-line render/parse) and test_content_identity.py (WP02) T009 Migrate the join allowlist to 4 lazily resolved ContentDescriptors matched by partition_findings (WP02) T010 Non-widening proofs (new join at formerly pinned line; duplicated key) and docstring rewrite (WP02) T011 Validation, tracer record of non-adopter matchers, quality gates (WP02)

Implementation sketch: commit 1 = acceptance tests only (RED); commit 2 = helper + unit tests; commit 3 = migration + proofs. Counter-based partition keyed on full (rel_path, qualname, token_line); stale = unresolvable or suppresses nothing (fail). Resolve lazily behind functools.cache.

Dependencies: none.

Risks: becoming an eighth matcher (docstring claims authority, lists non-adopters for FR-019(d)); set semantics reopening the neutrality/lint.py 379/380 widening hole; import-time resolution turning a stale entry into a collection error.

Estimated prompt size: ~260 lines.

Prompt file: tasks/WP02-content-identity-authority-and-join-allowlist.md

WP03 – Kernel + os-detect exemptions; retarget lock-ban/clock loaders

Summary

  • Goal: migrate the 2 kernel _PRE_EXISTING_EXEMPTIONS tuples and the 6 path:line rows in _exemptions/os-detect-ban-*.txt to content identity, and retarget the lock-ban and clock CALL: loaders to the same content shape at zero entries so no loader accepts or teaches path:line (FR-005, FR-007, NFR-001, D-OP-6).
  • Priority: P2.
  • Independent test: test_kernel_exemptions_survive_line_drift (1 file) and test_os_detection_exemptions_survive_line_drift (3 files) keep (unexpected, suppressed) identical under both mutations; each of the three loaders raises ValueError on a path:line line; stale entries fail with checked >= 2 / >= 6.

Subtasks T012 Red-first tests: kernel and os-detect drift tests plus three loader-rejects-line-pin tests (WP03) T013 Migrate the kernel exemptions to two descriptors; invert the real-violation test (WP03) T014 Migrate the os-detect loader, the three .txt files and the gate; invert the stale-removal test (WP03) T015 Retarget the lock-ban and clock CALL: loaders and their gates to the content shape (WP03) T016 Validation, tracer (#3206, #4727, D-OP-6, residual no-follow loader), quality gates (WP03)

Implementation sketch: one RED commit, then one commit each for kernel, os-detect and the loader retarget. Everything parses/renders/partitions through WP02's _content_identity. _exemptions/__init__.py and test_clock_call_ban.py are format-excluded (no ruff format).

Dependencies: WP02.

Risks: three private parsers (use the shared pair only); clock loader globbing every *.txt (prefix filter must ignore content lines); weakening the inverted stale-removal tests; test_clock_import_ban.py (not owned) must stay green unchanged.

Estimated prompt size: ~255 lines.

Prompt file: tasks/WP03-kernel-os-detect-exemptions-and-loader-retarget.md

WP04 – Census re-key (all 80) on _content_identity + composite key

Summary

  • Goal: re-key the 80 rel:lineno:op census keys (destructive 22, mutation 56, overwrite 2) to CensusKey(rel, qualname, token_line, op, op_ordinal), delete the census helper's own qualname algorithm (C-004), delegate diff_against_allowlist to partition_findings, keep warn-on-stale, and prove 80/80 site equivalence against planning base 3717c7ea (FR-006, C-004, NFR-001).
  • Priority: P1 (critical path; largest WP, size L).
  • Independent test: test_{destructive,overwrite,mutation}_census_survives_line_drift (15 / 2 / 21 derived files) RED on base and GREEN after; a second identical op in an exempted function fails; a token-changing argument edit fails and warns stale; census_rekey_equivalence.py --base 3717c7ea reports 80/80, a bijection and equal rationale hashes; mutation _ALLOWLIST still has 56 entries.

Subtasks T017 Red-first census drift tests and non-widening guards through extracted gate seams (WP04) T018 Delete the census enclosing_qualname, re-export the canonical one, add parse_with_source, CensusKey, census_keys; fix all callers in the same commit (WP04) T019 Delegate diff_against_allowlist / drop_one_entry generically to _content_identity.partition_findings (WP04) T020 Write the re-key/equivalence script (--head-root, --self-test) and generate census-rekey-map.csv; commit both on the planning branch before review (WP04) T021 Re-key the destructive-op gate (22 keys) (WP04) T022 Re-key the overwrite gate (2 keys) (WP04) T023 Re-key the mutation gate (56 keys), including the research.py prefix guard and literal re-builder (WP04) T024 Non-widening, ordinal and diagnostics proofs; baseline coupling check (WP04) T025 Run the 80/80 equivalence proof and its self-test; record both in PR and tracer (WP04) T026 Full tests/architectural/ run, blast radius, quality gates, campsite (WP04)

Implementation sketch: four lane commits (RED → helper → destructive+overwrite → mutation). Allowlist literals are generated in keyword form with rationales copied verbatim. CSV and equivalence script are mission artefacts written outside the lane (code_change WPs cannot own kitty-specs/).

