# Feedback

> Human ratings of the agents' analyses by pillar, model and week, the finished incidents waiting for a score, and what the aggregate counts before a quality review.

_Reference · Updated 2026-10-17 · Describes release 0.1.164_

## Feedback

**Agent quality → Feedback** is available with `report.view`. Submitting or changing a rating on an investigation needs `feedback.submit`; these are separate permissions.


### What the Feedback page shows

![Feedback: RCA scores by pillar, by model and by week, and the finished incidents still waiting for a rating.](/help/media/portal/portal-feedback.png?v=0.1.132)

Scoring itself happens on an incident, not here — see the **Incidents** chapter. This page is the aggregate and the worklist:

| Block | What it shows |
| --- | --- |
| `Scores (N scored)` | How many scored feedback rows are in the window. With none it reads `0 scored — rate an RCA on an incident`. |
| **By pillar** | Average score and count per pillar. |
| **By model** | Average score and count per model id, with unscored-model rows falling under `mock`. |
| **Weekly trend (week of)** | Average score and count per week, keyed by the Monday of that week, oldest first. |
| Finished incidents to rate | Every `resolved` or `escalated` incident in the window, newest opened first, linking to the incident so you can score it. |

Averages are tinted: 7 and above reads as good, 4 to 7 as a warning, below 4 as bad.

### Two things to know before you run a quality review

- **One score per investigation per user, and re-scoring overwrites.** The write is an upsert keyed on the investigation and the person. Changing your mind replaces your earlier score; it does not add a second vote. Two people scoring the same investigation do count twice.
- **The two left-hand tables count less than the total does.** The page reads a window — the most recent 500 feedback rows and the most recent 500 investigation summaries — and joins them. **By pillar** and **By model** can only count a score whose investigation is in that window, so on a high-volume tenant they under-count older feedback while the weekly trend still shows it. The same join is why a score you left on an estate x-ray or a chat turn appears in the total and the weekly trend but never in **By pillar** or **By model**: those runs are deliberately excluded from the investigation feed.

## Troubleshooting

Symptoms specific to this page. Sign-in problems, missing nav entries in general, stuck or queued investigations, and `$0.0000` costs are in the **Troubleshooting** chapter.

| Symptom | Cause | What to do |
| --- | --- | --- |
| **Feedback** cannot load | Your report access changed or the request failed | Check the displayed error, retry and ask an owner for `report.view` if access is missing. |
| A score you gave is missing from **By pillar** or **By model** | It was left on an x-ray or chat run, or its investigation fell out of the 500-row window | Expected. The total and the weekly trend still count it. |

## Next steps

- [Knowledge base](/help/knowledge-base.md) — a correction waits in the Review queue until an approver accepts it.
- [Benchmarks and scores](/help/benchmarks-and-scores.md) — Min human score filters a back-test on these ratings.
- [Incidents](/help/incidents.md) — rate an analysis on the incident itself.
- [Knowledge, x-ray and insights](/help/intelligence.md) — the hub for every page of this group.