Heisenbug Reproduction Strategy

ROLE: You are a senior debugger who insists on a reliable reproduction before changing any code.

CONTEXT: We have an intermittent bug in [SYSTEM] that appears roughly [FREQUENCY] and only in [CONDITIONS, e.g. production under load]. We cannot reproduce it locally. Available evidence: [LOGS/METRICS/TRACES/REPORTS].

KNOWN FACTS:
[PASTE_WHAT_YOU_KNOW]

TASK (think like a scientist):
1. Form 3-5 hypotheses about the conditions that trigger the bug (timing, specific data, concurrency, environment, scale, dependency version).
2. For each hypothesis, design the cheapest experiment that would confirm or refute it, and state the expected observation under each outcome.
3. Specify the instrumentation to add (targeted logs, correlation IDs, metrics, feature-flagged debug mode) that would make the bug observable without flooding logs.
4. Propose how to compress the reproduction: controlled load test, replayed production traffic, fault injection, fuzzing, or a seeded deterministic harness.
5. Define the exit criterion: what observation proves you can reproduce it on demand.

OUTPUT FORMAT:
- 'Hypotheses' (ranked, with rationale).
- 'Experiment plan' (table: hypothesis | experiment | expected result if true/false).
- 'Instrumentation to add' (specifics).
- 'Reproduction harness sketch' (pseudocode or steps).

CONSTRAINTS: Do not propose code fixes yet. Prioritize experiments by information gained per effort. Ensure added logging is privacy-safe and rate-limited.