Dependencies: WP02.

Risks: widening via a weak key (guarded by ordinal + token-line tests); laundering coincidental blessings (equivalence against base, not head); rationale drift (hash check); _baselines.yaml coupling (keep a 56-entry mapping, never edit the YAML); review load (staged commits, generated literals).

Estimated prompt size: ~305 lines.

Prompt file: tasks/WP04-census-rekey-content-identity.md

WP05 – Inert-slot retirement (#3026, #3962)

Summary

  • Goal: retire the dead inert-slot machinery that #3285 orphaned (FR-010, DM-01M3EW4PB6).
  • Remove the uncalled caps, owner predicates and code-only suppression record from _inert_slots.py, along with its materialize_snapshot src import.
  • Drop owner/provisional/mission data and the 2 stale rows (styleguide-references, model) from _inert_slots_baseline.yaml, taking it from 38 to 36 rows.
  • Delete the unassigned_entries and masking_suppressions leaves from _baselines.yaml and set baseline_entries from 38 to 36. This is a sequenced edit into the WP06-owned file.
  • Wire the calibrated per-walk floors (150 schema names, 120 model names).
  • Priority: P2 (US2). It unblocks WP06.
  • Independent test:
  • pytest tests/architectural/test_no_inert_schema_slots.py -q -rw passes with no "baseline shrank" warning; on base the warning names styleguide-references, model.
  • The retired-token search returns 0 live hits.
  • test_walk_floors_fail_on_a_collapsed_walk reds both walks through the real scanned_slots.
  • test_load_baseline_rejects_retired_keys is RED on base (first commit) and GREEN after the prune.

Subtasks T027 RED first: test_load_baseline_rejects_retired_keys (4 cases, RED on base); base evidence (0 callers, stale-row warning, #5117 reproduction) (WP05) T028 Delete dead machinery in _inert_slots.py (caps, owner predicates, code-only record, src import) (WP05) T029 Prune _inert_slots_baseline.yaml + parser (36 rows, no owner/provisional/mission/code_only; reject retired keys, turning T027 GREEN) (WP05) T030 Sequenced _baselines.yaml edit (38→36, delete 2 inert leaves) + test_reference_enum_ratchet.py:192 prose (WP05) T031 Wire per-walk floors with self-mutation; token search 0; green evidence (WP05)

Implementation sketch: commit the failing-first retired-key test, gather base evidence and confirm #5117, then do the atomic parser+YAML+module prune (T028+T029 in one commit), then the _baselines.yaml block edit, then the floors test and the final token search.

Dependencies: none (#5117 filed as the FR-019(b) precondition).

Risks:

  • The import-time load_baseline() breaks if the parser and YAML diverge. Land T028 and T029 atomically.
  • The only model detector is lost unless #5117 carries its reproduction.
  • The excluded test file must not become format-clean.

Estimated prompt size: ~290 lines

Prompt file: tasks/WP05-inert-slot-retirement.md

WP06 – Baseline leaf enforcement

Summary

  • Goal: the charter ratchet refuses any _baselines.yaml leaf that no comparison enforces (FR-011, NFR-003).
  • Hoist one _SIZE_RATCHETS table (19 rows) and make both arms iterate it.
  • Derive _REQUIRED_TOP_LEVEL_KEYS and _REQUIRED_NO_DEAD_MODULES_CATEGORIES from it, and retire the grandfather set.
  • Remove the derived category_1 and advisory skip_marker_blocks leaves (D-OP-2).
  • Prove every row reads its own leaf by lowering each to live − 1.
  • Priority: P1 (US2, SC-004).
  • Independent test:
  • test_every_baseline_leaf_is_enforced_by_a_size_ratchet is RED first. On the lane base it names 2 leaves; on the planning-base YAML, 4.
  • test_lowering_an_enforced_leaf_below_live_fails has 19 ids, all red-capable.
  • Planting a leaf fails and names it.

