Both drafts from one week, each copyable as text or as markdown. There is no send button, and there is not going to be one.
The problem
Two updates a week, from the same seven days, for audiences that need different things: the team wants what shipped and what is blocked, the board wants bets and outcomes. Neither of them wants the raw tracker.
So the work is translation, it happens on a Sunday night, and the first thing to go when the week is bad is the update explaining why the week was bad.
What it does
Paste
The raw week — tickets, commits, whatever the tracker gives you.
Answer
The handful of things the week cannot tell it. It asks rather than guessing.
Draft
Both updates, from the same source, in the register each audience reads.
Keep
A continuity check against last week: what you promised, and whether it appeared.
What it showed
Audiences from one week
2
Team and investor, drafted from the same fifteen items, in different registers.
Things it will send
0
By design. A person looking at the update before it goes is the point of having one.
The interesting constraint turned out to be refusal. The draft is only trustworthy if it declines to invent the parts you left out, so the tool asks its questions first and leaves gaps where the answers are missing rather than writing something plausible into them.
The continuity check does the other half: it reads last week’s update back and asks whether what you promised then appeared this week. That is the part a human drafter skips, and it is the part the reader remembers.
What it cannot do
From the tool’s own panel: the spec calls for read-only OAuth into the tracker, a weekly job and an email on the day you choose. This runs in the browser, so the week arrives by paste and the schedule is a preference rather than a cron.
What is real: the raw week, the questions, both drafts, the continuity check, and the rule that nothing is ever sent.
The long version of the argument behind it is The 2026 product leader’s toolset.