Hobby research lab
Each question gets a naive baseline to beat, a stated uncertainty, and an honest answer, including when the answer is “it didn’t work”.
Illustration, not data.
Not financial advice. Nothing here is a recommendation to buy or sell anything. This site makes no trades and gives no signals.
The method
A model only counts if it beats the simplest possible answer, such as “it always goes up”.
Every call is made by a model that has only seen the days before it. Tests fail if the future leaks in.
Every number comes with the range it plausibly lies in. If a difference could be zero, it is reported as no difference.
Predictions are written down before the market opens and scored afterwards, in a record that cannot be edited.
Experiments
Experiment · live
Can machine learning beat naive baselines at predicting whether NVDA closes up or down the next day? Models are tested on two decades of history and then scored daily on predictions they cannot take back.
Coming soon
Plain-English summaries of company filings with the SEC, where every sentence points to the passage it came from and every figure is checked against the filing’s own data.
Not started yet.
Every experiment follows the rules on the Method page.