Either is correct, but it is much easier to test using residual plots.

Thanks a lot

In repeated measures ANOVA, the normality assumption applies to the residuals, not the raw outcome values at each time point.

You should check normality of the model residuals (e.g., using a Q-Q plot or Shapiro-Wilk test), not the data at each level of the repeated factor.

Checking raw scores can be informative, but it’s not required for meeting the assumption.