Subtasks T032 RED first: _SizeRatchet/_SIZE_RATCHETS, _enforced_leaves/_yaml_leaves/_leaf_drift, leaf-enforcement test (WP06) T033 Rewire growth and shrink arms onto _SIZE_RATCHETS via _size_comparisons (WP06) T034 Derive hand-kept key lists; retire _GRANDFATHERED_UNREGISTERED_KEYS; fold unregistered-key tests (WP06) T035 Remove category_1 leaf + derived machinery/tests; survivor mutation proof; ADR bullet supersession (WP06) T036 Remove skip_marker_blocks leaf + advisory machinery/tests; round-trip comment edits (WP06) T037 Lower-below-live mutation tests (19), planted-leaf self-mutation, table floor, shrink-only proof (WP06)

Implementation sketch: a red-first table and test commit, then a pure rewire, then the derived registries, then the two leaf removals with NFR-006 survivors, then the mutation tests.

Dependencies: WP05, the sequenced _baselines.yaml inert-leaf edit.

Risks:

  • The rewire could silently drop a row. The 19-id lowering test and the unchanged arm tests guard against this.
  • Fast-tier import hygiene: never import the round-trip corpus at module scope.
  • Cross-lane coupling on len(test_mutation_ownership_routing._ALLOWLIST) with WP04.
  • The excluded file must not become format-clean.

Estimated prompt size: ~330 lines

Prompt file: tasks/WP06-baseline-leaf-enforcement.md

WP07 – Surface-resolution converter retirement

Summary

  • Goal: retire the orphaned converter, inventory and audit entry point (FR-008, FR-009, DM-01M3EW4J8E).
  • git mv audit.py to tests/architectural/_surface_resolution_scan.py, stripped to the scanner functions the survivor uses. Fix the _REPO_ROOT depth and import CompositeKey from _ratchet_keys.
  • Delete the rest of surface_resolution_audit/.
  • Rewire test_single_mission_surface_resolver.py to a normal import.
  • Fix the FR-009 references, including the dangling untrusted_path_audit/inventory.md:76 cross-reference.
  • Remove its own two pyproject format-exclude lines.
  • Priority: P1 (US2, #3011 is P1).
  • Independent test:
  • Charter exception (plan D-OP-4, DM-01M3F3T1G2RYW7P0ZVWQS41GEZ): the tracer records the base defect (audit.py exit 1 with 8 missing + 8 ghost rows; rekey_inventory.py --check STALE), and the reviewer re-runs it from the prompt's reproduction block.
  • The survivor test_single_mission_surface_resolver.py, re-pointed at _surface_resolution_scan, keeps the same 8 node IDs as base, all green.
  • The path-qualified token search (surface_resolution_audit|rekey_inventory|write_candidate_classification) finds hits on base and 0 after.
  • test_untrusted_path_containment.py stays green.

Subtasks T038 Charter-exception evidence via spec-kitty agent tracer-append: base audit exit 1 (8 missing + 8 ghost), --check STALE, base token count, survivor 8/8 baseline; no import-error RED commit (WP07) T039 git mv audit.py → _surface_resolution_scan.py, strip to scanner, fix root depth, drop bootstrap/noqa, re-point the survivor's import in the same commit, pyproject L954 (WP07) T040 Delete rekey_inventory.py, inventory.md, RULESET.md, audited-surfaces.md, write_candidate_classification.yaml; pyproject L955 (WP07) T041 Finish the survivor rewire (import switch committed with T039); fix its docstrings, floor comment and failure messages (WP07) T042 FR-009 references (_ratchet_keys.py, test_no_worktree_name_guess.py:155, untrusted_path_audit/inventory.md:76), token search 0, validation (WP07)

Implementation sketch: record the charter-exception evidence in the tracer, then do move + strip + survivor import switch + pyproject line in one commit (survivor stays green), then delete the remaining files with their pyproject line, then the reference edits and the final search.

Dependencies: none.

Risks:

  • A wrong _REPO_ROOT depth makes the scan silently empty. The survivor's floor of 15 catches it.
  • A pyproject entry could point at a moved or deleted file mid-series. Remove each line in the same commit as its file.
  • The untrusted_path_audit/inventory.md table is parsed, so edit the rationale cell only.

