Stock forecasts you can audit, not prompt.
Enter a ticker. The engine pulls full daily OHLCV history plus QQQ as the market-regime benchmark, discretizes the state space, selects parents by mutual information, fits BDeu-smoothed conditional probability tables, runs exact inference for the 15- and 20-trading-day horizons, and validates walk-forward against climatology. Every run is saved, so a rerun gives you a delta table instead of a fresh guess.
No card required. Full access for 5 days, then a free tier of 3 analyses a month.
Deterministic, not generative
Indicators, discretization, mutual-information parent selection, BDeu-smoothed CPTs and exact inference run in code — the same inputs always give the same posteriors.
Validated against climatology
Every run is walk-forward tested against a base-rate and naive-Bayes benchmark, so you see whether the network has an edge before you act on it.
Catalyst overlay
A news layer maps upcoming and recent catalysts onto the probability picture, kept strictly separate from the quantitative engine.
What a run gives you
- Evidence table with every discretized state
- 3–4 week posterior distribution and barrier probabilities
- Walk-forward log-loss vs climatology and naive Bayes
- Historical analogues and median forward path
- Monte Carlo touch probabilities for your levels
- Support / resistance map and CSV export