The weekly loop
No hour budget, only a floor that keeps the pipeline alive when other work comes first.
The floor
- One
/scout-onlinerun a week. - At least two prospect contacts logged per running experiment, with
/log-experiment.
That is all. Everything else (shaping ideas, interviews, writing a landing page) happens when it can. The floor exists because discovery dies quietly: nobody decides to stop, the next session just never comes.
How the workspace keeps you honest
The brief at every session start prints:
Days since last signal: 9
Days since last log entry: 3
The workspace CLAUDE.md tells Claude to say so before anything else when either number passes 7. Experiments past their deadline appear under Overdue: until they are closed.
A 45-minute week
| When | What | Command |
|---|---|---|
| Monday, 20 min | Scout; keep the signals worth keeping | /scout-online (or with a focus) |
| Any day, 5 min each | Send or log two contacts per running experiment | /log-experiment exp-NNNN <note> |
| Friday, 15 min | Cluster new signals; close or extend experiments past their deadline; commit | python3 scripts/engine.py check, git commit |
When to shape an idea
The /scout-online digest suggests it when a cluster holds three or more signals with evidence 3 or more. Until then, a cluster of evidence-1 and evidence-2 signals is a lead for interviews, not an idea. The fastest way to raise evidence is a conversation: one call that confirms the pain and its cost is worth more than ten more forum threads.
When to kill
When a dated kill criterion fires, or when an experiment ends below its metric and no pivot is more promising than the next idea in line. Killing is the normal outcome; git log keeps the record, and a killed idea’s signals stay available for the next cluster.