{"version":1,"pages":[{"id":"CtIKzEVGgQt1gZeaiIhA","title":"Introduction","pathname":"/","siteSpaceId":"sitesp_QcEbZ","description":"","breadcrumbs":[{"label":"Getting Started"}]},{"id":"1sESioz3Txhj0T3gbj12","title":"Glossary / Core Concepts","pathname":"/getting-started/glossary-and-core-concepts","siteSpaceId":"sitesp_QcEbZ","description":"Core terminology for three.dev — use cases, sessions, experiments, assessments, prompts, and quality metrics.","breadcrumbs":[{"label":"Getting Started"}]},{"id":"KrnaPb8gyoJmk9greNEx","title":"Quickstart: Record your requests","pathname":"/getting-started/quickstart-record-requests","siteSpaceId":"sitesp_QcEbZ","description":"Set up the three.dev proxy to record your LLM requests for observability and experimentation.","breadcrumbs":[{"label":"Getting Started"}]},{"id":"tQb0QAHhwazb3iCAjj1p","title":"Quickstart: Run an offline experiment","pathname":"/getting-started/quickstart-run-offline-experiment","siteSpaceId":"sitesp_QcEbZ","description":"Create your first offline experiment — replay historical requests, review AI Judge scores, and assess a sample to reach statistical confidence.","breadcrumbs":[{"label":"Getting Started"}]},{"id":"5y74e8YRCoBPLE7z27Yy","title":"Quickstart: Run a live experiment","pathname":"/getting-started/quickstart-run-live-experiment","siteSpaceId":"sitesp_QcEbZ","description":"Run a live experiment to test AI configuration changes on production traffic with statistical rigor.","breadcrumbs":[{"label":"Getting Started"}]},{"id":"eWIlFzB6dukprv6HIq2E","title":"Overview","pathname":"/sending-requests/sending-requests","siteSpaceId":"sitesp_QcEbZ","description":"Route your LLM requests through the three.dev proxy for observability and experimentation.","breadcrumbs":[{"label":"Sending Requests"}]},{"id":"eSVyUHtGZZNdZc99D0y0","title":"AI Providers Integration","pathname":"/sending-requests/ai-providers-integration","siteSpaceId":"sitesp_QcEbZ","description":"Configure the three.dev proxy for OpenAI, Anthropic, Gemini, Azure OpenAI, and LiteLLM — with code examples per provider and language.","breadcrumbs":[{"label":"Sending Requests"}]},{"id":"Ru5GF59WXzYbNXb0JQpL","title":"Supported paths","pathname":"/sending-requests/supported-paths","siteSpaceId":"sitesp_QcEbZ","description":"Advanced observability for Chat Completions and Messages APIs, session grouping, and request-level tags.","breadcrumbs":[{"label":"Sending Requests"}]},{"id":"iFyXAKFQj3qlkiapNfNS","title":"Overview","pathname":"/offline-experiments/offline-experiments","siteSpaceId":"sitesp_QcEbZ","description":"Test model and prompt changes offline — replay historical traffic through variants and score outputs against Acceptable Response before touching production.","breadcrumbs":[{"label":"Offline Experiments"}]},{"id":"3qPtPko3z0qN5uCZ59me","title":"Creating experiments","pathname":"/offline-experiments/creating-experiments","siteSpaceId":"sitesp_QcEbZ","description":"Configure an offline experiment — pick variants, set the dataset size, and launch the replay.","breadcrumbs":[{"label":"Offline Experiments"}]},{"id":"cwiGG4gisGySfPIp1YTp","title":"Quality Metric","pathname":"/offline-experiments/quality-metric","siteSpaceId":"sitesp_QcEbZ","description":"Offline experiments measure quality through Acceptable Response — domain experts and an AI Judge assess whether each replayed output is good enough to show to end users.","breadcrumbs":[{"label":"Offline Experiments"}]},{"id":"dF65p7hYsdQsZ1hu9OsG","title":"Assessments","pathname":"/offline-experiments/assessments","siteSpaceId":"sitesp_QcEbZ","description":"Assessments are pass/fail judgments on replayed requests — from the AI Judge (automated) or domain experts (human). Together they establish offline experiment quality.","breadcrumbs":[{"label":"Offline Experiments"}]},{"id":"xFXzsE5AQ6gdZv8HhMRA","title":"Understanding results","pathname":"/offline-experiments/understanding-results","siteSpaceId":"sitesp_QcEbZ","description":"Read offline experiment results — Observed and Statistical stages, variant performance badges, and what to do when results aren't conclusive.","breadcrumbs":[{"label":"Offline Experiments"}]},{"id":"TyJKajbBcwBQg2McGoYE","title":"Example: Model comparison","pathname":"/offline-experiments/example-model-comparison","siteSpaceId":"sitesp_QcEbZ","description":"Worked example — compare GPT-5.5 against GPT-5.2 on a historical set of hotel-booking requests, using Acceptable Response as the quality signal.","breadcrumbs":[{"label":"Offline Experiments"}]},{"id":"MJi2TPcfyxfVSPmXpTV1","title":"Example: Prompt comparison","pathname":"/offline-experiments/example-prompt-comparison","siteSpaceId":"sitesp_QcEbZ","description":"Worked example — compare two prompt