MASH Journalhttps://mash-project.euWorking notes for teams building modest, inspectable machine-learning experiments. Follow the data, establish a baseline and keep the final test honest.Define prediction time before selecting featureshttps://mash-project.eu/guides/prediction-time/https://mash-project.eu/guides/prediction-time/A field is useful only if it exists when the decision is made.Build the baseline before tuning the modelhttps://mash-project.eu/guides/baseline-first/https://mash-project.eu/guides/baseline-first/A comparison is useful when both sides get the same test.Keep preprocessing inside the training boundaryhttps://mash-project.eu/guides/preprocessing-leakage/https://mash-project.eu/guides/preprocessing-leakage/The split applies to transformations as well as the estimator.Select a heuristic without spending the final testhttps://mash-project.eu/guides/heuristic-selection/https://mash-project.eu/guides/heuristic-selection/A runnable synthetic example with separate fitting, selection and final evaluation.A result card another person can rerunhttps://mash-project.eu/guides/experiment-result-card/https://mash-project.eu/guides/experiment-result-card/Keep the task, split, baseline, error counts and reproduction details with the score.