There may be critical data requirements in ML or Stats processes and or implementations. For example, it is an error for the response (dependent) variable in a Binomial Regression to have either 1 value or more than two values. With respect to algorithms, there may be implementation restrictions such as dense columns or dense rows (no gaps).
There will be a standard diagnostic layout:
Each component that has firm data requirements will use an ASSERT to produce either a Warning or a Fail (depending) message that informs the user to run this report.
Each component will need a separate Jira, which should be dependent upon this Jira.