Results, samples and logs

Selecting a node opens two panels. The inspector on the right describes the node: its parameters, compiled SQL, documentation and change history. Resolved variables and compiled SQL follow the selected environment. The results panel under the canvas shows the data, charts, tests and logs for that environment.

The results panel of Paid orders on a sample of partner orders above 300, sorted by amount

The header names the node and the dataset shown, its Valid or Stale badge, the time, duration and commit of the run that produced the rows, and the row and column counts. With several outputs, or inputs, the selector lists the node’s outputs and, under Inputs, to compare, the datasets connected to its inputs: showing an input keeps the current sample when its columns allow it, to check an operation before and after. A sample filter is kept only when the other dataset has every column of the current one; otherwise that dataset’s own sample is restored, as when opening failing rows from an input. An export node shows its connected input in Data, Columns and Chart. The header names the upstream dataset; with several connected inputs, the Inputs selector lets you choose. These are the retained input rows, not a readback of the destination file or database; inspecting them does not execute the export, and before they exist Run ancestors computes them without writing the destination. Charts belong to the upstream dataset, so a chart added here is also available at its source node. Log, Tests and downloads still describe the export itself: a failed export shows its error above the input rows, and the header keeps the download button in every tab. Without a connected input, only Log and Tests are available. A component instance shows its exposed outputs.

The inspector’s result summary links to Data, Log and Tests, showing validity, warnings and the latest failure for the selected environment.

A failed node shows its error above its results, with actions to open the code at the failing line and the execution log. A blocked node, or a component with a failure inside, names the node that failed and shows its error, however many dependencies away, whether or not earlier rows remain. The error belongs to the failed run and clears after a successful one.

While you edit a node’s code, Preview adds a Preview tab with the outputs computed from samples of its inputs. Nothing in it is saved or published; see previews.

Result lifetime

Results belong to the branch and environment selected in the top bar. Switching either can show different results or none at all; it never starts a run.

State	What it means	What to do
No result	This context has not run, or its in-memory data was cleared, for example by a backend restart	Run the node with its ancestors, or choose Run all
Valid	The retained data matches the saved processing configuration	Inspect, sample and explain cells
Stale	The retained data is from an earlier configuration or attempt	Inspect the old rows if useful; rerun before using them as current results or exploring lineage
Failed or canceled run	Nodes that finished before the failure published their results; the failed, blocked and canceled nodes did not	Inspect Log; their previous outputs can remain visible as stale

SQL, parameter and upstream processing changes can invalidate results. Editing layout or documentation does not. Saving never launches a run, and changing an external source file does not refresh an existing result: run again to read it.

History keeps execution records, SQL and outcomes after a restart; it is not a stored copy of every historical result table. Use an export when you need a result file to keep or share. Job results follow the same context lifetime; see Jobs.

Data and samples

The Data tab pages through the retained result, 100 rows at a time. Select a value to explain it; filtered, sorted and random rows keep their identity, so lineage works on any cell of a valid result. Search all columns finds text in every column. Clicking a column header sorts by that column.

Sample chooses which rows are displayed:

Rows: all rows, the first rows, or a random sample with a seed, up to 10,000 rows after the filter and sort. A seed always shows the same rows of the same result.
Filter: a SQL condition on the dataset’s columns, written without WHERE, such as channel = 'Partner' AND amount > 300. It is validated like a Filter node’s condition: one expression, no query or statement, only this dataset’s columns. {{ var.name }} references take the selected environment’s value as a bound parameter. The number of matching rows updates as you type.
Sort: one or more columns, ascending or descending.

The Sample popover with a filter, its matching rows and a sort

Chips under the tabs show the active filter and sort; the cross removes one. A sample only changes the display: runs always compute on all the data, and descendants never read a sample. The sample is remembered in your browser for each node output; it survives reruns and environment switches. A filter that no longer applies, after a column was renamed for instance, is reported with Reset instead of being dropped. Reading a sample stops after 30 seconds.

From the Sample popover:

Add as a Filter node inserts a Filter node reading this output, with the same condition, when a check should become part of the pipeline.
Create a test from this filter appends -- @test <name>: <severity> with SELECT * FROM {{ output }} WHERE <filter>; to the node code: rows matching the filter then fail the test. For a node linked to a template, the test goes into the template and checks every linked node, unless you detach this node first. Node code cannot reference project variables, so a filter using them must be rewritten with values.
Copy SQL copies the query of the sample, written with {{ port }} rather than internal table names.

Save as view… names the current sample and saves it in views.json, so the whole team finds “Large orders” or “Partner orders above 300” next to All rows above the table. A view is a normal save: one commit, merged view by view between branches, listed in comparisons and included in project ZIP archives. Views never change results, so saving one makes nothing stale. Deleting a node deletes its views in the same save. Readers and operators use views and their own samples but cannot save views.

For external editing, see the saved view file format.

Chart

The Chart tab: revenue by region, the leading region first

The Chart tab draws the shown dataset with the Vega-Lite charts declared for it in charts.json. A chart plots the whole retained result, up to 5,000 rows, never the displayed sample. Editors can choose Add a chart on an output without one. See the charts guide for the format, the allowed specs, value conversion and the security model.

Columns and statistics

The Columns tab with descriptions and computed statistics

The Columns tab lists each column’s position, name, type and description, with the dataset description above, at once and without reading any rows. Descriptions are edited in place with the pencil and saved in documentation.json. Selecting a column name shows it in the Data tab. A chip compares the columns with the previous successful run of the node in the same environment: added, removed and retyped columns, or “Same columns as the previous run”.

Compute statistics scans the whole retained result once, never the displayed sample: null percentage, approximate distinct values, range and distribution of numbers and dates, and top values of text. The tab never starts this computation by itself, since it competes with runs for the CPU on large results. Statistics are kept for that result: reopening the tab, coming back to the node or another user shows them at once with the time they were computed, and the run that replaces the result discards them. The computation can be canceled and stops after 30 seconds.

Tests

The Tests tab with a failing warning test, its query and history

The Tests tab summarizes the passed, warning and error tests of the latest check, then lists each test with its severity, output, failing rows and its last ten runs in this environment. Select a status to see the test’s query and since when it fails. Show the failing rows opens the Data tab on the rows the test returns, recomputed from the retained result with the definitions recorded by the run: they are paged like a sample and never stored. It needs a valid result. Edit tests in the node code opens the code editor.

Failing rows of a test, recomputed from the retained result

Log

The Log tab: run summary, timeline and executed SQL

The Log tab shows the latest run of the node: its status, start and end, duration, commit and whether it matches the current configuration, who launched it, its scope and warnings. The timeline lists the inputs, each output with its row count and computing time, each -- @message result, the tests and the publication, or the error that prevented it, with Ask the assistant. A filter shows all entries, messages, warnings or errors. Executed SQL shows the statements, bound values and variables of that run, with secrets masked. Open the run in History opens its full record.

Download exported data

Use an Export CSV or Export XLSX node to write the full input table, including the format options configured on that node. After a successful run, select the export and use the download button of the results header, or choose Log → Download exported data, which also names the file, its time and its execution. For a first export, follow Use your own data.

Downloads apply to a single COPY ... TO statement using a destination parameter and a local CSV/XLSX target. They use a saved copy of the exported bytes, remain private to project members and survive a backend restart. Changed inputs/settings, upstream reruns, a failed export or a missing download copy require another successful export. A new export replaces the previous download. Database and remote-storage destinations do not offer browser downloads.

Code → Download project ZIP shares the pipeline definitions, excluding data and Git history. It does not download the displayed table. See project files and imports.