Spec Kitty

Implementation Plan: Doctrine Delivery Reachability

Branch: feat/doctrine-delivery-reachability | Date: 2026-07-28 | Spec: spec.md Input: Feature specification from kitty-specs/doctrine-delivery-reachability-01KYMXD6/spec.md

Branch contract: current branch feat/doctrine-delivery-reachability; planning/base branch feat/doctrine-delivery-reachability; final merge target feat/doctrine-delivery-reachability (the reconciliation target — a PR to upstream/main is raised after consolidation). branch_matches_target: true.

Summary

Close the delivery layer of the declared-but-inert defect class: a declaration that passes every check and still reaches nobody. Three strands share one mechanism — a projection table narrower than the enumeration it projects, silenced by a .get(…, default) or a literal [].

Per C-010 the mission extends the two class-closing mechanisms the repository already ships (derivation-from-model_fields; totality-or-explicit-exemption) rather than hand-fixing three instances. Concretely: a writer registry the field-completeness test iterates, and the totality guard extended over the project-tier kind mapping.

Delivery is then made real end to end — assets become resolvable and bundle-delivered, activated doctrine reaches the render on every load for every kind, and reachability becomes a named set under a named traversal at two named depths rather than a count that four independent lenses measured four different ways.

Technical Context

Language/Version: Python 3.11+ Primary Dependencies: pydantic v2 (model_fields, ConfigDict), ruamel.yaml, typer, rich, pytest Storage: On-disk YAML/JSONL — sharded DRG fragments (src/doctrine/*.graph.yaml), .kittify/charter/charter.yaml, .kittify/config.yaml, status.events.jsonl Testing: pytest; ATDD red-first per C-006 with executable, committed mutation proofs (model-extension fixtures, not narrated manual edits); named gate files only — never the full architectural suite in a WP loop Target Platform: Cross-platform CLI (Linux/macOS/Windows), distributed as a PyPI wheel; built-in doctrine ships as package data resolved via importlib.resources Project Type: single Performance Goals: Steady-state context render stays within the existing BUDGET_DEFAULT = 32_000 budget, handled by the degradation-to-fetch-stanza path — never by truncation (NFR-003). Delivered id set per boundary carries a shrink-only ceiling of 90. Constraints: Layer direction kernel ← doctrine ← charter ← specify_cli (C-001) — the derived serializer lives in doctrine and must reach specify_cli through a charter facade, not a function-local import. Zero new # noqa / # type: ignore (NFR-002). Complexity ceiling 15. Consumer mutation bounded to the activation surface (NFR-001). Scale/Scope: 310 DRG nodes / 781 edges; 184 activated artefacts; 24 action nodes; 18 agent profiles. Estimated 45–60 files across nine concerns.

Planning decisions taken (2026-07-28)

Recorded via the Decision Moment protocol; all four resolved, none deferred.

Decision	Outcome
Action-channel traversal	resolve_context asserted at both d=1 (compact, the steady state) and d=2 (bootstrap)
Delivery channels	Two — the action channel and the profile channel. Each carries its own named reachability set and its own delivery obligation (FR-020). Reach is never claimed without delivery (C-008)
Writer class-closure shape	Registry every edge-serializing writer joins — but in three shapes, not one (see IC-01)
Asset delivery	Assets are delivered, not activation-gated. Expressed as a total per-kind gate function, not an enumerated exception
FR-010 latency	The measured ~2x steady-state regression is accepted and recorded, not gated (NFR-007)

Corrected measurements (post-plan squad, re-derived with resolve_context)

The pre-squad figures in this plan were produced by a hand-written BFS — the method this plan's own binding-method note forbids. They are withdrawn. Re-derived:

measure	withdrawn figure	correct figure
activated-and-unreachable, action channel, d=1	136	103
activated-and-unreachable, action channel, d=2	87	96
contribution of profile seeds to resolve_context	+43 / +58	0 at both depths

Why profile seeding contributes zero to resolve_context: that function's step 1 walks scope edges only from the seed, and agent_profile nodes have 97 outbound requires, 4 specializes_from, and zero outbound scope. The profile channel is therefore not expressible as a seed set for resolve_context — it is a separate traversal over requires / specializes_from, which is why it is modelled as its own channel rather than folded into the walk.