Estimated prompt size: ~300 lines

Prompt file: tasks/WP07-surface-resolution-converter-retirement.md

WP08 – Runtime-parity bans, parity residue and next-no-unknown-state scan

Summary

  • Goal: make the runtime-parity bans non-vacuous and remove stale parity residue (FR-013, FR-016).
  • Convert the rich/typer ban to an AST helper over the live src/runtime/next/_internal_runtime (floor ≥ 16).
  • Retire the duplicate spec_kitty_runtime ban and the two surface-shape pins, with NFR-006 evidence.
  • Fix the vacuous src/specify_cli/next rglob in tests/contract/test_next_no_unknown_state.py:40, retargeting it to src/runtime/next (floor ≥ 31).
  • Remove dead strict-xfail machinery, stale docstrings, the name-only missing_seams arm, the self-referential docstring test and the --check-residual tombstone.
  • Priority: P2 (US3, SC-005).
  • Independent test:
  • test_rich_typer_ban_inspects_live_runtime_package and test_runtime_placeholder_scan_inspects_live_source are RED on base (0 files inspected).
  • The missing, empty and planted tests fail or flag through the real helpers.
  • test_equivalence_detects_planted_divergence reds _assert_equivalent.

Subtasks T043 RED first: hoist current targets, add AST/placeholder helpers and floor/missing/planted tests (WP08) T044 Convert rich/typer ban onto live _internal_runtime (AST, floor ≥ 16, load-bearing proof) (WP08) T045 Retire duplicate spec_kitty_runtime ban + 2 surface pins (NFR-006 survivor/mutation); relabel golden docstring (WP08) T046 Retarget test_next_no_unknown_state runtime scan to src/runtime/next (floor ≥ 31) + planted placeholder (WP08) T047 FR-016 residue in test_surface_resolution_equivalence.py + planted-divergence test (WP08) T048 FR-016 residue in execution-context, transition-gate and docs-CLI parity suites; tracer record; validation (WP08)

Implementation sketch: a red-first commit with floor tests, then retarget both scans, then the retirements with evidence, then the residue deletions in four format-excluded files (not reformatted).

Dependencies: none.

Risks:

  • Six format-excluded files must not become format-clean; escalate if one does.
  • Floors are pinned to planning-base counts, so a deliberate future shrink needs a one-line edit.
  • The rich/typer scope must stay _internal_runtime, because runtime_bridge_retrospective.py:158 is sanctioned.

Estimated prompt size: ~320 lines

Prompt file: tasks/WP08-runtime-parity-bans-and-residue.md

WP09 – Status parity retirement and relocation

Summary: Delete tests/status/test_parity.py (740 LOC of 0.1x cross-branch scaffold) so that no live invariant is lost (FR-014, NFR-006). Each of the 21 test functions gets a disposition:

  • retire (duplicate), naming the survivor and a mutation that turns both tests red;
  • retire (scaffold), citing the reason plus a mutation;
  • relocate, into test_reducer.py or test_transitions.py, but only where no survivor exists.

The five transition-matrix tests are presumed duplicates of test_transitions.py. That presumption is confirmed by mutations M2–M6 before any test is retired. The WP also removes its own pyproject.toml format-exclude line. Priority: P2 (US3). Independent test: pytest tests/status/ tests/architectural/test_no_retired_subsystems.py is green after deletion. The recorded mutation matrix M1–M10 reds each named survivor or relocated test. Red-first runs under the operator-approved charter exception (plan D-OP-4 / Complexity Tracking): the committed mutation script is the evidence substitute; no tombstones.

Subtasks T049 Reproducible mutation matrix M1–M10 (research/wp09_mutation_matrix.py, committed on the planning branch); record its verbatim output via tracer-append (WP09) T050 Relocate test_sorted_keys_in_json_output and the realistic-log fixture plus 2 tests into test_reducer.py; retire the determinism duplicates, including dispositions for L407 and L660 (WP09) T051 Transition-matrix dispositions: retire as duplicates with named survivors; relocate the lane↔enum checks into TestConstants only if M5/M6 find no survivor (WP09) T052 git rm tests/status/test_parity.py and remove its pyproject exclude line in the same commit; retire the backport/phase scaffold with cited survivors (WP09) T053 NFR-006 tracer record and blast-radius validation, including the full tests/architectural/ run (WP09)

Implementation sketch: run the mutation matrix → relocate unique tests in the existing unformatted style → delete the file plus its exclude line → record everything in the tracer. Dependencies: none. The pyproject hunk is about 1,600 lines away from WP07's, so it needs no sequencing. Risks: a mutation leaking into a commit (check git diff src/ is empty); the lane-enum tests being treated as duplicates by mistake (M5/M6 decide); relocated code becoming format-clean and flipping the exclude ratchet; #4506 landing mid-mission. Estimated prompt size: ~259 lines. Prompt: tasks/WP09-status-parity-retirement.md

