Files
Assistant/dev/tasks/DEV-000208-task-aware-model-evaluation.json
m feea909d69
CI / test (push) Successful in 57s
CI / test (pull_request) Successful in 1m0s
feat(capability): add task-aware model evaluation evidence
2026-08-23 11:57:47 +03:00

24 lines
3.6 KiB
JSON

{
"task_id": "DEV-000208",
"task_revision": 1,
"architecture_revision": "ADR-0018",
"base_repo_revision": "b78a026ee169047feb8468f0799fa1b9f4e27562",
"risk_tier": "E",
"intent": "Make Domovoy's model qualification and routing task-aware through parallel durable specialist work: typed evaluation profiles/evidence, durable task scores, and explainable selection without hardware/provider heuristics.",
"requirement_ids": ["task-specific model evaluation", "parallel specialist evaluation", "no main execution thread", "model fleet/failover", "hardware-neutral routing", "explainable model rating"],
"allowed_change_scope": {"modules": ["capability", "core-runtime"], "paths": ["contracts/capability/**", "modules/capability/**", "adapters/postgres/capability_repository.go", "cmd/core-runtime/main.go", "database/migrations/039_capability_task_evaluations.sql", "tests/model_qualification/**", "dev/ACTIVE_TASK.json", "dev/tasks/DEV-000208-task-aware-model-evaluation.json"]},
"forbidden_change_scope": {"paths": ["architecture/**", "security/**", "deploy/**", ".gitea/**", "contracts/authority/**", "modules/identity/**"]},
"allowed_architecture_delta": {"new_cross_module_dependencies": 0, "new_canonical_owners": 0, "new_privileges": 0, "new_third_party_dependencies": 0, "new_database_owner_roles": 1},
"contracts": {"consume": ["CognitiveRequirement", "ResolvedCognitiveBinding", "existing capability evidence", "existing capability owner repository", "Work Graph durable fan-out/fan-in"], "may_change": ["task profile/evaluation contracts", "task-aware selection request/response serialization"]},
"semantic_owner": "capability",
"canonical_state": ["capability task evaluation evidence keyed by task profile and binding revision"],
"state_transitions": ["GENERAL profile -> first bounded evaluation", "unresolved task evidence -> ADDITIONAL profile with parent and trigger evidence", "profile result UNKNOWN -> SUPPORTED/UNSUPPORTED; evidence is append-only by binding revision", "task-aware requirement -> eligible candidates with explainable task rating", "evaluation request -> durable fan-out -> specialist evidence -> fan-in aggregate"],
"side_effects_and_privileges": ["core capability database role writes only; no provider calls and no secret access in this task"],
"persistence_and_migrations": ["039 adds capability.cognitive_task_evaluations owned by capability"],
"failure_retry_restart": ["missing, stale or unknown task evidence excludes a task-aware candidate; no score defaults to success", "one specialist failure does not block unrelated tasks; stale leases are recoverable and fan-in preserves partial/unknown state"],
"assumptions": ["a later Work Graph evaluation runner will produce per-case evidence; this task establishes the durable contract and routing gate", "the chat/request path never waits synchronously for the full evaluation graph"],
"acceptance_scenarios": ["general profile runs before additional profiles", "an unresolved task creates only a bounded additional profile linked to failure evidence", "task-aware requirement rejects a model without current supported task evidence", "task scores are included in explainable rating", "provider kind, endpoint and hardware do not affect score", "unknown/failed task evidence never silently becomes supported", "existing requirements without a task profile retain current behavior"],
"required_tests": ["go test", "go vet", "preflight", "check-diff", "check-architecture", "independent", "git diff --check", "migration/schema checks", "CI", "production public health check"],
"terminal_gap_states": ["BLOCKED_BY_SPEC_GAP", "BLOCKED_BY_ARCHITECTURE_GAP", "BLOCKED_BY_MISSING_CAPABILITY"]
}