This does not make profiles irrelevant — the opposite. Profiles are loaded by the implement loop on every work package, so they are a first-class entry vector. What the measurement shows is that the profile channel resolves artefacts it cannot render: _render_profile_directives and _render_profile_tactics are the only profile render paths, so a profile-resolved procedure, styleguide, toolguide or asset reaches nobody. procedure:onboard-external-agent-to-pack — PR #3007's exemplar, reached from agent_profile:doctrine-daphne by a requires edge — is a live instance. FR-020 owns it.

Two further corrections: the d=1↔d=2 spread is 7 nodes, not 49 (which withdraws Complexity row 3's original justification), and 25 of the 184 activated identifiers are not graph nodes at all, so roughly a quarter of any "unreachable" count is C-009 normalization that must not be banked as progress. SC-005 now partitions the pinned sets accordingly.

Binding-method note (unchanged and now proven necessary). The assertion MUST call doctrine.drg.query.resolve_context for the action channel and a named walk_edges for the profile channel. Every hand-rolled BFS in this mission's history has produced a wrong number, including the one in this plan's first revision.

Recorded performance acceptance (NFR-007)

Measured on this branch:

build_charter_context (compact):   982 ms   <- today's steady state
_load_action_doctrine_bundle:      961 ms   <- what FR-010 moves onto that path
  filter_graph_by_activation:      517 ms   <- 552 ruamel YAML loads per call, uncached
  resolve_context d=1 / d=2:         0.2 ms

FR-010 takes the steady-state render from ≈982 ms to ≈1.94 s, per action, per agent load, in every consumer project. Operator ruling 2026-07-28: accepted, not gated. No caching obligation is imposed. The figures are recorded here so a later regression is distinguishable from this one, and so the memoization option (_resolve_activated_urns_by_kind per PackContext; the merged+filtered graph per (repo_root, org_root, pack_context)) is on the record as available rather than unknown.

Consequence of the asset-delivery choice. asset:common-docs-structural-lint has four inbound requires edges and all four sources are unreachable — the common-docs cluster is a strongly-connected island no action scopes. Building bundle delivery for assets without wiring that cluster ships the delivery path while the only shipped asset still fails to arrive. The cluster is therefore folded into FR-015's enumerated table, subject to C-007's source-reachability precondition.

Charter Check

GATE: passed before Phase 0. Re-checked after Phase 1 — see below.

Charter gate	Status
ATDD-first discipline (binding, C-011 of the charter)	Satisfied by design — C-006 binds nine requirements to red-first with executable mutation proofs
Canonical sources, never improvise	Satisfied — C-010 mandates extending the shipped mechanisms; FR-016 mandates calling resolve_context rather than reimplementing it
Single canonical authority	Directly served — FR-017 collapses the activation stores to one
No direct pushes to origin/main	Satisfied — lane work consolidates into the reconciliation branch; a PR follows
Complexity ceiling 15 / no suppressions	NFR-002; context.py is 3,227 lines and several concerns land in it — see Complexity Tracking
Terminology canon (Mission, not Feature)	No new user-facing vocabulary; the terminology guard runs before push
Tiered rigour + tracer files	Nine concerns carry explicit sequencing; mission tracer maintained per standing orders

Post-Phase-1 re-check: no new violations. The one pre-existing tension (module size) is recorded in Complexity Tracking with a bounded response rather than a refactor this mission does not own.

Project Structure

Documentation (this mission)

kitty-specs/doctrine-delivery-reachability-01KYMXD6/
├── plan.md              # This file
├── spec.md              # Revised post-squad
├── research.md          # Phase 0 output
├── data-model.md        # Phase 1 output
├── quickstart.md        # Phase 1 output
├── contracts/           # Phase 1 output
├── checklists/          # requirements.md (run 2, post-squad)
├── research/            # Discovery + squad artefacts (5 documents)
└── tasks.md             # Phase 2 — NOT created by /spec-kitty.plan

