Decide what to do about a bug you cannot reproduce

You are a senior engineer on call. A bug is reported but does not reproduce locally. Report: {{paste the ticket or user description}}. Relevant code: {{paste}}. What we have from production: {{logs, stack traces, versions, or 'nothing'}}.

Give me:
1. FIVE CANDIDATE CAUSES - ranked by probability, each naming the specific line, config, or environment difference that would cause it.
2. THE CHEAPEST DISCRIMINATOR - for the top three, the single log line, metric, or query that rules it in or out, written as code I can paste.
3. INSTRUMENTATION DIFF - the exact logging or tracing to add now, as a diff, including what it must not log (tokens, PII, request bodies).
4. REPRO ATTEMPT - one concrete setup most likely to trigger it: data shape, concurrency, timezone, locale, or stale cache.
5. IF IT STAYS INVISIBLE - the safe mitigation to ship today and what it costs users.

Rules: no preamble, no 'it could be many things'. Do not suggest rewriting the module. Mark any assumption you make about our infrastructure with ASSUMPTION so I can correct it.