Output Nodes
Output nodes make workflow results visible or exportable.
Nodes
| Node | Use When | Inputs | Outputs | Key Configuration |
|---|---|---|---|---|
Plot (output.plot) |
Show spectra, scores, loadings, diagnostics, dendrograms, or metrics-derived plots. | default: Any |
visualization |
plot_type; colorscale; x_axis; y_axis. Many model outputs carry plot-ready payloads. |
Contour Plot (output.contour) |
Show a 2D heatmap or contour view of matrix-like spectral data. | default: Any |
visualization |
colorscale; plot_type; reverse_x; transpose. |
Data Table (output.data_table) |
Inspect numeric arrays, metrics, sample tables, or selected variables as rows and columns. | default: Any |
visualization |
max_rows; transpose; show_index. |
Statistics (stats.summary) |
Generate deterministic descriptive summaries for datasets and typed scientific results. Inferential decisions remain in dedicated diagnostic nodes. | default: Any |
statistics: StatisticsSummary |
max_samples. |
Prepare Export (output.export) |
Prepare exact, verified bytes for a scientist-controlled download or generated-script materialization. | data: Array2D/1.0 |
artifact: ExportArtifact/1.0 |
Safe basename filename; closed format (csv, json, or jdx). |
Good Output Hygiene
All plot, table, statistics, and export surfaces are governed by the Scientific Result Surface Contract. It defines the required authority, population and transformation disclosures, lifecycle behavior, refusal states, and acceptance cases. A successful render or download alone does not establish scientific fidelity.
For chemometrics, useful output includes the numbers and the context: sample IDs, target names, units, metric names, split design, and model parameters.
Choosing the Right Output
- Use plots for pattern recognition: spectra, scores, loadings, residual diagnostics, dendrograms, and acceptance regions.
- Use tables for auditability: sample IDs, target values, predictions, selected variables, peak lists, and metric dictionaries.
- Use statistics summaries for quick health checks, not as a substitute for validation nodes.
- Use export only after confirming the dataset carries the sample and feature labels, units, and role needed by the recipient. CSV and JSON preserve the complete two-dimensional dataset. JCAMP-DX is admitted only for one spectrum with a measured numeric feature axis and declared units; the node refuses to invent a wavelength axis, spectral units, or a multi-spectrum JCAMP dialect.
- The DAG node prepares a closed artifact containing the filename, format, media type, exact content, byte length, source digest, and content digest. It never chooses or writes a filesystem path. The workbench download and the generated-script host independently verify that contract before handing the exact bytes to their host. The generated-script materializer additionally refuses to overwrite an existing destination.
Saved model handoff
A reusable fitted model must appear in Runs → Artifacts with its source run and be available to local Deploy after execution. Cohort-only clustering remains inspectable but does not predict new observations. See the canonical artifact acceptance contract.