Reproducibility: Seeds, Environments and the Ablation
A result that cannot be rerun is a claim, not a finding. Fixing seeds, pinning the environment and recording the exact command make a run repeatable, and an ablation is what shows which part of a method the gain came from.
What a learner can do afterwards
- List what has to be recorded for a stranger to repeat a run
- Design an ablation that isolates one component's contribution
- Say why a single seed is not a result and what to report instead
1 · Read
A result nobody can rerun is a claim, not a finding. Record everything a stranger needs: the fixed seeds, the pinned environment, and the exact command that produced the numbers.
An ablation shows where the gain came from. Change or remove one component while holding everything else fixed, and the change in the result is that component's contribution.
You add a new part to a method and the score rises. Run the full method, then run it again with only that part removed and the same seed. The gap between the two runs isolates what your part added.
One seed is not a result; it may be luck. Run several seeds and report how the numbers vary, so readers see the finding and its wobble.
Pin the seeds and the setup, isolate each gain with an ablation, and never report a single lucky seed.
2 · Watch
Take it off screen
Where it sits
8 questions wait behind this lesson, each with its answer explained. Every answer feeds the sky: stars light as they are learned, and dim when it is time to come back.