Spec Kitty

Research: Dead-Port Disposition

Mission: dead-port-disposition-01M1VRA2 · Date: 2026-09-06 Purpose: resolve every Technical Context unknown and every domain rule before design. All findings were re-derived from source at main (3a4f92b41), not taken from the ADR text.

R-1. Is the concrete class load-bearing anywhere?

• Decision: No. It is safe to delete once the two capabilities the bridge uses are promoted.
• Rationale: Production importers are exactly runtime_bridge.py:195 (binding) and runtime_bridge_engine.py:80 (TYPE_CHECKING only). The bridge uses two members beyond the eight emit_* methods: for_feature (:1552, :2739) and seed_from_snapshot (:1614, :2745, plus runtime_bridge_engine.py:344). The resolved mission_id is stored but never read (event_emitter.py:40,51). Tests: one live import (tests/specify_cli/events/test_decision_log_coord.py:17), 14 patch sites on the bridge-module attribute, two reservation comments.
• Alternatives considered: keep the concrete class renamed NullRuntimeEventEmitter (ADR transitional form) — rejected; the two-class state the ADR names as target has one name, and no transition is needed because all callers are in-repo.

R-2. Should for_mission / seed_from_snapshot be added to the Protocol?

• Decision: No. Add them to NullEmitter (and keep them on _BufferingRuntimeEmitter, which already has seed_from_snapshot); leave the Protocol at the eight emit_* methods.
• Rationale: The engine types its emitter parameter against the Protocol (_internal_runtime/engine.py:191,251,513,775) and only calls emit_. Seeding and construction are bridge-side concerns on the object the factory returned. Widening the Protocol would force every structural adapter — including DecisionGitLog, which today implements only the emit surface — to implement seeding, and would make the MagicMock(spec=Protocol) fixtures in test_decision_log_coord.py grow methods they never exercise. A future producer that needs seeding implements it; the factory contract (contracts/emitter-seam.md) says the returned object may* expose seed_from_snapshot and the bridge calls it inside its existing try (runtime_bridge.py:1611-1616, :2744-2751).
• Consequence for fix two: DecisionGitLog must gain a seed_from_snapshot pass-through, because advance_run_state_after_composition seeds whatever emitter it receives (runtime_bridge_engine.py:344) and after the fix it receives the wrap. One delegating method; no behavior.
• Alternatives considered: widen the Protocol (rejected above); special-case the composition path to seed ctx.sync_emitter and emit via ctx.emitter_for_engine (rejected — two emitter references on one path is the defect class being closed).

R-3. What exactly is the flush-target defect, and what is the right red-first fixture?

• Decision: The defect is on strict-retrospective-policy decision_required advances (and, independently of policy, on composition dispatch). The red-first fixture is a decision_required advance, never a terminal one.
• Rationale: DecisionGitLog persists only emit_decision_input_requested / emit_decision_input_answered (decision_log.py:147-160); every other emit is pass-through. The engine's decision_required and terminal branches are mutually exclusive (runtime_bridge_engine.py:299-305; engine :448,476). Under strict retrospective policy the buffer is installed for every advance (runtime_bridge.py:2129-2150) and flushed into the plain seam (:2187), so a DecisionInputRequested raised on the gated path is dropped; a terminal advance buffers only MissionRunCompleted, which the log ignores. The composition path passes ctx.sync_emitter outright (:1976), so its _emit_decision_required (runtime_bridge_engine.py:226) bypasses the log regardless of policy.
• Strict policy shape for tests: _retrospective_blocks_completion is true iff enabled and timing == "before_completion" and failure_policy == "block" (runtime_bridge_retrospective.py:384-390). Tests monkeypatch _resolve_retrospective_policy_for_runtime (pattern already used at tests/runtime/test_bridge_retrospective.py:200-204).
• Ordering / double-emit: the terminal gate (:2178-2181) runs before the flush and discards on refusal; flush is one-shot (runtime_bridge_retrospective.py:138-149); on the gated path the engine wrote only into the buffer, so the flush is the first and only DecisionGitLog write. Rollback is preserved.
• Alternatives considered: flush into the plain seam and replay decision events into the wrap (rejected: double bookkeeping); make DecisionGitLog the engine emitter even under strict retrospective policy and drop the buffer (rejected: re-opens the unretractable MissionRunCompleted problem the buffer exists to solve).

