03 · Score after close

The market closes, then the forecast gets scored.

The target-date forecast remains frozen until the realised Nasdaq Composite close becomes available. Scoring adds the outcome to the record; it does not rewrite the prediction.

The actual close belongs to the scoring stage, not the forecasting stage.

Before openForecast locked
After closeActual observed

Keeping those stages separate is what makes the prospective record meaningful: the answer is not available when the prediction is created.

How far away was the forecast?

\[AE_t=\left|\widehat S_t-S_t\right|\]

Absolute error is measured in Nasdaq Composite index points. Lower is better.

\[Error\%_t=\frac{\left|\widehat S_t-S_t\right|}{S_t}\times100\]

Percentage error provides a scale-free interpretation of the same miss.

The same chronology every session.

01Forecast exists

The target-date prediction is already locked.

02Market closes

The realised Nasdaq Composite close becomes known.

03Join records

Match forecast and actual on the same target date.

04Calculate error

Record absolute and percentage error.

05Update evidence

Add the scored date to the prospective history.

One forecast tells us very little. A sequence is more useful.

Each new forecast is generated under the same pre-specified rules. As that common sample grows, Graham Says can compare models using mean absolute closing-level error without pretending historical backtests were live forecasts.

Historical model selection and prospective model evaluation answer different questions. The site keeps those records separate.