TDM provides the training data substrate used by the forecasters. It pulls UIP's daily universe inventory, fetches a configured number of days of one-minute OHLCV bars for each symbol, then computes a normalized, daily-wide derivative dataset (forward-filled OHLC, summed volume, anchored to interval buckets, dropped partial bins, pivoted long→wide). SIG and MAG train on TDM's derivative dataset; nothing downstream re-derives bars.