Reproduction steps / minimal reproducible example (MRE)
The shortest exact steps or code that trigger the bug, plus how often it fires. What maintainers ask for before they will look at your issue.
See it
What it is
Reproduction steps are the numbered actions that make the bug happen, plus what you expected, what actually happened, and the environment (browser, OS, version, account state). A minimal reproducible example goes further: the smallest runnable thing that still fails, with your app, your auth, and your data stripped out. Stack Overflow popularized the sibling acronym MCVE (minimal, complete, verifiable example), and every serious maintainer asks for one before reading anything else.
Build it by halving. Copy the failing code into a fresh StackBlitz, CodeSandbox, or bare repo, then delete half and re-run. Still broken? Delete half again. Fixed? Put it back and cut elsewhere. Ten minutes of that usually lands on 15 lines, and the collapse itself often reveals the cause before anyone else looks. It is also the single biggest quality upgrade you can give an AI debugging prompt: paste the 15 lines, not the repo.
Gotcha: 'it doesn't work' plus a screenshot of your whole dashboard is not a repro, and neither is a private repo nobody can run. For intermittent bugs, state the failure rate ('3 runs in 10') instead of pretending it is deterministic, and always pin exact versions. A version difference between reporter and maintainer is one of the most common reasons a report cannot be reproduced at all.
Ask AI for it
Turn this bug into a minimal reproducible example for [LANGUAGE / STACK]. Produce the smallest runnable artifact that fits that ecosystem: a single file if one file is enough, otherwise the smallest set of files, services, or containers that still reproduces the failure. Keep every condition the failure actually depends on, and strip everything else: no auth, no routing, no styling, no app-specific helpers, data hardcoded as literals. Pin exact dependency versions in whatever manifest the ecosystem uses, and state the exact command that runs it. Then write the report above it: numbered steps to reproduce, expected result, actual result (with the verbatim error and stack trace), environment details, and the observed failure rate ('fails 3 runs in 10') rather than claiming it always fails. Keep cutting until removing one more thing makes the bug disappear.