diff --git a/public/images/docs/evaluation/explore-playground/usage-analytics/usage-tab.png b/public/images/docs/evaluation/explore-playground/usage-analytics/usage-tab.png new file mode 100644 index 00000000..3c16373c Binary files /dev/null and b/public/images/docs/evaluation/explore-playground/usage-analytics/usage-tab.png differ diff --git a/src/pages/docs/evaluation/guides/explore-playground/usage-analytics.mdx b/src/pages/docs/evaluation/guides/explore-playground/usage-analytics.mdx new file mode 100644 index 00000000..6d9477ce --- /dev/null +++ b/src/pages/docs/evaluation/guides/explore-playground/usage-analytics.mdx @@ -0,0 +1,39 @@ +--- +title: "Usage & analytics" +description: "See how an eval has performed over time, and dig into any single run" +--- + +The **Usage** tab on an eval's page shows how the eval has performed over time, and lets you open any single run. A summary and chart sit at the top, and the full evaluation log runs below. + +## Open the eval + +From the **Evaluations** list, open the eval you want to inspect and switch to its **Usage** tab. + +The Evaluations list, searched for hallucination, with a prompt to select the eval you want to open +*From the Evaluations list, open an eval to reach its Usage tab* + +## Pick a time range + +A row of ranges at the top scopes everything below it: **30 mins**, **6 hrs**, **Today**, **Yesterday**, **7D**, **30D**, **3M**, **6M**, and **12M**, or **Custom** for your own dates. Narrow it to a spike you are investigating, or widen it to watch the long trend. + +The eval's Usage tab with a time-range selector, summary stats, a volume and completion-rate chart, and searchable evaluation logs below +*The Usage tab: pick a time range, read the summary and chart, then dig into any run in the evaluation logs* + +## The summary and chart + +For the chosen range you get the headline numbers, **Runs**, **Success**, **Errors**, and **Task Completion Rate**, alongside a chart that plots **Volume**, how many runs, against the **Task Completion Rate** over time. A rate that dips or a volume that drops is your cue to look closer. + +## Evaluation logs + +Below the chart, **Evaluation Logs** lists every run the eval scored, one row each: its **Score** and **Result** (passed or failed), the **Input** it saw, the **Reason** for the verdict, the **Source** it ran under (**Playground**, **Dataset**, or **Tracer**, or the name of the project), and when it **Ran**. Use the **Search** box to find a run by a free-text match across its id, input, result, and reason. + +## Dive deeper + + + + The library of evals you can browse and open + + + Attach an eval to your data so it starts recording usage + +