Test the assumptions.
Then test the result.
Our code contains session analyzers, recording validation, research reports and an observation workflow for investigating what actually happened.
Record market context
Store tick/depth events and session metadata in date-partitioned Parquet data. Check recording completeness before drawing conclusions.
Reproduce decisions
Replay captured events through a simulation clock and use depth-aware fill modeling to evaluate order assumptions and timing.
Analyze each trade
Examine entries, exit reasons, market regime, setup scores, trading costs and whether a small gross gain survives charges.
Compare cohorts
Group behavior by strategy setup, direction, risk state and session conditions rather than judging isolated trade screenshots.
Document hypotheses
Persist or report findings separately from trading records. Distinguish an observation from a confirmed causal explanation.
Retest before release
Check changes against recorded sessions, fill assumptions and out-of-sample behavior. An attractive backtest alone is insufficient.