Plan · cascade
PLV — Plan-Level Verification
Checks whether an agent’s plan or multi-step output still matches the stated goal and evidence — step by step, before you trust the result. Built for compliance, audit, and prompt/trace validation workflows.
Call example
curl -X POST https://verify.thoughtproof.ai/v2/verify \
-H "Content-Type: application/json" \
-H "X-API-Key: YOUR_API_KEY" \
-d '{"question":"Should the agent approve invoice #4421?","answer":"Approve €500 payment to vendor-42.","trace":"Checked invoice, vendor history, and amount limits.","plan_steps":[{"index":0,"description":"Validate invoice","criticality":"critical"}]}'Standard and Thorough pricing: PLV on /pricing/.
When to use it
- Plan vs. goal drift The agent produced a multi-step plan. Does each step still serve the mandate, or did it quietly reframe the job?
- Trace / output validation Prompt- and trace-level checks for compliance and audit workflows that need more than a single exit gate.
- Cascade depth Standard and Thorough tiers trade latency for scrutiny — pick the depth the decision warrants.
What you get
Native ALLOW / BLOCK / UNCERTAIN outcomes against a bounded decision package. Optional attestation paths where configured. Measured false-allow posture is documented on the evidence note.
Related
Typical agent exit: Sentinel. Adversarial multi-model panel: RV. Five-axis scoring: DQL.