Monolith Research
A working file kept by a research desk has two readers in mind. The first is the desk itself, six months from now, looking back at the reasoning behind a decision that has by then produced its own evidence. The second is anyone willing to take the time to read it carefully, whether a counterparty, a future collaborator, or an honest observer of the work. The entries here are written for both readers and recommend nothing to either.
Each entry is dated and tied to a versioned notebook with a frozen data snapshot. Where the snapshot is no longer recoverable the entry will be withdrawn rather than re-cut. Numbers reported are after the friction the desk would have actually paid, not the friction the model assumed when it was tuned. Strategies that did not survive the friction will be filed alongside the strategies that did. That is the value of the record. A clean back-test in a drawer is worth less than a clean back-test next to the three near-identical attempts that failed before it.
The discipline this enforces is mundane. A research record is a constraint on the future as much as a transcript of the past. What is written here, the desk has to live with later. The hope is that the writing makes the later choices clearer; the working assumption is that it will, often enough, make them harder. Both outcomes are part of the point.