WP10 – Context parity conversion to an on-disk packs fixture

Summary: Convert tests/charter/test_context_parity.py from patching src/ internals to a copied SPEC_KITTY_PACKS_ROOT packs mirror, with an empty directives/ and an on-disk fixture profile, driven through the public entry points. Then rename it to test_context_bootstrap_markers.py (FR-015, D-OP-5, D-OP-10). Priority: P2 (US3). Independent test: test_context_markers_use_no_src_patch_targets. It is RED on base with exactly 10 entries (6 patch( sites, 2 private imports, 2 private calls) and GREEN with 0. The four behavioural markers hold, and a control shows a real directives mirror changes the miss cause.

Subtasks T054 Red-first AST scan _src_coupling_offenders, forbidding all first-party patch targets and private calls, plus a planted self-mutation test (WP10) T055 _mirror_packs copy fixture (never symlink) and the on-disk parity-fixture-agent.agent.yaml (WP10) T056 Convert the marker tests; add the pack-root-under-tmp_path/no-UserWarning assertion and the fixture-mutation control; run the two-order cache check (WP10) T057 git mv to test_context_bootstrap_markers.py; update 4 references, including the context_contract.py comment with a D-OP-3 AST-equality proof (WP10) T058 Blast radius, tracer record, and the deferral decision record (WP10)

Implementation sketch: red-first scan → mirror fixture → drop every patch while keeping the assertions byte-identical → rename in its own commit → format. Dependencies: none. Risks: process-wide caches (checked in both orders); #3251 fail-open masking a broken mirror (the positive missing_artifact and pack-root assertions guard this); scan evasion through aliases (planted alias case). If the markers cannot be reproduced without src/ patches, defer FR-015; never re-patch privates. Estimated prompt size: ~276 lines. Prompt: tasks/WP10-context-parity-conversion.md

WP11 – Bridge parity split: P0 board-authority tests out of the oracle module

Summary: Move the 13 P0 board-authority tests and the 2 fail-closed direct-call tests (15 functions, 16 nodes) from tests/runtime/test_bridge_parity.py into the new tests/runtime/test_next_board_authority.py. Their helper closure moves into the new tests/runtime/_next_mission_scaffold.py. The oracle module keeps 8 nodes, and the oracle stays untouched (FR-017, C-002, NFR-004). Priority: P2 (US3). Independent test: test_board_authority_module_does_not_import_the_oracle is RED on base (fewer than 15 test functions) and GREEN after the move. Node-ID set equality (func[param]) holds: 16 in the new module, 8 in the oracle. --setup-plan shows no ledger_results, and the module runs in under 60 s (recorded, not asserted).

Subtasks T059 Capture the base node-ID sets; add the red-first guard module with _oracle_coupling_offenders, an expected-name-set floor, a --setup-plan oracle-fixture check and a planted self-mutation test (WP11) T060 Verbatim move of the helper closure into _next_mission_scaffold.py (hoisting the imports drops 2 noqa: E402); the oracle re-imports from it (WP11) T061 Verbatim move of the 15 test functions and _assert_reason_has_runnable_recovery_command (same "move" commit as T060) (WP11) T062 Commit 2: ruff format the new modules; commit 3: rename helpers to public names and update call sites (WP11) T063 Update references in tests/next/test_finalized_task_routing.py:269 and tests/runtime/fixtures/bridge/README.md (WP11) T064 Evidence: node-ID diffs, oracle byte-identity, --setup-plan, durations, tracer entries (WP11)

Implementation sketch: red-first guard → move (1 commit) → format (1 commit) → rename (1 commit) → references → evidence. Dependencies: none. Risks: an unreadable ~1.6k-line diff (mitigated by 3 commits plus --color-moved); hidden fixture dependencies (--setup-plan); renames touching oracle assertions (a C-002 breach, so revert). _bridge_oracle.py is not edited; its bridge:NNNN anchors are deferred to #5116. Estimated prompt size: ~288 lines. Prompt: tasks/WP11-bridge-parity-split.md

WP12 – Parity verdict catalog (#2631 sweep, 45 modules)

Summary: A planning artifact that records a verdict for every module in the #2631 parity/equivalence sweep. The sweep is 49 find hits minus the 4 _support/coverage_safety helpers, giving 45 modules. The catalog lives at kitty-specs/<mission>/research/parity-verdicts.md in #2620 format, with churn (all/src/mass/pcm/pcsrc) and an NFR-006 survivor column. A tracer-append pointer records it (FR-018). Priority: P2 (US3). Independent test: research/parity_catalog_check.py --base 3717c7ea exits non-zero on base (no catalog) and prints 45/45 OK at WP end. It checks set equality against git ls-tree, and that row shape and survivors are present.

