Stock backtesting on minute-resolution history

There is no pipeline to stand up and no notebook to maintain. The series is fetched, the rule is evaluated and the result is rendered client-side, which is why iterating on a condition costs seconds rather than an afternoon.

"Exit at the close" is resolved through the exchange's own time zone, recomputed per hour rather than pinned to a fixed UTC offset — so a test spanning a daylight-saving change does not put half its exits an hour wrong and call it alpha.

The tools that rank for portfolio backtesting test one asset class each. DepthFeed joins equities, ETFs, crypto spot, perpetual futures and recorded prediction-market order books onto a single time grid, so a rule can read an equity gap and act on a crypto position — or read either and act on a Kalshi contract — without you reconciling three vendors' timestamps by hand.

History follows your plan, and bars use the same window as order-book depth: 7 days on the free Explorer tier, 30 on Quant, 90 on Research, and the full available history for each instrument on Desk. Long-horizon work — multi-year allocation studies, regime comparisons across cycles — is a Desk workload. Shorter windows are enough to validate a rule before you commit to the archive.

These are source-attributed bars, not a consolidated (SIP) feed. For research, backtesting and signal work that distinction rarely matters; for best-execution reporting or regulatory record-keeping it does, and you should use a consolidated tape instead. One-minute is the finest resolution — the coarser frames are rolled up from it, so a 1-hour bar and sixty 1-minute bars always agree.