BRIEFLOOP / GUIDES

Feedback and reuse

Turn specific revisions into inspectable, reversible experience.

Current installers: Mac 0.22.2, Windows 0.22.2. Labels and capabilities vary by channel; the installer version is authoritative. Development features are not implied.

Before you start

Use a workspace you can back up. Model tasks require an authenticated runtime and permission to process the chosen materials.

Steps

1. Explain the desired writing behavior or reason for a correction, including when it applies, rather than leaving only a score.
2. Choose whether to enable automatic learning in settings and set the maximum rounds K. Additional rounds can increase model usage and time.
3. Inspect Wiki experience and proposed skills. The Wiki holds method knowledge, not current-period evidence.
4. Inspect validation results and adoption status. A candidate file does not mean it is active, and a running task does not demonstrate a benefit.
5. If adoption is unsuitable, use the rollback control available in your version and confirm the active skill state. Preserve original feedback and comparison records.

What success looks like

You should be able to distinguish feedback, candidates, validation and active content. Feedback can become experience evaluated for adoption; scores are not guaranteed to improve every round.

If something goes wrong

When paused or out of quota, keep records and follow recovery instructions. Policies for explicit human requirements versus automated scoring suggestions vary by version; do not assume a development policy exists in an older installer.

← All guides · Installation options →