Monolith Research
A new research candidate at this desk is not allowed to move from notebook to live size in a single step. Between the two are four gates, each with its own evidence requirement and its own veto. The structure is named because the names are doing work; once a gate exists, it is harder to skip than a habit.
The first gate is walk-forward. A static-sample back-test is treated as a hypothesis only; the strategy is then required to be re-tuned and re-tested across rolling time folds with stressed costs. The metric that matters at this gate is not the headline Sharpe of any single fold but the consistency of the worst-fold return and the worst-fold drawdown. A model that wins on average and loses badly in a single fold is a model whose live behaviour will look like the worst fold sooner than its author expects.
The second gate is paper. The candidate is run forward in real time against the live feed, with no orders placed, for a period long enough to observe at least one regime change or one event window. The metric at this gate is not return; it is the agreement between the signal the live process produces and the signal the research notebook produced from the same data. Disagreement at this gate is almost always a systems problem rather than a strategy problem, and the systems problem has to be fixed before any further work is done.
The third gate is shadow. The candidate runs alongside a production strategy, on the same feed, producing real orders against a separate account at a deliberately small size. The metric at this gate is the difference between simulated fill and real fill. The gap between the two is the part of the strategy that the back-test could not observe.
The fourth gate is small size. The candidate runs at a fraction of the size the back-test would suggest, against the live book, with the deterministic shell already in front of it. The metric at this gate is whether the live equity curve agrees with the shadow curve under real friction. If it does, size is increased on a schedule, not on a feeling.
None of these gates produces a profitable strategy on its own. Together they produce the operational record that decides whether a profitable strategy in a notebook is allowed to become a profitable strategy in a market.