Files
Assistant/dev/tasks/DEV-000072-semantic-interpretation-deferral-boundary.json
m ce666bbc43
CI / test (push) Successful in 18s
CI / test (pull_request) Successful in 17s
DEV-000072 semantic interpretation deferral boundary
2026-08-17 23:02:00 +03:00

147 lines
5.5 KiB
JSON

{
"task_id": "DEV-000072",
"task_revision": 1,
"architecture_revision": "DEV-000071-baseline",
"base_repo_revision": "64870798b9a50f895c99d24441703f366df5c98a",
"risk_tier": "E",
"intent": "Implement semantic interpretation and deferral boundary V1 so deterministic interpretation acts as a fast-path candidate rather than unquestionable final truth; preserve provenance between deterministic and cognitive candidates; support mixed intent and safer deferral without activating new providers.",
"requirement_ids": [
"DEV-000071 measured baseline",
"Deterministic candidate can decline/defer",
"Mixed intent representation",
"Cognition may challenge uncertain deterministic candidate",
"No provider hardcoding",
"No Ollama activation",
"Known and holdout evaluation comparison",
"Production chat acceptance"
],
"allowed_change_scope": {
"modules": [
"interaction",
"work",
"capability",
"epistemics"
],
"paths": [
"contracts/interaction/**",
"contracts/work/**",
"contracts/cognitive/**",
"modules/interaction/**",
"modules/work/**",
"adapters/postgres/interaction_repository.go",
"cmd/core-runtime/**",
"cmd/creator-web/**",
"cmd/cognitive-worker/**",
"database/migrations/**",
"tests/**",
"docs/PROJECT_STATUS.md",
"docs/evidence/**",
"dev/ACTIVE_TASK.json",
"dev/tasks/DEV-000072-semantic-interpretation-deferral-boundary.json"
]
},
"forbidden_change_scope": {
"paths": [
"architecture/ARCHITECTURE_CONSTITUTION.md",
"architecture/architecture-registry.yaml",
"architecture/architecture-registry.json",
"architecture/adrs/**",
"security/**",
"modules/authority/**",
"modules/commit/**",
"deploy/**",
"ci/**",
".gitea/**"
]
},
"allowed_architecture_delta": {
"new_cross_module_dependencies": 0,
"new_canonical_owners": 0,
"new_privileges": 0,
"new_third_party_dependencies": 0,
"new_provider_adapters": 0,
"new_database_owner_roles": 0
},
"contracts": {
"consume": [
"contracts.identity",
"contracts.interaction",
"contracts.work",
"contracts.capability",
"contracts.cognitive",
"contracts.epistemics",
"contracts.situation"
],
"may_change": [
"contracts.interaction",
"contracts.work",
"contracts.cognitive"
]
},
"semantic_owner": "interaction",
"canonical_state": [
"Interaction-owned human utterance interpretation semantics",
"Interaction-owned interpretation provenance between deterministic and cognitive candidates",
"Owner-local persistence expansion for semantic candidate provenance",
"Existing Work planning/help-path semantics updated to consume richer interaction interpretation without changing canonical ownership"
],
"state_transitions": [
"human utterance -> deterministic semantic candidate",
"candidate accepted or deferred",
"deferred candidate -> cognitive candidate",
"candidate selection -> persisted selected interpretation + provenance",
"mixed/consequential ambiguity -> clarification or bounded safe response"
],
"side_effects_and_privileges": [
"No direct external mutation",
"Cognitive Gateway only through existing provider-neutral runtime",
"No direct CPU endpoint calls from chat path",
"No provider activation changes"
],
"persistence_and_migrations": [
"Interaction owner-local migration expected",
"No cross-owner SQL writes",
"No cross-owner FK introduction",
"No new DB owner roles"
],
"failure_retry_restart": [
"If cognition is unavailable, obvious deterministic paths continue to work",
"Uncertain consequential intent must fail closed via clarification or explicit limitation",
"Model output must remain typed candidate evidence, never direct canonical fact/action"
],
"assumptions": [
"Current production has only one proven eligible cognitive binding for human chat: CPU CANARY",
"Ollama/GPU remains inactive for routing in this milestone",
"DEV-000071 artifacts remain the baseline comparison set"
],
"acceptance_scenarios": [
"ask-model request no longer finalized as assertion by default",
"think/investigate wording can defer to cognition without lexical patching as the sole strategy",
"mixed investigate+modify retains both semantic needs",
"correction wording plus new directive preserves both aspects",
"uncertain consequential request defers safely",
"suite_v1 plus DEV-000071 development/holdout comparisons are reported honestly",
"production /chat acceptance covers greeting, cognition question, named implementation request, mixed work, correction, correction+new directive, consequential ambiguity"
],
"required_tests": [
"go test ./...",
"go vet ./...",
"go run ./cmd/devctl preflight -active dev/ACTIVE_TASK.json",
"go run ./cmd/devctl check-architecture",
"go run ./cmd/devctl check-diff -active dev/ACTIVE_TASK.json",
"tests/cognitive_advisory/cognitive_worker_env_policy_test.ps1",
"tests/owner_acl/verify.ps1 -DockerContext docker-prod -PostgresContainer assistant-postgres-1",
"interaction domain tests",
"work planning tests",
"reality evaluation comparisons",
"production /chat acceptance"
],
"terminal_gap_states": [
"BLOCKED_BY_SPEC_GAP",
"BLOCKED_BY_ARCHITECTURE_GAP",
"BLOCKED_BY_MISSING_CAPABILITY",
"BLOCKED_BY_CONFLICT",
"REQUIRES_MIGRATION_DECISION"
]
}