Source Code (repository root)

src/
├── doctrine/                          # lowest layer — hosts shared vocabulary
│   ├── artifact_kinds.py              # kind enumeration + project-tier dir mapping (hoisted here)
│   ├── assets/
│   │   ├── models.py                  # AssetManifest (exists)
│   │   └── repository.py              # NEW — tiered resolution, rglob overlay, resolve_path
│   ├── service.py                     # + assets property; _PROJECT_KIND_DIRS retired to artifact_kinds
│   └── drg/
│       ├── models.py                  # + DRGGraph model_config
│       ├── merge.py                   # org-tier edge bridge joins the registry
│       ├── query.py                   # resolve_context — the FR-016 authority (read-only here)
│       └── migration/extractor.py     # derived helper + the writer registry it anchors
├── charter/
│   ├── drg.py                         # facade so specify_cli reaches the derived helper (C-001)
│   ├── pack_context.py                # activation authority; absence semantics
│   ├── resolver.py                    # the four hardcoded [] literals
│   ├── compact.py                     # steady-state rail — must carry all kinds
│   ├── context.py                     # bundle, slot table, render, reference block
│   └── synthesizer/project_drg.py     # writer 3 — joins the registry
├── specify_cli/
│   ├── migration/rewrite_opposed_by.py    # writer 2 — joins the registry
│   ├── cli/commands/doctrine.py           # asset subcommands + scaffold kind map
│   ├── cli/commands/agent/workflow.py     # FR-012 grain
│   ├── cli/commands/agent/workflow_executor.py
│   └── upgrade/migrations/                # NEW — activation authority + absence normalization
└── runtime/next/prompt_builder.py     # exception policy only (grain already correct)

tests/
├── doctrine/drg/                      # registry field-completeness, kind totality, reachability
├── charter/                           # bundle delivery, render, compact rail, reference block
├── docs/                              # asset consumer repointed off _REPO_ROOT
└── architectural/                     # named gate files only

Structure Decision: Single project, existing four-layer structure preserved. The only new module is src/doctrine/assets/repository.py; the only new facade is src/charter/drg.py. Two mappings are hoisted downward to src/doctrine/artifact_kinds.py — the module whose docstring already declares that no second kind enumeration may exist — and imported down by charter and specify_cli, which is C-001-legal in all three directions.

Complexity Tracking

Violation	Why Needed	Simpler Alternative Rejected Because
Four concerns land in src/charter/context.py (3,227 lines)	The bundle, the slot table, the render and the reference block all live there; splitting the module is a different mission with its own blast radius	Extracting a resolution surface first would double this mission's diff and put a refactor on the critical path of B1's unblock. Bounded response: no concern may grow context.py net — each extracts the logic it touches into a named helper, so the module shrinks or holds flat
The writer registry is hosted in specify_cli, the top layer, rather than beside the derived helper in doctrine	A registry naming charter and specify_cli members cannot live in doctrine — it reds test_layer_rules.py:282 and :293, two of this mission's own named gates. charter fares no better (:311). Only the top layer can statically hold all members	Import-time self-registration is the obvious alternative and is worse: membership becomes import-order-dependent, which re-opens the "a writer that never joins is invisible" gap the contract already concedes. Tests are not layered, and the precedent exists — tests/doctrine/.../test_model_strictness_roundtrip.py:542 already imports specify_cli.migration
Four named sets rather than one: two channels × two depths on the action side	Reach without delivery is this mission's defect class (C-008), so each channel must be measured where it delivers. Compact is the steady state and the stricter action-channel depth	A single unioned set was the original plan and is unsound — three independent lenses showed it certifies artefacts that no channel renders. The d=1↔d=2 spread turned out to be 7 nodes rather than 49, so the two depths are cheap; the two channels are what carry the cost, and they are not optional

Implementation Concern Map

> Note: Implementation concerns are NOT work packages and are NOT executable units. > /spec-kitty.tasks translates these into executable WPs.

IC-01 — Writer registry and derived serialization

edge/node-serializing site joins, with the field-completeness test iterating the registry rather than a hand-written list.

src/doctrine/drg/models.py (DRGGraph.model_config), src/doctrine/drg/merge.py (_bridge_org_edge_to_drg_edge), src/charter/synthesizer/project_drg.py, src/specify_cli/migration/rewrite_opposed_by.py, src/charter/drg.py (new facade), tests/doctrine/drg/test_model_strictness_roundtrip.py, tests/charter/synthesizer/test_project_drg.py

(C-005)

single edge_to_mapping Protocol: project_drg._serialize_graph is (DRGGraph) -> str with the dicts built inline (extracting them is a prerequisite task, not a registry join); _dump_graph_document is (DRGGraph, Path) -> None and owns document-level keys; and the org bridge constructs a DRGEdge from a fragment edge rather than serializing one. Model as MappingWriter, DocumentWriter, ModelBridge.

