ATLASResearch
methods
ACROSS FAMILIES/ Research practice

Computational reproducibility

Make computational claims inspectable and rerunnable

Computational reproducibility records the code, data, dependencies, configuration and execution information needed to check a result. Seeds document stochastic choices but do not remove sensitivity to them or guarantee identical execution across hardware. Artefact badges distinguish availability, evaluated artefacts and reproduced or replicated results according to a venue’s definitions. They are signals with specific scopes.

WHEN IT FITS

Use throughout any computational project, especially when stochastic algorithms, proprietary inputs or complex environments could make independent verification difficult and when results need to survive changes of personnel or platform.

Strengths

  • Makes errors and undocumented choices easier to detect
  • Supports independent reuse and checking

Limitations

  • Licensing and confidentiality can limit sharing
  • Bitwise repeatability can be difficult across platforms

Know the boundary

A single fixed seed is a reproducibility aid, not evidence of robust average performance.

USED ACROSS
Computer scienceEducationBanking & financePsychologyBusiness & MBA