10 February 2026 · method, audit

Backtesting before you trust a model

A short discipline for checking whether your forecasting process would have helped last quarter.

Notebook and pen beside planning documents

Before adopting a new forecasting method, replay it against the last two planning cycles. Would the ranges have covered reality? Would the assumptions have been updated in time?

Backtests do not prove the future. They expose whether your inputs, definitions, and review cadence are honest. Many “broken models” are actually broken event definitions or delayed data.

Document the backtest in one page: method, window, error pattern, and what you will change. Keep that page next to the live forecast so trust has a paper trail.

Back to the journal · Discuss a forecast review