charter and specify_cli members reds two of this mission's own gates.

already imports through it. No facade is created; the work is exporting the derived helper.

empty list*, and _model_to_dict drops the key — proven: a novel impacts: str|None = None or impacts: list[str] = [] field is dropped by both derived writers, while is_symmetric: bool = False survives. Since B1's impacts is plausibly list-shaped, W-1 must assert over a fully-populated instance and carry a separate obligation for empty values, or the flagship gate is vacuous for the field it exists to protect.

document with an unknown top-level key from silently-accepted to a hard load failure on upgrade. NFR-001 bounds file mutation and is silent on pack rejection. Either give it a typed error + migration, or move it off the lane that lands first and alone.

needs membership, not a new test. Per C-005, FR-002's tests must cover the merged-graph read path, because the project-tier graph is merged into the graph every agent's context traverses once IC-06 lands.

attribute injection and extra kwargs are both rejected, and DRGGraph preserves the subclass all the way to the writers. Name this in the contract so three WPs do not each rediscover it.

• Purpose: Close the graph-serialization defect class by construction — a registry every
• Relevant requirements: FR-001, FR-002; SC-004, SC-009; C-006, C-010
• Affected surfaces: src/doctrine/drg/migration/extractor.py (derived helper + registry anchor),
• Sequencing/depends-on: none — this is the B1 unblock and must carry no inbound dependency
• Risks (revised post-squad):
• The registry needs three shapes, not one. Three of the five named members cannot satisfy a
• Host the registry in specify_cli. See Complexity Tracking — a doctrine-layer tuple naming
• charter/drg.py already exists (532 lines, 24-entry __all__) and rewrite_opposed_by.py:97
• The derived helper is not yet total. _render_for_yaml returns None for None *and for an
• extra="forbid" on DRGGraph is an unowned consumer break. It converts any org-pack graph
• project_drg is the unguarded writer and leads; rewrite_opposed_by is already guarded and
• The mutation fixture works via subclassing — models are not frozen (only extra="forbid"),

IC-02 — Kind mapping totality

hosting it once in artifact_kinds.py instead of four hand-maintained copies.

src/charter/kind_vocabulary.py:79, src/charter/pack_manager.py:136, src/specify_cli/cli/commands/doctrine.py:442, tests/doctrine/drg/test_kind_mapping_totality.py

charter/kind_vocabulary.py:79 and charter/pack_manager.py:136 are dict[ArtifactKind, str] and are explicitly exempted. But doctrine/service.py:19 is dict[str, str] keyed by plural strings and cli/commands/doctrine.py:442 is dict[str, str] keyed by singular — the totality guard's AST scan matches only EnumName.MEMBER attribute keys, so it never discovers them at all. Converting those two call sites to ArtifactKind keys is part of the definition of done, or the hoist leaves the guard blind at precisely the two sites #3038 names.

against the enum and can never detect that two copies disagree — which is the actual defect (I-A4). A work package could satisfy "extend the totality guard" by adding asset to all four copies and never hoisting, and go green.

against _CANONICAL_KIND_SINGULAR_TO_PLURAL (:431, 8 of 12 kinds), and then needs a stub from _stub_template (:464) — an eight-arm if-chain over kind strings ending in raise ValueError. :442 is consulted only at :626, after both gates. So IC-02 and IC-03 can land fully green and doctrine new --kind asset foo still exits 2. The if-chain is a kind projection the totality guard cannot see because it is not a dict. _CANONICAL_KIND_SINGULAR_TO_PLURAL also duplicates ArtifactKind._PLURALS eleven lines above the dict this concern edits.

the hoist must declare the canonical form and the adapter at each call site.

