Frozen protocol first
Each project gets an identifier and a written protocol — question, portfolios, horizon, metrics — before any code runs. That is what stops a study from quietly becoming a search for a flattering result.
Paired experiments
Where two strategies are compared, both are evaluated on identical simulated paths — same crashes, same rallies, same order. Any difference is caused by the thing under test and nothing else. This removes far more noise than adding paths.
Real returns, honest option pricing
Market data is resampled in blocks from real daily index history, so volatility clustering and fat tails survive. Options are priced with Black–Scholes at a delta-targeted strike, including the volatility skew that makes out-of-the-money calls cheaper than the headline number. Skipping that one detail is enough to make covered calls look free — see RET-CC-001 for what that error is worth.
Assumptions you can break
Where a conclusion depends on an assumption, that assumption is a control on the page, not a footnote. If turning a knob flips the result, you should be able to turn it. Every project also states plainly what it does not model — usually taxes, always skill.
What this is not
Not investment advice, not a recommendation, not a prediction. These are models, and every model is wrong somewhere. The aim is to be specific about where.