Subtasks T065 Red-first stdlib checker parity_catalog_check.py (set equality against git ls-tree 3717c7ea, row-shape and survivor checks) (WP12) T066 Author parity-verdicts.md: 35 keep rows plus 10 acted-on or deferred rows, each with a named owning WP and survivor (WP12) T067 Tracer pointer entry, and a reconciliation section handed to WP13 (WP12)

Implementation sketch: checker first → catalog from the grounding Part 2 data (a shallow clone, so the churn provenance is cited) → tracer pointer. Dependencies: none. WP13 reconciles the WP08–WP11 rows against the landed code. Risks: unreproducible churn numbers (cite provenance and the command); verdict drift if WP10 defers (the owning-WP column plus WP13 reconciliation catch this); scope creep into consolidating the deferred rows. Estimated prompt size: ~229 lines. Prompt: tasks/WP12-parity-verdict-catalog.md

WP13 – Closeout: pin exemptions empty, retire orphan gate data, format-exclude drain, follow-ups, issue matrix

Summary: The mission closeout. It covers:

  • pinning _POSITIONAL_ANCHOR_EXEMPTIONS == frozenset() and flipping stale rows from warn to fail (FR-003 final, SC-001);
  • retiring tests/architectural/resolution_gate_allowlist.yaml and its references, including a D-OP-3 comment-only edit to src/specify_cli/status/aggregate.py:543 (FR-012);
  • ruff formatting every format-excluded file this mission rewrote and removing their exclude lines in one commit;
  • filing the FR-019(d) follow-up for the remaining hand-rolled matchers and verifying #5116/#5117/#5118;
  • finalizing the pre-seeded in-mission issue-matrix rows (#5085, #2631, #2972, #5104, and any an owning WP left open) to terminal verdicts via spec-kitty agent issue-verdict (FR-020, SC-006);
  • the final full sweep.

Priority: P1 (it gates the mission PR). Independent test: test_positional_anchor_exemptions_are_pinned_empty is RED with 94 rows and GREEN with frozenset(). The full tests/architectural/ run, make test-fast and census_rekey_equivalence.py --base 3717c7ea are green. SC-001..SC-006 are evidenced in the tracer.

Subtasks T068 Red-first test_positional_anchor_exemptions_are_pinned_empty (after confirming 0 live findings remain) (WP13) T069 Empty the exemption set; flip row-without-finding from warn to fail, with a self-mutation test; NFR-003/NFR-004 records (WP13) T070 FR-012: delete resolution_gate_allowlist.yaml, add a _YAML_ALLOWLISTS floor, update prose/docs/aggregate.py comment with an AST proof, run the orphan-data audit (WP13) T071 Format-exclude drain: format each AST-changed excluded file and remove its line, in one formatter-only commit (WP13) T072 File the FR-019(d) follow-up plus the plan's out-of-scope follow-ups; verify #5116/#5117/#5118 (WP13) T073 Finalize the pre-seeded in-mission issue-matrix rows to terminal verdicts via spec-kitty agent issue-verdict; close referenced-but-missing gaps; record the FR-020 campsite (WP13) T074 Final integration sweep (full tests/architectural/, make test-fast, integration list, equivalence script) (WP13) T075 SC-001..SC-006 verification, WP12 catalog reconciliation via the tracer, tracer assess step (WP13)

Implementation sketch: pin red → empty and flip → retire orphan YAML → formatter-only drain → tracker work → matrix → sweep → SC evidence. Dependencies: WP01, WP02, WP03, WP04, WP05, WP06, WP07, WP08, WP09, WP10, WP11, WP12. Risks: a migration WP leaving a live finding (stop and report, never delete that row); formatting hiding behaviour edits (a formatter-only commit plus git diff -w); #4506 landing mid-closeout (re-derive candidates via AST comparison); issue-matrix classification surprises (use not-applicable, never hand-edit); C-005 breach (AST proof mandatory). Estimated prompt size: ~335 lines. Prompt: tasks/WP13-closeout.md