What the person brings that the models cannot

The terminal answers three questions: which names, which regime, and when. Everything else on the screen is arithmetic. The question worth asking is what is left for the person.

The measurements are honest about their own limits

A permutation null says a result could not come from shuffling within the window. It says nothing about the next quarter. Ninety-three days of sentiment, four and a half years of prices. The models know what worked inside those windows and nothing else.

What a person knows that the data does not

First, what changed in the world. A delisting, a legal action, a project that lost its team, a chain upgrade. The news feed carries some of it. The rest lives outside the corpus and arrives in the judgement.

Second, whether this quarter looks like the ones we measured. Regimes repeat with variation. A person can say this tape is like that one, except, before any model sees enough data to notice.

Third, what can be afforded. The model ranks names. It does not know your capital, your margin, or what a drawdown would do to you. Position size and the stop are not inputs to the model. They are the decision.

Fourth, the discipline to hold when the model is quiet. The longer-term read names a horizon and an entry rule. Sticking to the rule when the screen disagrees is a person's job, not the model's.

What the model does that the person should not try

The model sees every name every day. It ranks without moods, without favourites, without the hope that the losing position turns around. A person hand-ranking nine hundred names is a model without a GPU.

The division of labour

The terminal filters, ranks and times. You decide, size and stop. A model that decides is a desk without an owner. The product is the line between the two.