> ## Documentation Index > Fetch the complete documentation index at: https://docs.aide.app/llms.txt > Use this file to discover all available pages before exploring further. # Quality: see whether Aide's replies meet the bar > The Quality page says whether the Aide AI Agent's replies are good enough, as a pass or fail against a fixed bar, with the numbers behind it. The Quality page says whether the Aide AI Agent's replies are good enough, as a pass or fail against a fixed bar, with the numbers behind it. Open **AI Agent > Quality** in the sidebar. Owners and admins see it. ## How the page is laid out Pick an agent at the top of the left column, then one of the pages under it: **Overview**, **Conversations**, **Issues**, **Evals** and **Learned changes**. This article covers the Overview. See [Evals: the tests Aide replays every night](https://docs.aide.app/aide-ai-agent/evals.md) and [What Aide learns every night](https://docs.aide.app/aide-ai-agent/nightly-learning.md) for the others. Use the period menu at the top of the page to switch between **Last 7 days** and **Last 28 days**. The Overview opens on the last 28 days. ## What gets reviewed Every night, Aide reviews a random sample of its drafts against the replies your team sent. It also reviews drafts that stand out: one your team rewrote or discarded, a handoff, a gap in your knowledge. A draft is reviewed once your team has replied, up to four days later, so a weekend's drafts are counted when they're answered. Each reviewed reply is acceptable or not; there is no partial credit. When a teammate gives a verdict on a draft, it replaces the judge's (see [Mark Aide's draft as good or wrong](https://docs.aide.app/aide-ai-agent/review-a-draft.md)). Until the first reviews are in, the page says "No reviewed replies yet". ## The headline The headline counts only the random sample, so it is an estimate for all of Aide's replies. * **Meets the quality bar**: at least 90% of the reviewed replies were acceptable, over at least 100 reviews in the period. * **Below the bar**: 100 reviews or more, with under 90% acceptable. The line below says how many more acceptable replies would reach the bar. * **Not enough reviews yet**: fewer than 100 replies were reviewed in the period, so a few reviews can't swing the result. ## The four numbers Each shows the change from the previous period. * **Acceptable replies**: the share of the random sample judged acceptable, with how many were reviewed and how low the true share could plausibly be. **See replies** opens them. * **Drafts sent**: of the reviewed drafts whose outcome is known, the share your team sent as is or with minor edits, with first replies and follow-ups shown apart. A draft in a conversation your team closed without replying is "not needed" and is left out of the rate. A draft Aide wrote when it handed off counts like any other. * **Resolved without a person**: conversations Aide resolved without a teammate stepping in, for the whole workspace, as Analytics > AI performance counts them. * **Eval pass rate**: the share of eval cases the latest run in the period passed. Cases that need knowledge are shown beside it and are not in the rate. ## Compared with your team Aide also grades a sample of your teammates' replies on the same criteria. **Compared with your team** shows Aide, your team's median and your top quarter of teammates, overall or **By criterion**, on topics both have replied to. Team figures appear once at least 2 teammates have 5 graded replies each. Until then, the page says what is missing. **Who replied** splits graded conversations into **AI only**, **Hybrid** and **Teammate only**. **Team** lists each teammate with Aide in a **Leaderboard**. Teammates' scores are a benchmark for Aide, not a performance review. ## Topics: draft or send One row per topic says whether Aide's drafts on it are good enough for Aide to send replies itself. Over the last 30 days, a drafting topic reads **Ready to send** when all of these hold: * Aide has at least 15 graded drafts on the topic. * Aide's score, at the low end of its range, is at or above your team's median on the topic. * At least 90% of its drafts were sent as is or with minor edits. * Your team agrees with the judge often enough for its grades to be trusted. * The topic's eval cases pass 3 times in a row. * No missing knowledge is open on the topic. Otherwise the row reads **Below your bar**, **Needs knowledge** or **Not enough data**; open it for the checklist of what is unmet. **Turn on sending** sets the topic so Aide sends its replies, after a confirmation. **Back to drafts** reverses it. Aide never switches a topic on its own. For the levels themselves, see [Control what Aide is allowed to do for each topic](https://docs.aide.app/aide-ai-agent/topic-ai-levels.md). ## Judge trust, trend and issues * **Judge trust**: how often the judge agrees with your team's own reviews. It shows once your team has reviewed 30 replies, with at least one marked acceptable and one not. * **Daily trend**: the share of reviewed replies judged acceptable each day (UTC), against the 90% bar. * **Top issues**: what the unacceptable replies have in common, grouped into themes, most frequent first. **All issues** opens the full list, with **Missing knowledge** beside it (see [Find and fix gaps in your knowledge](https://docs.aide.app/knowledge/knowledge-gaps.md)). ## What it costs Reviewing and grading use Aide's own allowance. They don't count towards your AI replies or your daily AI usage limit. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.