• Purpose: Extend the totality-or-explicit-exemption guard over the project-tier kind mapping,
• Relevant requirements: FR-006, FR-007; C-001, C-010
• Affected surfaces: src/doctrine/artifact_kinds.py, src/doctrine/service.py:19,
• Sequencing/depends-on: none
• Risks (revised post-squad):
• The four copies are not four of a kind, and two are worse than exempted — they are invisible.
• The hoist is the obligation; the guard is the ratchet. The guard checks each dict independently
• Scaffold parity targets the wrong surface. doctrine new --kind asset is rejected at :606
• The three copies use three different key/value conventions and three different .get fallbacks;

IC-03 — Asset resolution and operator surface

src/doctrine/service.py, src/specify_cli/cli/commands/doctrine.py, tests/docs/test_docs_structural_lint.py

paths anchor at its parent, while org/project dirs anchor at themselves; getting it wrong yields .../assets/built-in/built-in/.... BaseDoctrineRepository._project_scan is non-recursive, so an org-pack manifest at the ADR-mandated nested path is never found — override to rglob. Containment must call doctrine.drg.org_pack_config.resolve_relative_path_within_root directly, because the validator's helper lives in specify_cli and C-001 forbids the import. SC-003 must run against a built wheel in a clean environment, since in-repo the dev-layout fallback cannot fail — and that harness has no home in this plan. Name the test path and the CI job before tasks; a criterion with no artefact is this mission's own defect class. The _project_scan override must return list[Path], not Iterable[Path] — the base declares list[Path] and widening a return type in an override is a Liskov violation mypy will flag, which NFR-002 forbids silencing.

• Purpose: Make a shipped asset addressable by identifier from a consumer installation.
• Relevant requirements: FR-003, FR-004, FR-005, FR-008; SC-003; C-002
• Affected surfaces: src/doctrine/assets/repository.py (new), src/doctrine/assets/__init__.py,
• Sequencing/depends-on: IC-02 (the project-tier directory for assets is part of the hoist)
• Risks: Anchor asymmetry — _built_in_dir("assets") returns <root>/assets/built-in and blob

IC-04 — Reachability as a named, pinned property

named depths, seeded from actions and activated agent profiles.

(_INTENTIONAL_ORPHANS, _ACTIVATED_BUT_UNREACHABLE, _orphan_urns), a reachability helper, _CURATED_ARTIFACT_EDGES or authored *.graph.yaml fragments

regenerates up to 14 src/doctrine/*.graph.yaml fragments through _dump_graph_document, the exact function IC-01 rewrites, and test_shipped_graph_is_fresh_and_byte_identical pins them. Landing IC-04 first regenerates them twice, with a changed serializer in between. Ledger both regeneration events under NFR-004.

incidence set to a named reachability set, (b) build the per-channel assertions, (c) author FR-015's edges. (c) is a production-data change with a 14-file regeneration side effect; (a) and (b) are gate changes with different red-first shapes.

UNREACHABLE currently measures edge incidence despite its name — landing a genuine reachability set beside it is a readability trap; rename in the same WP. FR-015's table must show each proposed source's own reachability (C-007b) or "obvious" collapses into implementer taste. C-009 forbids banking the id-normalization swing (~20 artefacts) as progress. The common-docs cluster is in scope here because IC-06 delivers assets and the only shipped asset sits behind it. Edges land in _CURATED_ARTIFACT_EDGES`, which mission B2 retires — the destination and handoff must be recorded.

• Purpose: Replace incidence counting with named-set assertions under a named traversal at two
• Relevant requirements: FR-015, FR-016; SC-005; C-007, C-008, C-009
• Affected surfaces: tests/doctrine/drg/migration/test_extractor_projection.py
• Sequencing/depends-on: IC-01 — undeclared in the first revision. Authoring FR-015's edges
• Expected split: this is three concerns and will become three work packages — (a) rename the
• Risks: The assertion must call resolve_context, not reimplement the walk. `_ACTIVATED_BUT_

IC-05 — Activation authority and absence semantics

_load_charter_activation_source), .kittify/config.yaml surface, a new upgrade migration, tests/doctrine/drg/migration/test_extractor_projection.py::_charter_activated_urns

NFR-006). This is an authority invariant — the surface that raises owns the guarantee that the error surfaces — so splitting raise and catch across two concerns left nobody owning the end-to-end assertion. IC-05 carries it, including prompt_builder's except Exception: pass, with an end-to-end test that an activation failure reaches an operator.

be repointed before the mirror is deleted" — but that is IC-05's own work, not a dependency. The real coupling is a file collision: IC-04 renames _ACTIVATED_BUT_UNREACHABLE while IC-05 rewrites _charter_activated_urns, both in the same 604-line test file, with test_activated_but_unreachable_orphans_are_really_activated consuming both. Assign to one lane or serialize — they are otherwise parallel, which shortens the critical path.

