Summary data from a three-year prometryn exposure experiment on durum wheat (Triticum durum)

Treatment-level summary data from a three-year pot experiment in which durum wheat (Triticum durum Desf.) was grown on soil treated with the photosystem II-inhibiting herbicide prometryn at 0.2x, 1x and 10x the field-recommended rate, plus an untreated control (six pots per dose). The primary replicate-level records of this experiment were lost following hardware failure. The values deposited here were recovered from the cached numeric series inside the Microsoft Word chart objects of the archived manuscript files, and were validated by reconstructing the previously published one-way ANOVA (recovered F = 1.096, p = 0.3738 against the published F = 1.10, p = 0.372). The deposit contains 36 treatment-level rows in tidy long format: four soil organic matter doses with means, standard deviations and replicate counts, and thirty-two plant means spanning four biochemical endpoints (total chlorophyll a+b, total soluble sugars, total soluble proteins, proline), two growing seasons and four doses. Standard deviations survive for the soil endpoint only and have not been imputed for the plant endpoints. A 24-row robustness table gives, for each dose-versus-control contrast, the largest pooled coefficient of variation under which that contrast would still reach p < 0.05 at the realised replication. A codebook defines every field. The complete analysis script regenerates every statistic and figure in the associated article from these files in about six seconds. Intended reuse: inclusion in meta-analyses of photosystem II-inhibitor effects on non-target crops, and reapplication of the robustness screen to other legacy datasets whose replicate variances have been lost. The dataset is too small to train or validate a predictive model and should not be used as a machine learning benchmark; the robustness verdicts are derived classifications under an assumed coefficient of variation, not ground truth.