Reference

What needs the Source panel or an agent

Most editing has a form in the app: adding and arranging blocks, the authoring blocks around every selected block, the form of a layered or composite chart, creating params, the whole datasource lifecycle (SQL, run, snapshot, rename, delete), field labels and formats, dashboard duplicate/import/export, and comments.

These things have no form, and you write them in the dashboard's JSON:

• A chart's advanced Vega-Lite: layers, selections, and custom transforms. The chart's authoring blocks cover the mark, the x/y/color/size encodings, aggregate, sort, stacking, and tooltip. For the rest, "Edit spec" in the chart's Code block opens the chart in the Chart workspace, with the cursor in the code pane that holds the raw spec.
• Marks and encoding channels that the menus of the chart's authoring blocks leave out. The menus hold the common ones. Vega-Lite has more marks, such as tick, rule, text and boxplot, and more channels, such as x2, row and column. The canvas draws whatever is in the spec. When you select a chart, the Code block beside it has the parts that you cannot edit in the other blocks.
• Field thresholds and value mappings. A measure and a custom column each have their own editor in the datasource's Fields section.
• A datasource's source, which says where its rows come from. bigquery queries the warehouse. derived runs SQL over the other snapshots, in your browser and on the server. file runs SQL over a file that you brought in: a file you picked, dropped or pasted, or one in your files. imported holds rows from a dashboard made before every import was a file, so there is nothing to run again. url runs SQL over a Google Sheet or a file at an address. database runs SQL in a database that the operator of this site registered, as a registered service account. To change a source after you create the datasource, edit it in the Source panel.
• A datasource's runsAs, which says whose credential the run uses. On arkush.app, a BigQuery datasource runs as user: the query runs in the browser of the editor who selects Refresh, under that editor's own Google account, and no schedule runs it. On an org install, the Runs as field of a BigQuery datasource has each registered service account that you may use, once the dashboard is saved on the server. Whoever manages IAM in Google Cloud decides who may use an account, except for an account that the operator gave a key file: for that account, an admin decides in the app, with the account's list. When the field has no account, it is greyed, and its note has the reason and who can change that. With an account in the field, set runsAs to a service-account email that you may use, and the server runs the query as that account, so a schedule or an agent can refresh it. The server refuses a run as an account that you may not use. With no runsAs, a BigQuery datasource cannot run on the server. On an org install with a registered database, a database datasource always shows the field, and takes a service account only.
• A rich tooltip's layout. With the Rich tooltip switch in the Tooltip row, the layout follows the chart. To set the series list or the mark's context yourself, write usermeta.tooltip.mode as series or context.
• Composite charts: facet, repeat, and concat.
• New connector blocks: a line drawn between two blocks. The palette inserts every other block type. You write a new connector in the Source panel or through an agent. After that, the Connector card beside a selected connector sets its two ends and its title.

The Source panel edits the same JSON that agents edit, so anything an agent can write, you can write there too.