Spec Kitty

└─ kitty-specs
   └─ model discipline dispatch binding

Mission Run:

📚 Docs ↗

Research & Brownfield Checks — Model-Discipline Dispatch Binding

Consolidates pre-spec grounding (2× sonnet) + post-spec squad (alphonso+paula, opus) + post-planning brownfield. HEAD 71b2787e8.

1. Code-state (gaps confirmed)

Catalog dead: model_task_routing/models.py has no loader; whitelisted in test_no_dead_modules.py:251 as schema-gen only; no populated instance.
Dispatch seam: ProfileInvocationExecutor.invoke() (executor.py:168-284) returns InvocationPayload synchronously, never spawns a model (:150,181); no model slot (:105-116); no task_type concept (grep task_type invocation/ → empty).
Charter model_task_routing reference dangles (not an ArtifactKind; absent from generated references.yaml). autonomous-operation-protocol tactic exists (graph.yaml:367) but lacks an inbound directive suggests edge.
File-scope correction (grounding): the issue's delivery/dispatcher.py is the SaaS drain, NOT the seam — real target is invocation/{executor,router}.py + cli/commands/dispatch.py.

PRIMARY #2364; PARENT #1799 (OPEN, confirmed); #2196/#2216 ruled out.
#1049 (OPEN) — static per-step model config (agents.models.steps.*); complementary boundary, scoped against (C-003), not folded. Stays a separate track.
#1841 (OPEN) — same "rule instructed not structurally bound" pattern (profile-load); cross-referenced in design, not folded.
#240 (CLOSED) — the 2.x model-discipline research ancestor (archived docs/archive/2x/model-discipline-routing.md).
#1438 (CLOSED, PR #1545) — schema/Pydantic parity fix; the schema this mission consumes. No consumer wiring then.
Origin: mission 057 bootstrap (623057f97) shipped the schema; consumer never scoped.

3. Brownfield: dead-symbol / duplication / deprecation

Dead-symbol ratchet: model_task_routing.models currently allowlisted; FR-001 removes it as the loader lands (the wiring is the fix). Watch the ratchet.
Duplication: none introduced — one loader, one evaluator, one task-class map (new canonical seams).
Deprecation: no deprecated surfaces touched. invocation/* current. NOTE the SEPARATE pre-existing base failure test_no_dead_modules.py::test_no_new_dead_modules_under_src (stale allowlist for migration m_3_2_3_encoding_provenance_gitignore_backfill, drifted via unshim #2292) — red on clean upstream/main, unrelated to this mission; flagged for a standalone campsite fix, NOT folded here.

4. Brownfield: generated-artifact / bundle regen risk

agent-profile.schema.yaml is GENERATED from schema_models.py — regenerate, never hand-edit (C-005).
references.yaml + _LIBRARY/synthesis-manifest are generated from the DRG — FR-006/007 add suggests edges then regenerate the bundle (a generated-lockfile step; reds CI if skipped, like the docs page-inventory pattern).

5. Campsite (#1931) — none

No domain-matched hygiene items; touched files carry no live TODO/FIXME/skip/xfail (grounding). The dead-module base failure (§3) is tracked separately, not a #1931 fold.

6. Design decisions resolved (spec rev 2)

Decision	Resolution
Advisory vs enforced	Advisory-only — the seam never spawns a model (C-001)
Capability vs cost/latency tier	`objective: quality_first` + `weights.quality` = the capability lever (no schema change)
Profile field placement	`profile.py` domain model (kebab alias) + `schema_models.py` regen — not schema-only
Reference resolution	DRG `suggests` edge + token repoint (not manual references.yaml)
`autonomous-operation-protocol`	One-line suggests edge to the existing tactic (FR-007)
Scope	Full evaluator (operator ruling) — loader + map + scorer + payload + field + doctrine chain
#1049	Complementary static-config track; out of scope