After a campaign goes quiet, Deloosh writes itself one short, factual memory of what happened — grounded strictly in your sending ledger. It is not a report you read; it is how the advice you get next time gets better.
When a campaign “closes”
A campaign is not closed the moment the last email goes out. Deloosh closes it on a fixed clock started by the last send: once a quiet window — about seven days — has elapsed since the most recent email in that campaign, the campaign counts as closed.
It is worth being precise about this, because the intuitive reading is wrong. Deloosh is not waiting for all the activity to die down. It does not watch for the last bounce or the last unsubscribe to arrive and then decide things have gone quiet. It measures from the last send, and when the window is up the campaign is closed — whether or not anything is still trickling in. A late click does not reset the clock.
The retrospective then runs on a schedule, without being asked. It is one of the few things Deloosh does entirely on its own, and it is on that list because it neither sends anything nor changes anything you own. See TRUST-1, What Deloosh does on its own — and what it always asks first.
What the memory is allowed to say
The retrospective is ledger-only. It may record what your own sending records actually show, and it is checked against a gate that rejects the kind of language that smuggles in an unproven outcome — claims of lift, ROI, click-through rates, or “strong returns”.
If the content does not pass that gate, it is not written. Deloosh would rather hold no memory of a campaign than a flattering one.
The result is a memory that reads like a fact, not a press release: what was sent, to whom, what the ledger recorded afterwards, and what is worth carrying forward. No invented performance numbers, because Deloosh does not have the evidence to support them and will not manufacture it. See TRUST-2, Why the AI won’t make things up.
Opens are not part of the retrospective or the advice it supplies to future drafts. That evidence ban does not hide them: when the assistant surface is available, you may separately ask for a recorded open count, and the assistant will disclose it with the mandatory inaccuracy caveat. The count stays isolated from this memory and every downstream decision. See SEND-17, Why Deloosh doesn’t show open rates.
Where it shows up: in your next draft
Not as a dashboard. There is no retrospective report screen, no campaign scorecard, and no post-campaign email — and none is planned as part of this.
The memory has one consumer today, and it is a specific one: when Deloosh writes your next draft, the relevant past retrospective is pulled into the context the drafting model works from. Deloosh matches on the play, so a retrospective from a previous sunset campaign informs the next sunset draft.
That is the whole of it, and it is worth stating narrowly rather than generously. Retrospectives do not currently feed the Signals engine — a Signal does not know that this cohort was mailed last month. They do not stop the assistant from suggesting something you already tried. The compounding happens in the copy Deloosh writes for you, not yet in what it decides to raise.
Why remember at all
An email brain that forgets every campaign the moment it ends is a brain that writes you the same generic email forever. The retrospective is what lets the drafts Deloosh writes become specifically yours over time — which is also why it matters so much that the memory is honest. A brain that compounds invented outcomes gets confidently worse, and does it faster the longer you use it.