Experiment reproducibility
Retain the problem, configuration, execution context and analysis needed to examine a result.

Could someone tell whether a different result came from the method or the setup?
An experiment record should connect the question with the exact setup used to investigate it. Capture the inputs, implementation versions, execution settings and the environment in which the work ran.
Distinguish repetition from a conclusion that generalises. Repeating one configuration helps examine variability within that setup; it does not establish behaviour across every problem instance or hardware condition.
Keep raw observations and the analysis steps together. Record exclusions and failed runs rather than preserving only the most persuasive outcome. A reproducible workflow makes a claim easier to challenge and improve.
What to take forward
- Versioned inputs and configuration
- Clear execution context and observations
- Transparent analysis and exclusions
What should I bring to a conversation?
Bring the decision you need to make, the constraints you already know and a small, non-confidential example. We can then agree whether a resource, a focused review or a more detailed discussion is the right next step.
