Quantitative Researcher

The problem.

Most backtest edges are the largest draw from an uncounted search. Crypto derivatives data is short, non-stationary and expensive to hold out. You own the judgement that separates signal from artefact. You own what happens when the market disagrees.

What you would do.

• Take a hypothesis from a sentence to a specification a simulator can execute. Write down what would falsify it before you run it.
• Build the validation: purged cross-validation, walk-forward windows, an embargo that is justified rather than conventional.
• Keep the trial ledger honest. Every configuration you evaluate is counted, including ones abandoned early.
• Model cost properly — taker and maker fees, spread, documented slippage, funding, maintenance margin — before you believe anything gross.
• Put the result in front of the market yourself. Watch it in production. Write the note on what it did and what you had not anticipated.
• Retire it when the evidence says so, and say why in writing.

What we would want evidence of.

• You are not done with a backtest until you have tried three ways to break it.
• You can say what you do when the out-of-sample window is short and the regime changed inside it.
• You are comfortable concluding that a result does not hold, including on your own work.
• You can explain why a number is significant without using the word "significant" as the answer.
• You write code that another person can run in a year, because you have had to run your own.
• You have an opinion about multiple-testing adjustment that survives a follow-up question.

What would make this a bad fit.

This role is a poor fit if you want a specification handed to you and a result graded at the end. The question is yours to find.

This role is a poor fit if your instinct on a disappointing result is to widen the search. We will disagree early.

This role is a poor fit if you want to hand a signal to an engineering team and move on. There is no team to hand it to. You build it.