Literature-anchored factors. Every factor is a construction with a known economic mechanism and a decades-long public record — momentum, value, quality, issuance. Sign known before testing; no mined interactions.

Validation before deployment. Shuffle-null tests, walk-forward folds, multi-offset entry perturbation, parameter-flatness checks. Over 90% of researched ideas fail and are archived, not retried until they pass.

Hold, don't fiddle. Exit research converged on a hard result: no stop-loss or ML exit policy beat holding to the rebalance date. The system trades on schedule, not on emotion — because it has no emotions.

Risk lives at the portfolio layer. Entry gates and exit rules failed every test; volatility-complex overlays and regime weighting passed. Risk is managed where the statistics say it can be — above the stock picks, not inside them.

Honest expectations. Live performance is internally budgeted conservatively against research results — expectations are set below what the numbers show, never above.