Work · Research
Quantitative method
A validation process built so a negative result can be trusted.
The statements are not published. Their order and their status are.
Three independent out-of-sample checks, on every line.
- H-01Hypothesis oneFalsified
- H-02Hypothesis twoFalsified
- H-03Hypothesis threeFalsified
- H-04Hypothesis fourFalsified
- H-05Hypothesis fiveFalsified
- H-06Hypothesis sixFalsified
- H-07Hypothesis sevenFalsified
- H-08Hypothesis eightFalsified
- H-09Hypothesis nineFalsified
9/9Falsified
Nine hypotheses pre-registered, nine negative results, and not one reinterpreted afterwards. That is the only thing that makes a check worth trusting.
The problem
Analytical results are easy to make look convincing and hard to trust. Most failure modes are ways of fooling yourself: fitting noise, testing until something passes, or measuring against the wrong benchmark. The engineering problem is a process that can return a negative answer and be believed.
Approach
- Each hypothesis is registered in writing before it is tested, so a result cannot be reinterpreted afterwards. Nine consecutive hypotheses have been recorded and falsified.
- Validation runs out of sample in three independent ways, including a holdout of instruments never used during development.
- An empirical null control establishes what a false positive looks like in this pipeline before any true positive is trusted.
Measured
Built with
- Python
- pandas
- NumPy
- SQLite
- Amazon EC2
- systemd
Stated plainly
Presented as methodology only. No performance figures are shown and none should be inferred — a research process transfers between domains, a track record does not.