Human-Adjudicated Construct-Level Distress Phenotyping Dataset and Benchmark Materials Using Dreaddit

This dataset contains a human-adjudicated construct-level annotation and benchmark package derived from Dreaddit, a public online distress-discourse corpus. The package includes final 1,500-row gold labels, label provenance, reliability summaries, adjudication reason distributions, construct prevalence and co-occurrence summaries, baseline classifier benchmark outputs, threshold and calibration summaries, codebooks, and reproducibility scripts. The annotations operationalize appraisal-level distress-language constructs, including rumination-like distress, magnification-like distress, helplessness-like distress, perceived injustice language, perceived invalidation language, support expectation mismatch language, and self-threat/identity disruption language. The dataset is intended for research on mental-health informatics, construct-level distress phenotyping, annotation reliability, and computational benchmark development. It is not a clinical diagnostic dataset and should not be used for individual-level clinical inference.