feat(cognitive): add bounded model qualification probe
This commit is contained in:
@@ -24,8 +24,8 @@
|
||||
|
||||
- `COMPLETION_MATRIX = docs/evidence/audit-completion-matrix-2026-08-23-v3.json`; all seven objective areas are mapped to current repository/production proof, with external bridge, GPU promotion/quality and deep-acceptance gaps explicit.
|
||||
|
||||
- `ACTIVE_TASK = dev/tasks/DEV-000234-orchestration-snapshot-replan.json revision 1`; orchestration waves now persist their GoalID/search snapshot in immutable Work Graph revisions, and verified graph completion returns the next durable search decision. Dispatch-intent discovery/materialization remains an explicit planner hand-off rather than fabricated authority.
|
||||
- `ACTIVE_TASK = dev/tasks/DEV-000235-model-qualification-probe.json revision 1`; cognitive-worker now exposes a bounded provider-neutral qualification probe that discovers actual model IDs, invokes them in parallel, reports reachability/latency and keeps quality `UNRATED` until independent task tests. Hardware presence does not grant promotion or priority.
|
||||
- `LINEAGE_DECISION = docs/GOVERNANCE_LINEAGE_DECISION_2026-08-22.md`; historical packets remain provenance and are not current authority.
|
||||
- `REGISTRY_STATUS = YAML and JSON semantic mirrors agree`
|
||||
- `PRODUCTION_DOMAIN = docker.cin.su`; shared docker-test is not used for this project.
|
||||
- `NEXT_WORK = connect the returned replan decision to capability/agent-intent discovery and automatic next-wave materialization; then add bounded stale-wave recovery and continue remaining audit gaps.`
|
||||
- `NEXT_WORK = use qualification evidence in the provider-neutral Capability rating/selection loop, then continue authenticated UI acceptance, reflection recovery and governed self-development gaps.`
|
||||
|
||||
Reference in New Issue
Block a user