Manifest-bound lineage

Every result is tied to a manifest—a complete audit trail of: which model? which dataset version? which code kernel? Run it again and you get identical results. No "I can't reproduce this" or "the model drifted."

Reproducible results

Deterministic. No hidden randomness. Audit teams can verify what risk team found, what research team backtested, what execution actually did.

Risk, research, and audit teams read the same deterministic geometry

Everyone has one source of truth. No spreadsheet copies, no version conflicts, no "my model says X but your model says Y because we ran different code."

Signal generation through execution and evidence

The lineage traces the entire flow: how the trading signal was generated (RL model? rules engine?) → how it was executed (what slippage? what fills?) → what the actual evidence shows (did it work as backtested?).