gate ∩ reachable. Identifier normalization (I-V4 / C-009) moves the set by ~25 artefacts, so extract normalization as its own slice landing before IC-04's pin, and give IC-04 an explicit re-pin obligation if IC-05 moves the set.

charter_file:, which exists nowhere. Minting the second name reintroduces the defect FR-017 removes. _charter_activated_urns reads the dead mirror; deleting it first makes that gate's floor assertion fail and its stray guard vacuously true. SC-007 requires a divergent-mirror fixture** — the only case that proves which store won. The absence migration mutates consumer configs, which is why NFR-001 was amended.

survived both. m_unify_charter_activation.py made config.activated_<kind> the single authority; m_unify_charter_activation_finalize.py folded it into charter.yaml and minted the charter: pointer — yet .kittify/config.yaml on this checkout still carries the activated_* keys alongside the pointer. FR-017 is the third pass. IC-05's definition of done must include reading both, stating why the finalize migration did not take, and declaring the ordering constraint its docstring already carries.

• Purpose: One activation store; absence means empty, not "all built-ins".
• Relevant requirements: FR-017, FR-018; SC-007; NFR-001, NFR-006
• Affected surfaces: src/charter/pack_context.py (_read_activated_*,
• Also owns (moved from the dissolved IC-07): the fail-closed policy (FR-012's error half,
• Sequencing/depends-on: none. The first revision claimed IC-04, justified by "the gate must
• Ordering caveat that is real: IC-05 owns activated, and the delivery predicate is
• Risks: **The pointer field is charter:, which already ships and is honoured — not
• Two prior migrations already fought over this surface, in opposite directions, and the mirror

IC-06 — Delivery rail carries every kind, to the render

the rendered context, on every load.

_ACTION_BUNDLE_SLOT_BY_KIND, _render_bootstrap_text, the compact returns), src/charter/compact.py, src/charter/resolver.py

workflow_executor.py:459. It selects the bundle IC-06 delivers and is inert without it, exactly as resolver.py's [] literals are. Keeping it downstream would let IC-06 pass SC-001 through a test that supplies the grain itself while the shipped CLI still emits an empty bundle on every real invocation — _load_action_doctrine_bundle degrades to empty when resolved_type is None. US3 scenario 9 exists to close that hole.

change) and FR-010 (delivery on every load; a control-flow change, and the mission's only NFR-007 exposure).

bundle is computed. Cost is measured and accepted under NFR-007.

BOOTSTRAP_ACTIONS, returning before state is prepared), :237 (depth below minimum), and :3201 / :3207 in the --json payload builder, which has its own mode handling and its own bundle call at :3213. The --json path is what quickstart.md uses to observe the defect and where SC-001/SC-002 will be measured — it was absent from the first revision's surfaces.

rendering (:1149-1157), and resolve_context's docstring confirms depth "also controls extended artifact inclusion". At the pinned depths d=1 and d=2, I-B1 is structurally unachievable for exactly the two kinds this mission calls the sixth delivery defect. IC-06 owns retiring or repointing that gate, and splitting the overloaded parameter (a suggests hop cap and a render verbosity tier are two concepts on one integer).

surface"; _render_compact_governance (:2788) accepts only directive_ids/tactic_ids. Widening the steady-state rail is a signature change and a documented-contract widening, not only a payload change. Assertions must read the rendered compact text, not the bundle.

dict[str, list[str]] and destroys it at the return boundary (:748) into a positional 4-tuple. There are already five parallel per-kind projections in this path with five different subsets, and every subset difference is a live delivery defect. Return the mapping — ids_by_slot: Mapping[str, tuple[str, ...]] with the slot set derived from _ACTION_BUNDLE_SLOT_BY_KIND — so totality and delivery become one statement and the compact rail widens for free. If this is deferred, record the deferral: a sixth projection is the tax the next mission pays.

action_bundle_bucket raising on unmapped kinds. The work is flipping PROCEDURE and ASSET from None to real slots, plus per-kind rationale — not building the guard. This reverses PR #3007 WP03's recorded verdict ("state them, not start rendering them") for two of twelve exclusions; record the reversal and the criterion, or the next mission flips two more by the same unstated reasoning.

construction, so an equality stated as activated ∩ reachable makes asset_ids = [] the conforming implementation. gate(kind) must be a column of the same NodeKind-keyed table — activated(kind) for activatable kinds, ALL for delivered-but-ungated ones. TEMPLATE's None then acquires a stated reason instead of being ASSET's untreated twin.

fifth activation reader. It also still lacks an asset field while the bundle gains one — a narrower projection is this mission's defect class.

