READER EDITION / ISSUE 002 / Revised 2026-11-20
Creative Agentic Computing
CodeFarmer / Issue 002
RESEARCH RELAY / YOUR FIRST RUN

Follow five
visible handoffs.

1 / Ask the planner to choose notes

Click Run next step once. Expect WORK, 1 / 5 calls, and a planner event selecting S1, S2 and S3. Those are labels for the supplied sources. Refresh the browser: the same progress should return from its saved checkpoint.

2 / Collect three quoted observations

Click Run next step three more times. Each worker step adds one claim and a quotation from a chosen note. A claim is what the reply says; its quote supplies the words you can check. After these steps, expect REVIEW, 4 / 5 calls and three evidence notes.

3 / Read the review, then choose approval

Click Run next step again. The prepared reviewer leads to READY, 5 / 5 calls. Check every claim against the question and its source. A quotation appearing in a note proves those words were present; you still judge whether the claim follows from them and whether the source is reliable.

4 / Keep the result

Click Approve local report, then Save Markdown report. Open research-relay-report.md in a text editor. Markdown is readable text with simple heading marks. It records Mode: scripted and model: none (scripted fixture). Run remaining steps is a shortcut through the remaining jobs.

We verified approval, report download, reload and Python restart. Prepared replies teach the sequence; real-model replies need their own inspection.