R-4. Factory + registry: which idiom, and where does the env gate live?

• Decision: Module-level runtime_emitter_for_mission(...), register_runtime_emitter_factory(f), reset_runtime_emitter_factory() in _internal_runtime/events.py; the SPEC_KITTY_SYNC_MINIMAL_IMPORT check runs at call time inside the factory. (Decision Moment 01M1VSEWHJM4WAQD7EGN5PA4XJ.)
• Rationale: Mirrors the one proven registration idiom in the tree, status/adapters.py:201-214 (ensure_zeitgeist_moment_handlers / reset_handlers), which the ADR names as the pattern. Call-time env read lets tests toggle the gate with monkeypatch.setenv without reloading modules and costs one dict lookup. A bridge-module name to patch keeps the 14 rewrites mechanical (monkeypatch.setattr(rb, "runtime_emitter_for_mission", ...)) and the oracle's spy wrapping unchanged in shape (_bridge_oracle.py:471-481 wraps the real callable and returns a _RecordingProxy).
• Alternatives considered: registry-only (rejected: forces every test to go through the registry and the oracle to grow a registry adapter for no production benefit); class-shaped attribute for compatibility (rejected: violates ADR (a) and preserves the collision).

R-5. Layer rules: can _internal_runtime/events.py import specify_cli.mission_metadata?

• Decision: Yes.
• Rationale: The enforced direction is kernel <- charter <- {glossary, runtime, mission_runtime} <- specify_cli (CLAUDE.md §Modularity SSOT), i.e. runtime may import specify_cli but not specify_cli.cli / specify_cli.next (tests/architectural/test_layer_rules.py:323-340). _internal_runtime/planner.py:46 already imports specify_cli.mission_metadata.load_meta, and mission_metadata is on the runtime outbound ledger (test_layer_rules.py:132,192). Moving resolve_mission_identity (mission_metadata.py:234) into events.py adds no new ledger entry.

R-6. Supply chain (051-supply-chain-install-safety)

• Decision: Not applicable — no dependency is added, upgraded, or removed. spec_kitty_events 9.1.6 stays pinned and untouched; the six mission_next moment types it provisions are consumed as today.
• Adversarial evidence: no security-impacting dependency decision exists, so no challenge pass is required. Recorded here so silence is not mistaken for compliance.

R-7. Terminology

• Decision: Promote as for_mission; do not rename the feature_dir keyword parameter.
• Rationale: The Terminology Canon forbids minting new feature* aliases for the Mission domain object; the constructor is a new public surface. feature_dir is a pre-existing parameter name used across the bridge (:1553, :2740, resolve_mission_identity(feature_dir)) and the engine adapter; renaming it is a cross-module bulk edit outside this mission's boundary (DIRECTIVE_035). The terminology guard (tests/architectural/test_no_legacy_terminology.py) scans docs prose, not identifiers; NFR-006 is therefore enforced by the PR's added-lines grep, recorded in the PR body.

R-8. Existing tests that constrain the change

Test	Constraint it imposes
tests/next/test_internal_runtime_coverage.py:173-174	emitter.py re-export identity and exact __all__ — must be extended with the factory/registry names
tests/runtime/test_bridge_engine.py:408-427	advance_run_state_after_composition seeds the emitter it is given and emits step-issued on it — the wrap must accept seed_from_snapshot
tests/runtime/test_bridge_parity.py:1140-1166	answer-path sync sink records calls via the oracle spy — spy must wrap the factory's return
tests/runtime/test_bridge_retrospective.py:88-147	buffer records/flushes in order, discard drops, flush tolerates missing target methods — unchanged
tests/specify_cli/next/test_runtime_bridge_composition.py:55-68	composition tests substitute a NullEmitter subclass with seed_from_snapshot — becomes a factory patch
tests/status/test_producer_conformance.py, tests/contract/test_identity_contract_matrix.py	reserve the seam by comment only — comments updated, assertions untouched

Adversarial evidence

The governing ADR was itself challenged by a pre-merge adversarial squad (architect / debugger / reviewer) before acceptance; the MAJOR finding (terminal-advance framing of the bug) was changed in the ADR before this mission was specified. Plan-level contested findings: none raised; the seam design (R-4) was put to the operator as a three-option Decision Moment and accepted.