Know where your examples go.
When you use the Playground
Playground examples are supplied by JevBench. Their Jev scores were recorded in advance. Opening a scenario, viewing mistakes, and moving the decision line do not send a new request to Jev or OpenRouter.
When you explore the demo
The example contains fictional business inputs and illustrative scores. Exploring thresholds does not send rows to Jev. Metrics are calculated in your browser.
When owner-only testing is enabled
The own-data Workbench is currently closed to public visitors. An authorized owner can prepare a test; CSV files are parsed in the browser and the original file is not uploaded.
Expected human labels are used for analysis and are never sent to Jev. OpenRouter and the provider’s data policies apply to live requests. The provider API key remains on the server.
Product analytics
We use PostHog to measure which pages and recorded tests people explore, whether they inspect mistakes, and whether they try Builder. Events contain counts, cutoff values, and scenario identifiers. They do not contain your question, examples, CSV, or result text. Public visitors are measured with anonymous browser identifiers.
Session replay
PostHog may record a replay of page layout, clicks, and scrolling so we can understand where the Playground is confusing. Input values are masked. The own-data Workbench and its results are blocked from replay, including questions, examples, CSV input, and provider responses.
Provider policy
JevBench is an independent evaluation tool and is not affiliated with or endorsed by TypeSafe AI.