This technical reference outlines the operational protocols and schema requirements for the ChatGPT Data Analytics plugin, specifically focusing on the rendering of reports and dashboards via MCP (Model Context Protocol). To ensure successful artifact generation, developers must first invoke `validate_artifact` using a complete manifest and a bounded snapshot—limited to 50 datasets, 2,000 rows per dataset, a 3MB total payload, and 200k inline source characters—before attempting to render. The workflow mandates that `render_artifact` be used for hosting complete manifests inside the MCP app, while `render_chart` and `render_table` are intended for specific visual deliveries. In "work_mode," users should avoid standalone plugin widgets that lack `appContext` and instead utilize the trusted rendering path. Charting requirements specify the use of `encodings.x.field` and `encodings.y.field` (or `fields` for multiple y-axes), while legacy fields like `xField` are explicitly rejected. For complex visualizations, the `charts_widget_v2` should be emitted via a specific `genui` JSON structure without Markdown backticks. Additionally, the documentation emphasizes strict data integrity: SQL queries must be provided in `source.query.sql` rather than prose, and chart titles must remain neutral and descriptive, reserving subtitles for reader-facing insights rather than technical metadata.