← LevelsBuilder · B3
The human-in-the-loop matrix
Build-along · ~10 min at your own keyboard
Two questions decide whether a human stays in the loop: how bad is the outcome if it's wrong, and can you undo it? A label you can change in a second is different from an invoice a client has already read. Draft-for-review and send-automatically are different products. Anything a client or the public reads stays draft-for-review unless you've tested it and decided otherwise on purpose.
The long version: Human in the loop ↗
Do this
0/4Checkpoints
- ▸You should now see 6 real tasks listed from your week.
- ▸You should now see each task with a damage score and a sensitivity score.
- ▸You should now see your three buckets and a review checklist for one task.
The three ways this goes wrong
- Everything scores a 3 and you can't say why. — Write the worst realistic mistake for each task in one line and score that. If two tasks really carry the same risk, the same score is the right answer.
- Customer-facing tasks land in full-auto. — Anything a client reads goes to draft-for-review until it has run clean many times.
- The review checklist is vague ('check it looks right'). — Write checks you can answer yes or no: right client, right amount, right date.
Share what happened
Paste your 6-task matrix with buckets and your review checklist for the one task.
Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].
🔒 Don't paste confidential, personal-health, or payment info.