I think it would be in everyone's interests to require all publicly traded companies to put out 4 press releases during the year (one per quarter) with an obviously ludicrous claim. Either the algorithms would get better or they would go away.
Or the folks running the algorithms could put April 1st days in their test corpus?
I imagine fullwedgewhale was picturing them as being randomly distributed, not all on the same day every quarter.
And I imagine lallysingh meant that including April 1 headlines in the test corpus could help train the algorithms to correctly interpret false headlines, whether those headlines occur on a single day or not.