RESEARCH · 2026-09-10
The GDI score as a predictor
FALLEN Fallen — broke its own criteria; kept on purpose
Hypothesis and result
Trades taken with a GDI score of 80 or more were 31.0 % over break-even against 18.3 % otherwise — a result that cleared Bonferroni over 253 combinations and the half-split test. It was composition: one exchange contributed 556 of 557 trades to the low bucket. With instrument and day held fixed, p = 0.49, and 0 of 253 combinations survive.
A clean-looking result can be two populations mixed together. Control for the obvious before believing the striking.
The evidence card
| Sample | 281 trades · 52 days of data |
|---|---|
| Effect | +12.7 pp raw; p = 0.49 controlled |
| Half-split test | Passes — the sign holds in both halves of the record. |
| Multiple-testing status | p = 0. Judged against the whole family of 276 hypotheses Apex has logged (Bonferroni threshold 0.00018): clears it. |
| Stage | Fallen — broke its own criteria; kept on purpose |
| Re-measured | Re-measured when its source script is run — no schedule · last 2026-09-10 |
What would invalidate it
- the effect returns with instrument and day held fixed on 200 new trades
How to read this page
Every number describes the past under the stated conditions. A finding that passes today can fall tomorrow; Apex keeps fallen findings visible on purpose. Stages, tests and the family-wise correction are described in the methodology.