Why backtests can mislead
Timing, revised data, missing evidence, and the gap between simulation and execution.
Ask when the information was known
An observation’s market date is not necessarily its availability date. Later corrections or completed training can create information that was unavailable to the original decision. Reconstructing old prices with today’s knowledge is retrospective research, not automatically a point-in-time simulation.
Freeze the fitted artifact
Using evaluation data to choose thresholds can contaminate a result. The source regime workflow saves a fitted artifact with training cutoffs and parameter manifests. Evaluation consumes that frozen artifact. This prevents silent refitting but does not by itself establish out-of-sample performance.
Do not erase the hard cases
Missing observations, failed captures, unfilled orders, and unresolved source conflicts should remain in the record. Omitting them can make a strategy appear easier to execute than it was. The engine preserves explicit exclusions and blocks dependent measurements when required evidence is missing.
Performance is another validation problem
Costs, capacity, position sizing, instrument adjustments, and shared cash can materially change a simulated outcome. The source has bounded comparison workflows, while combined acceptance and performance validation remain incomplete. No market-beating or deployable performance claim follows from those research tools.
Sources and context
Horvath, Issa & Muguruza, Market regime classification with signatures. External references provide methodological context; implementation statements are based on the application’s retained technical documentation reviewed September 6, 2026.
Product interpretation and limitations are described in the implementation overview and internal API overview.