Additional examples are adjusted to the entries in an automated way - we cannot guarantee that they are correct.
This error is called a chance error (see Errors and residuals in statistics).
We can reduce chance errors by taking a larger sample, and we can avoid some bias by choosing randomly.
Jacob Wolfowitz, The coding of messages subject to chance errors, Illinois J. Math., vol.
One possible explanation for this is that because absolute pitch is subject to chance error, occasionally incorrectly inferred note names could compete with the pitch-induced color on particular trials.