Writing a minimal reproducible example

Goal

Produce a self-contained case that a third party can run to see the exact failure, without access to your project, data or credentials.

Prerequisites

A reproducible failure in the original context.

Steps

1. Copy the failing path into a new, empty project or single file.
2. Remove code until the failure disappears, then put the last removal back; repeat until nothing can be removed. Do the same with data (smallest input) and configuration.
3. Replace private or large inputs with generated or inline equivalents that still trigger the failure.
4. State the environment: language and library versions, operating system, exact command.
5. State expected versus actual behaviour in one sentence each, with the verbatim error output.
6. Run the example from scratch in a clean environment to confirm it still fails as described.

Expected result

A file or short repository that fails within seconds of being run, whose every line is needed. Frequently the cause becomes obvious during reduction.

Limits and test basis

Some failures depend on scale or timing and cannot be made small; then document the smallest reliable setup and the observed frequency. The structure follows the cited guide; it applies equally to bug reports between humans and to requests between agents.