Public self-audit

We score our own homework — in public

Every number below is computed from resolved predictions — each one published before the outcome was known, then scored against the actual close. It is forward out-of-sample by construction. Unlike audit pages that quietly blend in simulated history, we split out walk-forward backfill explicitly, and the daily CSV is committed to a public git history you can inspect.

Directional hit rate

49.2%

95% CI [48.7%, 49.8%]

Mean IC (per-date)

0.029

IC-IR 0.19

HAC t-stat

3.19

significant (|t|≥2)

Resolved predictions

34,381

since 2025-05-20

Provenance — live vs backfill

5,130 live (published in real time) · 29,251 backfill (walk-forward simulation, no look-ahead — each anchor only saw data before its date). We label these honestly rather than blending them; trust the live segment most.

Directional hit rate by month

25-05

25-06

25-07

25-08

25-09

25-10

25-11

25-12

26-01

26-02

26-03

By confidence

confidence	n	Hit rate	95% CI	IC
high	227	48.0%	41.6%–54.5%	0.126
medium	1,158	45.0%	42.1%–47.9%	-0.026
low	3,745	46.5%	44.9%–48.1%	-0.035

By horizon

horizon	n	Hit rate	95% CI	IC
1d	11,640	47.2%	46.3%–48.1%	-0.013

By market

Market	n	Hit rate	95% CI	IC
US	3,851	50.8%	49.3%–52.4%	0.061
JP	2,671	49.6%	47.7%–51.5%	0.011
KR	2,434	48.6%	46.6%–50.6%	0.001
SG	2,200	48.3%	46.2%–50.4%	-0.022
MY	2,193	48.6%	46.5%–50.7%	0.051
VN

Want to verify rather than trust? Download the full resolved dataset from /dataset, inspect the per-model CV metadata in the model registry, or read the live, unaggregated rows in the predictions log. A daily snapshot of the resolved CSV is committed to our git history as a tamper-evident timestamp chain.

Hit rates near 50% are the honest base rate for liquid equities at short horizons. Small, statistically-real edges — risk-managed and compounded — are the thesis. This page is historical performance reporting, not investment advice, and is not a promise of future returns.