• Purpose: Make resolution and delivery the same thing — every kind the bundle resolves appears in
• Relevant requirements: FR-009, FR-010, FR-011; SC-001, SC-002; NFR-003, NFR-006
• Affected surfaces: src/charter/context.py (_ActionDoctrineBundle,
• Also owns (moved from the dissolved IC-07): the mission-type grain at workflow.py:738 and
• Expected split: two work packages — FR-009 (render carries every resolved kind; a payload
• Sequencing/depends-on: IC-04, IC-05
• Risks:
• FR-010 is a control-flow change, not a payload change — the compact returns fire before the
• "The compact returns" is four sites in two functions, not one: context.py:208 (action not in
• depth means two things. _EXTENDED_CONTEXT_DEPTH = 3 gates styleguide and toolguide
• CompactView carries two kinds (compact.py:38) and its docstring calls them "the contract
• Do not create a sixth per-kind projection. _classify_artifact_urns (:729) already builds
• The slot table is already total and already guarded, with a verdict comment per group and
• Express the asset carve-out as a total gate function. activated(asset) is ∅ by
• Populating GovernanceResolution from anywhere but the canonical PackContext path creates a

IC-07 — Profile channel delivery

> Replaces the first revision's "Caller grain and error propagation", which was a leftovers bin: > two call sites and one except clause sharing only the word "caller", splitting an authority > invariant across the surface that raises and the surface that catches. Grain moved to IC-06, > fail-closed policy to IC-05, and the concern number was reused for the defect the squad surfaced.

implement loop on every work package, so they are a first-class entry vector — but only profile-cited directives and tactics render today.

_render_profile_directives / _render_profile_tactics (:2757-2758), the profile resolution path in AgentProfileRepository, and the profile-channel reachability helper

surface, same file, so treat as a file collision rather than an ordering dependency

have zero outbound scope edges, so resolve_context from a profile returns 0 artefacts at every depth. Its traversal is a separately-named walk_edges over {requires, specializes_from}. Any attempt to fold it into resolve_context will silently measure nothing.

agent_profile:doctrine-daphne by a requires edge — is the live instance of this defect. It is in scope here, and it is why the first revision's plan to "rescue" it via seed-union was unsound: the union asserted reach that no channel delivers.

configuration**. Measuring it as unconditional repeats the fail-open shape FR-018 retires.

Where the answer is not attested by the profile schema, defer under C-007 rather than inventing it.

• Purpose: Make the profile channel deliver every kind it resolves. Profiles are loaded by the
• Relevant requirements: FR-020, FR-016 (profile-channel set); SC-001, SC-005; C-008
• Affected surfaces: src/charter/context.py:2745-2764 (_render_profile_sections),
• Sequencing/depends-on: IC-04 (the channel's named set); parallel with IC-06 — different render
• Risks:
• The profile channel is not expressible as a resolve_context seed set — agent_profile nodes
• procedure:onboard-external-agent-to-pack — PR #3007's exemplar, reached from
• profile is str | None at context.py:137, so the channel is **conditional on caller
• Deciding which kinds a profile should deliver is a doctrine question, not only a render one.

IC-08 — Reference block distribution and resolvable pointers

a fixed kind order.

_filter_references_for_action:1484, _LIBRARY path resolution

references is sourced from the charter.yaml catalog and passed in, not derived from the bundle (context.py:1166-1172). Retain a file-collision note only: both edit context.py under the no-net-growth constraint.

nowhere in src/ and only from tests/charter/test_context.py:815 — a render path reachable only from the test suite, i.e. an instance of this mission's own thesis. Ruling: delete it; if a WP finds a live caller, that is a finding, not a reason to keep it. _filter_references_for_action is a no-op for every doctrine kind (213 → 213), so "filtering" is not the fix — distribution is. No test pins either cap (proven by mutation), so SC-006 needs a non-vacuity floor or it passes over an empty set.

