# Lumi AI Improvement Center

Open **Community > AI Improvement Center** to process feedback about Lumi's
answers.

## Normal review

1. Read the user's question, Lumi's answer, and any suggested improvement.
2. Choose **Apply improvement**, **No change needed**, or **Dismiss feedback**.
3. Applied improvements are activated immediately. Every finished record moves
   out of the active queue automatically.

**Apply improvement** asks for the answer or behavior Lumi should use for
similar requests. Audience and activation are the only normal settings. More
technical targets, route scopes, links, and evaluation options remain under
**Advanced options**.

## Finished feedback

Choose **Search finished feedback** to open the history dialog. It can search
questions, answers, tags, notes, implementation details, and outcomes. Admins
can restore a record to the active queue, edit its retained details, or
permanently delete it through Lumi's timed confirmation.

The page subscribes to Lumi's authenticated event stream. New feedback and
review actions from another administrator or trusted reviewer update the queue
without a full-page refresh and reconnect automatically after a temporary
connection loss.

Review actions use an authenticated JSON response even when Lumi is behind a
reverse proxy. If a save cannot complete, the page distinguishes an expired
session, insufficient access, a feedback record changed elsewhere, and a
server-side storage problem instead of showing a generic failure.
Lumi resolves the authored form endpoint directly, so fields named `action`
cannot replace the destination URL in the browser.

## Advanced tools

The collapsed **Advanced tools and access** section contains moderator access,
manual correction-bank maintenance, evaluation cases, and training exports.
These controls are not required for ordinary feedback processing.