debug

To investigate a root cause for a 500 error or timeout, collect logs, metrics, source code, and configuration details into an evidence ledger before diagnosing the offending layer. This evidence-first approach ensures causal claims are strictly grounded in cited artifacts.

The best way to triage test failures and deployment regressions is to correlate evidence artifacts from an evidence ledger to identify the offending layer. This method enforces a minimum evidence breadth gate across artifact categories, preventing unsupported causal claims during diagnosis.

Confidence scoring for unexpected exceptions and incorrect function outputs works by mechanically deriving a confidence tier (Confirmed, Probable, Suspected, or Unknown) based strictly on what was actually reproduced and inspected from the collected evidence ledger.

Yes, you can use read-only investigation for incident response without risking state-changing actions. This approach requires explicit per-turn confirmation before any mutation that could affect shared systems, ensuring safe investigation of live incidents and development bugs.

You need to gather configuration details and test output for a stuck job or deployment failure immediately upon reported breakage. Collecting these artifacts into an evidence ledger before making any causal claims ensures accurate root cause diagnosis across all project layers.