• Purpose: Every emitted pointer opens; composition varies by action instead of being exhausted by
• Relevant requirements: FR-013, FR-014; SC-006
• Affected surfaces: src/charter/context.py:1169 and :1531,
• Sequencing/depends-on: none — parallel with IC-06. The first revision declared IC-06, but
• Risks: Two cap sites, not one. :1531 sits in _render_bootstrap, which is called from

IC-09 — Documentation and CLI reference

docs/doctrine/create-a-doctrine-artifact.md, docs/api/cli-commands.md, CHANGELOG.md

IC-02 changes) and IC-06 (D4 requires the asset bundle-slot verdict be documented where the table records verdicts). The first revision named IC-05, which owns no surface this concern documents.

create-a-doctrine-artifact.md contains "asset" zero times while review-gates.md cites it as the asset how-to. New visible Typer paths from IC-03 trip REF-MISSING in the CLI-reference freshness gate against a 4,950-line reference. Run the terminology guard before push — it lives in a CI job the fast suites do not cover. Cite the sentence, not the line — this plan's own quickstart records that four discovery anchors drifted.

• Purpose: Make the documented remedy followable.
• Relevant requirements: FR-019; SC-008
• Affected surfaces: docs/doctrine/doctrine-kinds.md:50-52,
• Sequencing/depends-on: IC-03, IC-02 (doctrine-kinds.md is the kind-vocabulary reference
• Risks: doctrine-kinds.md states asset has "no built-in artifacts yet" — false, one ships.

---

Issue ledger

Added post-squad. The first revision cited ten issues and named none as closed, which reads as closure to a reviewer.

Disposition	Issues
Closes	#2977 (rewrite_opposed_by still declares its own _node_to_dict/_edge_to_dict; IC-01 removes them) · #3037 (DoctrineService exposes 9 repositories, none assets; IC-03 adds the resolution path)
Advances, does NOT close	#3038 — IC-02 fixes the _PROJECT_KIND_DIRS half; project_drg._KIND_TO_NODE_KIND stays deferred · #2981 — this mission touches only the project-tier mapping; the headline _activation_render.py 8-vs-10 drift is out of scope under C-004 · #3009 — _ACTIVATED_BUT_UNREACHABLE is its nine artefacts; IC-04 renames and re-measures, but closure depends on FR-015's table covering all nine, which is not yet asserted
Cited, not touched	#2986, #2994, #3026, #3036, #3039 — no concern names _inert_slots.py or test_no_dead_doctrine_paths.py. Listed so a reviewer does not read a citation as a closure

Corrected size estimate

The first revision's 45-60 files is ~30% light. Realistic: 65-85, concentrated in two concerns.

Concern	Named surfaces	Realistic
IC-04	3	18-22 — authoring FR-015's edges regenerates up to 14 *.graph.yaml fragments, plus the helper, the renamed/partitioned sets, NFR-004 ledger rows, and the common-docs sources
IC-03	5	12-16 — new repository, __init__, service property, the rglob override, containment call, two CLI subcommands, the test repoint, new unit tests, and SC-003's wheel harness, which still has no named home

File ownership — declare the lane, or serialize

Four files are claimed by two or more concerns. This is how lane execution splits its brain.

File	Claimants	Resolution
src/charter/context.py	IC-06, IC-07, IC-08	Three concerns in a 3,227-line module under a shared no-net-growth constraint. Assign one lane or serialize
src/doctrine/service.py	IC-02 (_PROJECT_KIND_DIRS), IC-03 (assets property)	Land IC-02 wholly first, or give IC-03 only the asset rows
src/specify_cli/cli/commands/doctrine.py	IC-02 (:431, :442, :464), IC-03 (asset subcommands)	Same
tests/doctrine/drg/migration/test_extractor_projection.py	IC-04 (renames the set), IC-05 (repoints _charter_activated_urns)	One 604-line file, one lane

Deferrals still needing a home

Recorded rather than filed — filing is outward-facing and awaits operator confirmation.

Deferral	Current state
Agent-facing asset delivery	Stale row in spec.md — says "Filed" with no number, and D4 has since reversed it into scope. Delete the row
The four action:plan/* nodes resolving to zero artefacts of any kind	Verified real and exact at d=2. No issue number. SC-001 requires such gaps be named; without an issue it is invisible work
The AST gate on dict-literal edge payloads	No owner, no number — and it is the compensating control for the registry's one conceded blind spot. Deferring the compensating control with no home is this mission's defect class applied to its own governance