Reports and delivery
Each kind of job has its own report shape. The server, not the model, decides whether the thing you asked for was delivered, and says so plainly.
Updated 2026-10-13
Report shapes
Reports use short paragraphs or bullets that start with a bold lead-in. Headings appear only when there are two or more sections, and tables only for real comparisons. “Done” alone is never an acceptable reply.
| Kind | What the report contains |
|---|---|
| Research | About 150 to 350 words. The answer first. Then the evidence (a compact table if items were compared, otherwise bullets with source markers). Then Confidence and gaps, then What I would do, then what was produced. |
| Build | About 100 to 250 words. What now works, with the live link or file. Then How I checked it: the commands run and what they showed. Then what changed, and what is left. |
| Data | About 120 to 300 words. The answer with the number. Then How sure: sample size, data quality, assumptions, ranges. Then two to four findings. Then the charts, workbook and script, and the one thing to do about it. |
| Writing | The piece lives in a file or Library item. The chat report is about 80 to 200 words: who it is for, the angle, anything assumed or unproven, and what to test first. |
| Outreach | About 100 to 220 words: how many leads or emails, the signal behind the best ones, what is waiting for your approval versus what has actually been sent, any guessed email address, and when to follow up. |
| Ops | About 60 to 180 words: what is set up and where it lives, what is open with an owner, and the next date or trigger. |
The server decides what was delivered
When a run ends, the server weighs the evidence: the files that were archived, the tool steps, the approvals and the answer. Only archived deliverables count. Working files never do. The format and the read-back counts must match what was asked. A blank answer is a failure. Code pasted into the answer is not a file.
| Verdict | Meaning |
|---|---|
| Delivered | The thing you asked for exists and passed the checks. |
| Partial | Some of it was made. |
| Answered | You asked a question, and it was answered. |
| No deliverable | It was asked to make something and did not. |
| Delegated | The work was handed to someone else, who will report. |
| Failed | The run did not finish. |
When something is missing, the reason is one of: out of steps, no file, wrong format, empty file, a tool failed, not archived, a blank answer, or the budget ran out. A missing file is said first in the answer.
What it was expected to deliver
The expected result is worked out in this order: your choice in the hand-off dialog, then the task’s, then the brief’s, then rules that read the brief (a deck, a sheet, a document, N leads, a site, an email sent, code, or an answer).
| Hand-off chip | Expected | Minimum check |
|---|---|---|
| Deck | A .pptx | At least 3 slides |
| Sheet | A .xlsx or .csv | At least 1 row |
| Doc | A .docx, .pdf or .md | At least 50 words |
| Site | A live site | It is published |
| Answer | An answer in chat | Not blank |
Previews
Results come with a preview you can open: documents, PDFs, slides, spreadsheets, tables, images and HTML pages. Files also show who made them, their version and size. At most one full preview is shown per message.