Summary

Activates the existing PlanVerifier skeleton (introduced in PR #2235) with a full per-task and whole-plan verification pipeline based on the VMAO paper (arXiv:2603.11445). When verify_completeness = true, the agent now runs an LLM-judged completeness check after DAG execution and triggers a targeted gap-filling replan cycle for incomplete sub-tasks.

• Add completeness_threshold: f32 (default 0.7) to OrchestrationConfig with sanitizer clamping and startup validation
• Add verify_plan() for whole-plan verification after DagScheduler completes
• Add replan_from_plan() generating root TaskNodes for plan-level gaps
• Wire SchedulerAction::Verify handler in agent loop (was no-op since feat(orchestration): AdaptOrch topology-routing and Plan-Execute-Verify-Replan (#2219, #2202) #2235)
• Insert whole-plan verify step between Done{Completed} and aggregation
• Partial replan DAG runs in a separate DagScheduler with max_replans=0 and verify_completeness=false (INV-2: no recursive loops)
• Partial DAG outputs merged with original task outputs before Aggregator
• Output truncated to verify_max_tokens * 4 chars before verify_plan() call
• GapSeverity implements Display returning lowercase names consistent with serde snake_case and LLM system prompt expectations
• All LLM error paths are fail-open (verify → complete=true, replan → empty Vec, whole-plan → None)
• 25 new unit tests across verifier.rs and experiment.rs
• verify_completeness = false by default — no behavior change when disabled

Closes #2252

Research evaluations (in architect handoff

.local/handoff/2026-03-28T10-59-52-architect.yaml)

• AdaptOrch research(orchestration): AdaptOrch topology routing — DAG-based selection of parallel/sequential/hierarchical patterns, 12-23% gain (arXiv:2602.16873) #2320: P3 backlog — TopologyClassifier already exists; content-aware topology routing needs a feedback loop that VMAO completeness scores can now provide
• AgentRM research(orchestration): AgentRM OS-inspired scheduler — MLFQ lanes, zombie reaping, context hibernation, 86% P95 latency reduction (arXiv:2603.13110) #2334: MLFQ priority lanes P3, context hibernation P4 (separate zeph-memory scope)

Test plan

• cargo +nightly fmt --check — passed
• cargo clippy --workspace --features full -- -D warnings — 0 warnings
• cargo nextest run --workspace --features experiments --lib --bins — 6581/6581 passed
• Verify verify_completeness = false (default) produces no behavior change
• Verify completeness_threshold outside [0.0, 1.0] rejected at startup