critic-pass

To verify a deep-reasoning draft for logical and factual errors, you can run a critic pass using a script like `node critic.mjs` with your task and draft inputs to validate against unsupported claims and missed constraints.

Checking architectural proposals for missed constraints is best done using a critic pass that operates in report-mode, logging unsupported claims and logical gaps without blocking the commit of your high-stakes output.

The critic pass script does not block the commit of high-stakes outputs; it operates in report-mode, logging factual verification findings and logical errors so you can review potential issues early.

Running factual verification on drafts requires a Sonnet or llm_azure model to parse results and conduct the deep-reasoning checks needed to validate your task prompt against the draft.

A draft verification loop should be used for high-stakes deep-reasoning tasks like financial analyses or migrations, providing a low-cost way to catch logical gaps and factual errors before final commitment.

You can use this critic pass for clinical analysis draft verification, as it is specifically intended for deep-reasoning components to identify unsupported claims and factual errors in high-stakes outputs.