Monolith Research
A research notebook can be made to produce almost any equity curve by softening the friction model under which it runs. Tight spreads, instantaneous fills, no slippage, exit costs forgotten, point-of-entry assumed equal to point-of-decision; each is a small allowance, and stacked together they describe a market that does not exist. The first cost-model decision on a new strategy at this desk is therefore deliberately the wrong way around. The friction model is set to a pessimistic assumption set first, and the strategy is asked to survive it before it is allowed to graduate to a more careful one.
The construction is unfussy. Spreads are widened by a multiple of their median observed value over the test window. Slippage is added at entry and at exit. Stops and trails are simulated as if they are crossed mid-bar rather than at the bar close. Where a position would have had to be carried across a feed gap or a low-liquidity hour, the cost of doing so is paid in full. The point of the construction is not realism in the strict sense; it is the discomfort of competing with a friction model that will not flatter the strategy back.
Three useful things tend to fall out of the construction. First, strategies whose edge is borrowed from optimistic fill assumptions identify themselves quickly; the same code returns a negative curve under pessimistic friction. Second, the comparison between candidate strategies becomes monotonic. The variant that survives stressed costs by a wider margin is the variant that gets the next round of validation. Third, the conversation about leverage gets honest sooner. Sharpe does not scale with size, but drawdown does, and a stressed-cost drawdown of one percent is a different planning input than a low-cost drawdown of one percent.
The cost model is one of the gates between research and live, not the last one. But it is the gate that catches the most candidates, and it catches them earliest. That is enough to keep it as the first.