Systems

REPLAY-LAB

A bar-by-bar futures backtest engine that reproduces TradingView trade-for-trade, plus a visual replay lab for inspecting what it did. It exists to answer one question honestly: is this an edge, or is it noise? Most backtest code answers that wrong in ways that stay invisible until real money is on the line. It is the engine AlphaFactory's surviving candidates are validated in.

Why it is built this way

The engine owns the bar series. On step i a strategy receives a slice that ends at the current bar, and there is no API that yields a future one. Indicators are incremental state machines fed a bar at a time, so they cannot be precomputed over the full series — nothing downstream of the engine ever holds it. Lookahead is unrepresentable rather than merely avoided, which matters because lookahead is the single most common way a backtest lies.

Time is explicit for the same reason. Bar timestamps are always UTC at bar open, vendor exports convert at the boundary, and the futures session has exactly one definition. Timezone bugs in backtests are silent — they shift entries by an hour twice a year and quietly change the results — so that layer carries its own tests for DST transitions and session rollover.

ALPHAFACTORY

An LLM-driven research loop that generates, backtests and walk-forward validates strategy hypotheses for equity index futures. Over 1,700 hypotheses evaluated to produce 12 independently validated systems now deployed across funded accounts. The whitepaper covers the architecture, the validation protocol, and the research dead-ends.

replay-lab is the backtest engine those candidates are validated in. The two remain separate codebases: a strategy that survives screening is re-implemented in Rust to run there, so the validating pass executes an independent implementation rather than the code that produced the original result. A strategy that still reproduces trade-for-trade after that port is a stronger signal than one that only ever agreed with itself.