The chart found an edge. We asked whether chance could draw it too.
You've seen lots of them: a so-called “alpha” strategy and an equity curve that claims to beat the market. They explain the setup, promise there is no future leak, and show that it earns money.
Most “technical” indicators are essentially astrology with better charts.
AlphaVerify is built to put an end to all that bullshit.
| random OHLC histories | 20 NASDAQ candidates | raw pass—and not proof of alpha |
The figures below are historical results from the former selection-first pipeline; they have not been regenerated with validation of all bins. Their original percentage-point score axes are historical; current shifts store conditional_probability - baseline_probability, and current scores are one hundredth of the former scale. A shift of 0.233 displays as 23.3%, the same 23.3-percentage-point difference.
The "MA Cross" heatmap looks decisive: the classic MA Cross 50/200 produces a coherent 23.3 percentage-point upside/downside contrast.
Then the same full-grid score is applied to 10,000 fitted synthetic histories, and it turns out the decisive edge is just pure luck.
Run commands from the repository root.
The distribution, CLI, and Python package are named alphaverify.
The repository name remains alpha-verify. Reinstall the editable package
after updating an existing checkout.
python -m pip install -e .
alphaverify measure
alphaverify compare
alphaverify validate
alphaverify select
Each pipeline command also accepts --cuda when a CUDA-capable PyTorch installation and device are available.
| Stage | CLI Command | What it does | Mathematical form |
|---|---|---|---|
| 1 | measure | Measure conditional probabilities | π_condition(r, δ, k, t) |
| 2 | compare | Compare them with the baseline | G = 100 × (π_condition − π_baseline) |
| 3 | validate | Validate every supported bin against the null | p̂_k < 0.05 or p̂_k ≥ 0.05 |
| 4 | select | Retain cleared bins and render their shift heatmaps | K_selected = {k : p̂_k < 0.05} |
Validation writes 03_validation/; selection consumes only a current validation result and writes 04_selection/. Every cleared bin is retained—there is no top-k ranking or secondary economic gate.
- Source architecture — implementation boundaries, data flow, artifact contracts, and null mechanics.
- Test contracts — what the fast suite protects and what requires a real pipeline run.
- Workspaces — experiment declarations, plugins, generated artifacts, and safe workspace changes.
AlphaVerify is research software, not investment advice.
The complete numerical specification is available as LaTeX in
mathematics.tex.

