max / goingson

The email body column says which of its two formats it holds emails.body has always held one of two things, decided at sync by extract_body_with_html: a text/plain part arrives as it was sent, and anything else goes through pter::convert, which produces markdown. One column, two formats, no flag, so both readers guessed and both guessed wrong in opposite directions. utils.js:formatEmailBody escaped the body and re-linked bare URLs with two regexes, so pter's markdown reached the reader as literal ** and [](), and the quasi port carried Node::rich unconditionally, so a plain message whose asterisks looked like markdown rendered as markdown. 066 adds body_format, TEXT and not a bool for the reason tz_kind is: a body kept as raw HTML is a plausible third case and takes a value where a flag takes a rename. It is written by the one function that knows which branch it took, threaded through ParsedEmail, FetchedEmail and NewEmailWithTracking, and round-tripped by restore, which carries durable content. Existing rows default to plain, which is the conservative direction rather than the correct one. They cannot be reclassified without re-parsing source messages the database does not keep, and the rows that came from HTML are exactly the ones that read wrong today. Defaulting the other way would start rendering every genuinely-plain body's asterisks as emphasis: a new bug applied to the majority to retroactively fix the minority. They age out as folders resync. Both readers now read it. The quasi thread picks Node::rich or Node::text; EmailResponse gains bodyRendered, docengine::render_strict and present only for markdown, and formatEmailBody keeps the plain path where its escaping, quote collapsing and linkifying are still what is wanted. Strict rather than the standard a task description gets, because this markdown came from outside the app. Email::body_preview is format-aware too, flattening with render_plain before it cuts so the budget is spent on what the reader sees rather than on a URL, and EmailResponse calls it instead of slicing 100 characters of its own. That duplicate was the same two-readers-guessing shape as the defect itself. The projects card takes Prose::rich, so its description travels as the markdown it is instead of as a pre-flattened string.