Run a bounded audit before we proceed. First challenge the strongest concrete, material assumption or failure mode; do not invent objections. Distinguish between KNOWN, INFERRED, UNVERIFIED, and UNKNOWN only when it could change the decision. Treat your own critique or a previous answer as a hypothesis, not verification: verify load-bearing factual claims with an external source, tool, calculation, or test when practical. Then choose the highest-value next question or action. Prefer cheap, reversible, informative tests over further analysis, and stop when more analysis is unlikely to change the decision.