All docs
See what's happeningOverviewdashboard

Overview

The project front page: how answers are doing, the top issues, alerts, the live experiment and recent policy changes.

Overview is the project's front page. It answers two questions: how are my AI's answers doing, and where should I look first? Open it from Project › Overview.

Read it top to bottom

OverviewOverview

  1. Pick the range: Last 24 hours, Last 7 days (the default) or Last 30 days.
  2. Start at Needs attention: the share of answers where at least one check failed. A check is a yes/no test, such as "every claim is backed by the documents" (see Glossary).
  3. Read Top issues · last 24 h: the most frequent reasons answers needed attention. All issues → opens Issues, with where to fix each one.
  4. If an alert is firing, its card links to Open failing answers and Alert events →.
  5. If a number moved, check Recent policy changes. A newly published version of the policy can explain it: every version is listed with its change note, and History → opens them all.

What else is on the page

Overview, lower halfOverview, lower half

CardShows
Score trendAnswers per day and the mean overall score, so you can see a drop the day it happened
Mean overall, Mean faithfulnessThe averages for the range
CategoriesThe six categories (Grounding, Safety, Semantic, Coherence, Quality, Depth) and their means. Weights → opens the category weights
How claims were judgedHow many claims the standard checks settled, and how many went to the reasoning model under Max-Accuracy
Max-AccuracyClaims checked by the reasoning model, Events with at least one such claim, and any Claims where the reasoning model failed (standard result kept)
Live experimentThe experiment splitting live traffic now, and the variant leading
Recent policy changesThe latest policy versions, In force or Draft

A worked example

On Monday Needs attention is 4%. On Wednesday it is 11%. Top issues · last 24 h shows "Right version of the source" as the top issue, and Recent policy changes shows nothing new. So the scoring didn't change; the answers did. All issues → shows the fix belongs to retrieval: the search started returning the 2025 benefits guide for 2026 questions. You fix the search filter, and Needs attention drops back the next day.

Empty states

You seeIt meansWhat to do
No answers yet — send your first answerNothing has been logged in this projectConnect your app, or see Connect your app
No events in this range.Nothing was logged in the rangeWiden the range, or check your app is still sending
No failing checks in the last 24 h.Every answer passed its checksNothing
No experiment is running.No experiment is splitting trafficNew experiment, if you want to compare versions
No claims were checked in this range.Answers arrived without documents, so there were no claims to checkSend reference_materials with each answer