Debug the thing that only fails in production

This works locally and fails in production. Symptoms: {{describe, including frequency and pattern}}. Environment differences I know of: {{list}}. Logs: {{paste}}.

Work the problem:
1. List every difference between local and production that could plausibly matter — concurrency, data volume and shape, latency, clock, locale and timezone, memory limits, connection pooling, caching layers, load balancer behaviour, environment variables, permissions, and cold starts.
2. Rank them against the observed pattern. Explain why the pattern (intermittent? time-of-day? one region? only large payloads?) supports or excludes each.
3. For the top three, tell me the exact evidence to collect and where. Prefer evidence I can get without another deploy.
4. Give the safest way to reproduce it — locally if possible, in staging if not, and what to add to make production observable enough.
5. State the workaround that stops the bleeding now, and how it differs from the fix.

If the logs contradict my description, say so.