feat: train execution-mode priors separately from plan-key priors in offline policy builder
Done.
Implemented
- Extended
policy_stats.jsonwith anexecution_modesbucket. - Extended
policy_candidate.jsonwith learned priors fordirectandplan. - Updated
train-policy-offline…
feat: train execution-mode priors separately from plan-key priors in offline policy builder
fix: align shadow controller execution-mode semantics for single_tool vs run_plan
fix: align shadow controller execution-mode semantics for single_tool vs run_plan
Done.
Implemented
- Added explicit
execution_modeto shadow-controller outputs. - Added
execution_modeto typed trajectories. - Changed bandit bias so
single_toolprefers direct execution…
feat: bootstrap the system topology graph from infrastructure manifests and live inventory
perf: replace repeated analysis.update / return-analysis loops with a single idempotent observation commit
chore: emit structured counterfactual decision traces from the shadow meta-controller
feat: build an online-to-offline distillation loop for policy improvement beyond static sacred gating
feat: implement hierarchical memory tiers for semantic, episodic, procedural, and causal system knowledge
feat: add cross-system causal edges to the typed tool graph for root-cause traversal
refactor: split the orchestrator into policy-router, effect-executor, and observation-reducer modules
fix: align shadow controller execution-mode semantics for single_tool vs run_plan
openclaw
created branch main in openclaw/openclaw-intelligence-core-public
2026-03-21 08:35:56 +01:00