alphakernel · livedaily bars · long/shortpoint-in-timeturnover-aware costs

Backtest on the platform engine.

Cross-sectional momentum run by thealphakernel HTTP service — the same Python engine that processes a real OHLCV panel in research workflows, demonstrated here on a synthetic universe of 24 tickersover 750bars. Long-short quantile sizing, walk-forward semantics, turnover-weighted costs. For the same mechanics running entirely in your browser (no backend, planted signal), see/sandbox; for multi-seed A/B against the long-only baseline see/strategies/compare; for the strategy catalog see/strategies.

data · what universe?

universetickers24bars (days)750seed

signal · what window?

lookback (bars)60holding (bars)10long quantile (%)20

execution · what costs?

cost (bps / rebalance turnover)4

alphakernel · ready

sharpe

—

ann. return

—

ann. vol

—

max drawdown

—

avg turnover

—

rebalances

—

equity · strategy vs equal-weighted benchmark

base = 1.0. Long top quantile, short bottom, rebalanced every 10 bars.

strategy benchmark

drawdown · peak-to-trough on strategy equity

lineage card · from the alphakernel response

run_id —

engine —

elapsed —

strategy xs_momentum

universe —

window —

cost_bps —

sharpe —

max_dd —

how this is wired

Click run and the browser POSTs to alphakernel.dev/v1/backtest. On the server,alphakernel.server.apibuilds a synthetic L1 panel, runs the cross-sectional momentum strategy throughalphakernel.backtest.run (the same vectorised engine that's used for real OHLCV panels), and returns equity, drawdown and stats.

Right now the server still uses synthetic prices — what changes here vs the sandbox is theengine (Python / polars / numpy, walk-forward semantics, the same code path real models exercise), not the data. When real-data sources land,/v1/backtest swaps inputs without touching this page.