Answers computed from real US market data.
But what data?Here's what I found searching the web…
0 numbers computedAsk it to do the work
Every order and every cost, against the market.
Plain-English rules, checked after every US market close and sent to your inbox, only when something changed.
3 S&P 500 stocks fell more than 8%Triggered daily change ≤ −8%
Pine Script you can download and add to any TradingView chart.
1234567891011121314
The instruments
Read-only SQL across 36 tables. Facts and aggregates come from a query, never from memory.
pandas, numpy and statsmodels in a kernel that stays alive for the turn, so data is loaded once and worked on in steps.
Market regimes, next-month relative return and quarterly report quality — each quoting its own out-of-sample metrics.
News and context for why something moved, and discovery of which companies a theme actually means.
Results come back as tables, charts and cards you can inspect, export or publish — not a wall of prose.
Hard-won corrections and validated query patterns are saved and recalled, so a mistake is made once.
Financial standards, built in
A quarter ends weeks before it is filed. History joins on the filing date, so a backtest never knows a number before the market did.
Every bar is split- and dividend-adjusted, and refetched in full when a corporate action restates the past.
Models are retrained on data strictly before each test window and judged only on what they never saw.
Rank IC, t-statistics and hit rates travel with every prediction. An IC above 0.15 is treated as a leak, not a discovery.
The universe is today's listings, and the agent says so — with the delisted archive available when history needs it.
Every step, query and chart is in the answer. Nothing is asserted that cannot be opened and checked.
Research
