## Summary Rather than contextual overlap, splitting on H2s (but keeping the H2 itself of course) can be useful in many cases when content editor discipline is good. The chunks can then be less disjointed ## Problem Contextual overlap is a fairly rough way of doing it, making the chunks less naturally split on topic ## Proposed solution *(optional)* Split on H2s; first detect if multiple H2s exist in the body, and if so, whether the chunks are similar in size