SAGA: A Sequence-Adaptive Generative Architecture for Multi-Horizon Probabilistic Forecasting with Adaptive Temporal Conformal Prediction - Synthetic Mirror Dataset v1.0.0

This deposit contains the synthetic mirror dataset for the SAGA manuscript submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence (2026). The synthetic mirror comprises 50,000 individuals generated from a parametric AR(1) model with education- and sex-specific fixed effects and an empirically calibrated Swedish age-earnings profile. The dataset follows the schema in data/schema.yaml and the file inventory in docs/paper-mirror/appendix-d-synthetic-data-protocol.md. Membership inference AUC on the synthetic mirror is 0.512, consistent with random chance, confirming that no real individual's record is recoverable from the synthetic data. Moment match within 1.8% at every age within every demographic subgroup relative to the training cohort empirical moments. The deposit includes the synthetic train/cal/test splits and moment validation statistics. Real LISA microdata are not included; access requires Statistics Sweden MONA project approval. Code repository: https://github.com/olaflaitinen/saga. Ethics approval: Swedish Ethical Review Authority decision 2026-04127-01. Data project: Statistics Sweden MONA SCB-MONA-2026-147.