FROM EVIDENCE TO LEARNED DYNAMICS
Data and model pipeline
The event stream connects the two: source material becomes a history that can be inspected, represented, and learned from.
SIX STAGES · SELECT A STAGE TO FOLLOW THE HANDOFF
↳Other usesThe event stream also feeds company knowledge, search, workflow discovery, and evidence inspection directly.
Stage outputs
Purpose
Code references
Example data
Model components
A trained encoder turns available history into a compact state. A predictor estimates change in that state. Fitted readout heads translate it into specific questions: outcomes, unusual changes, or action-conditioned next states.
See the research and model comparisons. The deliverable is a model bundle with its feature definitions, preprocessing, time splits, checkpoints, and evaluation evidence—not just a weights file.