Practical model how-tos

How to use these guides

The three guides below map to the order you should actually make the decision in. Start with the most fundamental question — whether to build a model yourself or run one that already exists. For most beginners the honest answer is “run one, at least to start,” and the build guide says why: the step that settles whether an edge is real is the forward record-keeping, and it is the step almost everyone skips. If you are going to evaluate any model — your own or someone else's — the second guide is the one that matters most: how to read a track record without being fooled, with a check you can repeat on any service. The third, rules-based versus discretionary, settles the question underneath both: why a process beats a hunch on every axis that can be measured.

None of them assume you will take the recommendation on this site on trust. Each is built so you could apply it to a competitor and reach your own verdict; the site simply argues that one set of tested models comes out the other side intact. Where a guide leans on a specific property — testability, measurability, repeatability — it links through to the matching criterion lesson so you can go as deep as you want.

The mistake these guides are meant to prevent

The most common and most expensive error a beginner makes is treating a big return number as proof. A return is the last thing to trust and the easiest thing to dress up: it costs nothing to print a spectacular figure and quietly omit the trade count, the losing months and the drawdown that produced it. By the time a subscriber notices that the live results do not match the homepage, the fee is spent. Each guide here is structured to move you from believing a claim to checking one — from “those numbers look great” to “I confirmed one of those calls myself.” That shift is the whole value of the cluster, and it is why the guides are deliberately short on opinion and long on procedure.