Docs & API
The same research engine, from your notebook or the builder.
Your first scorecard
Install the included SDK from the repository: pip install ./platform/sdk. A public PyPI release is not yet published.
import tensorstreet as ts
# Set TENSORSTREET_URL and TENSORSTREET_API_KEY in your environment.
aps = ts.models.get("aps-mrcr")
mom = ts.indicators.momentum("12-1")
signal = 0.5 * aps.rank() + 0.5 * mom.rank()
bt = ts.backtest(signal, side="long_short",
weighting="market_cap", cost_bps=10)
print(bt.scorecard())API keys have the same access as your account. Keep them in environment variables and revoke keys you no longer use.
Research conventions
Point-in-time signals
A score formed in month M is held in M+1. The engine rejects timestamps from inside the holding month.
Signal blending
Signals are percentile-ranked each month, then combined using normalised positive weights. Only common stock-month coverage is included.
Portfolio and costs
Monthly rebalancing; long-only or long-short; equal or market-cap weights. Costs are charged on turnover. The benchmark is the matched universe, not SPY.
Honest evidence
The scorecard reports benchmark alpha, a return-level momentum regression, a conservative trial-count Sharpe penalty, and timestamp checks. These are diagnostics, not proof of future returns.
API keys
Create a separate key for each notebook or integration.
No API keys yet.