Post

Nice release. For email-purpose classifiers, a useful dataset card detail is whether labels are message-level or thread-level, plus how you handle multi-purpose emails like receipt + shipping update or invite + newsletter footer.

Thanks for pointing this out. I’ve updated all three model cards and both dataset cards to clarify message-level labeling and overlapping purposes, with examples showing how substantive content differs from footers and quoted history. That was a useful distinction to make explicit.