According to the spec, "validation errors cannot be suppressed by points of uncertainty." However, the behavior of validation in this situation depends on the validation mode specified. If the mode of validation that waits until parsing is over is used, this is correct. However, if the "validate as you go" method is used, we only want the errors to come through for 2 scenarios:
1. the branch with the validation error parses successfully
2. all choice branches fail.
This is my understanding based on your e-mail. Please feel free to correct me if I've gotten it wrong or missed something.