versions on a single-turn listing-description generator, using Acceptable Response as the quality signal.","breadcrumbs":[{"label":"Offline Experiments"}]},{"id":"6Bvf8qL5lHV82eCMZv9e","title":"Overview","pathname":"/live-experiments/live-experiments","siteSpaceId":"sitesp_QcEbZ","description":"Test AI configuration changes on production traffic with session-level variant assignment, event-based quality metrics, and dynamic traffic allocation.","breadcrumbs":[{"label":"Live Experiments"}]},{"id":"yVOBP6L8F7zUL65u5ALo","title":"Creating experiments","pathname":"/live-experiments/creating-experiments","siteSpaceId":"sitesp_QcEbZ","description":"Create a live experiment — define variants, integrate assignment, start the test, and manage the lifecycle.","breadcrumbs":[{"label":"Live Experiments"}]},{"id":"Hy2B34NukOfoYxGGPRWQ","title":"Quality Metric","pathname":"/live-experiments/quality-metric","siteSpaceId":"sitesp_QcEbZ","description":"Live experiments measure quality through an event-based binary metric your application reports — a booking, a click, a resolution — attached to a session.","breadcrumbs":[{"label":"Live Experiments"}]},{"id":"GfFAmQwqy1Xi9DlgwrnM","title":"Understanding results","pathname":"/live-experiments/understanding-results","siteSpaceId":"sitesp_QcEbZ","description":"Read live experiment results — decision summary, variant performance table, traffic split history, and latency and cost analysis.","breadcrumbs":[{"label":"Live Experiments"}]},{"id":"TmLzqvtpu4oJXjlkD9bK","title":"Example: Reducing cost","pathname":"/live-experiments/example-reducing-cost","siteSpaceId":"sitesp_QcEbZ","description":"Worked example — use a live experiment to test whether a cheaper model can match your current quality.","breadcrumbs":[{"label":"Live Experiments"}]},{"id":"BUdJkyCH632zLXlVelt7","title":"Example: Changing the system prompt","pathname":"/live-experiments/example-changing-prompt","siteSpaceId":"sitesp_QcEbZ","description":"Worked example — use a live experiment to test which system prompt maximizes your quality metric.","breadcrumbs":[{"label":"Live Experiments"}]},{"id":"UY0HgO8vf98zRwWubLqy","title":"Overview","pathname":"/prompts/prompts","siteSpaceId":"sitesp_QcEbZ","description":"Store, version, and manage prompts in three.dev — test prompt changes in offline experiments without deploying code.","breadcrumbs":[{"label":"Prompts"}]},{"id":"nGRZD57UIUmhdXcALDWO","title":"Creating and versioning prompts","pathname":"/prompts/creating-and-versioning","siteSpaceId":"sitesp_QcEbZ","description":"Create prompts with message templates and variables, version them through draft to published, and activate the version three.dev renders by default.","breadcrumbs":[{"label":"Prompts"}]},{"id":"gzyGLPdYyJzV2DTFHr1p","title":"Overview","pathname":"/ai-judge/ai-judges","siteSpaceId":"sitesp_QcEbZ","description":"An AI Judge scores every replayed request in an offline experiment against the Acceptable Response quality metric. Each use case has exactly one Active judge.","breadcrumbs":[{"label":"AI Judge"}]},{"id":"3v2dbnBTY609qiTNfcgf","title":"Creating and managing your AI Judge","pathname":"/ai-judge/creating-and-managing","siteSpaceId":"sitesp_QcEbZ","description":"Create an AI Judge for your use case, set it as active so future offline experiments use it, and iterate on the AI Judge configuration with Edit a copy.","breadcrumbs":[{"label":"AI Judge"}]},{"id":"Kf9IxVeTsXmX71eryBaE","title":"Overview","pathname":"/api-reference/api-reference","siteSpaceId":"sitesp_QcEbZ","description":"Public API endpoints for three.dev — authentication, metric reporting, variant assignment, and prompt rendering.","breadcrumbs":[{"label":"API Reference"}]},{"id":"Xa7lHI7zUnpOIH3thdc1","title":"Report Metric","pathname":"/api-reference/report-metric","siteSpaceId":"sitesp_QcEbZ","description":"Report a binary quality metric outcome for a session.","breadcrumbs":[{"label":"API Reference"}]},{"id":"cA7x0dyqCIPNsfH94xAe","title":"Assign Session","pathname":"/api-reference/assign-session","siteSpaceId":"sitesp_QcEbZ","description":"Assign a session to an experiment variant for a running live experiment.","breadcrumbs":[{"label":"API Reference"}]},{"id":"LUDyiqHHTjZYTFUnnz7Z","title":"Render Prompt","pathname":"/api-reference/render-prompt","siteSpaceId":"sitesp_QcEbZ","description":"Render a prompt by substituting variables into the active (or pinned) published version.","breadcrumbs":[{"label":"API Reference"}]},{"id":"8WcgA9PWaywvdYqt9o1Y","title":"Overview","pathname":"/for-ai-agents/for-ai-agents","siteSpaceId":"sitesp_QcEbZ","description":"Machine-readable entry points for AI agents consuming three.dev documentation.","breadcrumbs":[{"label":"For AI Agents"}]}]}