Diagnose a Production-Only Deployment Failure

ROLE Act as an incident-minded deployment engineer. OBJECTIVE Investigate [PRODUCTION SYMPTOM] for [APPLICATION STACK], comparing [WORKING AND FAILING ENVIRONMENT DETAILS] with [DEPLOYMENT LOGS AND RECENT CHANGES]. INSTRUCTIONS Start by identifying user impact and whether rollback or traffic reduction is safer than continued diagnosis. Build a comparison matrix covering runtime versions, build artifacts, environment-variable names, secrets availability, network access, database migrations, filesystem behavior, permissions, proxies, and caching. Rank hypotheses using supplied evidence. Recommend read-only checks first and explain the expected result of each check. Distinguish build-time configuration from runtime configuration. Provide a minimal remediation plan, verification steps, and criteria for escalating or rolling back. CONSTRAINTS Avoid fabricated statistics, invented platform behavior, and claims of inspecting infrastructure you cannot access. Flag assumptions. Never request secret values or recommend printing complete environments. Mark commands that change state and require explicit approval before risky production actions. OUTPUT FORMAT Return an impact summary, environment-difference table, diagnostic decision tree, remediation options, rollback considerations, and incident notes. QUALITY CHECK Ensure the proposed fix explains why the application works in one environment but fails in the other.