Templates

Templates are runnable workflow starters. A good template is not just a graph of nodes; it is a scientific story with an input, a processing path, a primary plot or metric, and a next decision.

Use templates when you want to move quickly from a dataset to a defensible first result. After the first run, edit the workflow: change preprocessing, adjust model settings, add validation nodes, or replace the starter data with your own imported dataset.

flowchart LR
    D[Dataset] --> Q[Quality check]
    Q --> P[Preprocess]
    P --> M[Model or analysis]
    M --> V[Validate and interpret]
    V --> R[Report or export]

Strong First Templates

Template family	Use it when	Primary output	What to decide next
PCA exploratory analysis	You need to understand sample structure before modeling	Scores, loadings, explained variance, diagnostics	Whether preprocessing and sample grouping make scientific sense
PLS calibration	Spectra have quantitative reference values	Predicted vs measured, RMSEP/R2/bias, VIP, coefficients	Whether validation design and target alignment are trustworthy
Representative or peak-guided PLS	You want a compact calibration workflow with variable interpretation	Calibration metrics plus selected regions or peak-driven features	Whether the reduced model remains chemically plausible
KNN, PLS-DA, SIMCA	Samples have class labels or QC acceptance categories	Confusion matrix, class metrics, SIMCA acceptance diagnostics	Whether class design, rejects, and holdout samples are appropriate
MCR-ALS	Mixtures contain overlapping spectral components	Resolved spectra, concentration-like profiles, lack-of-fit	Whether the resolved components are chemically defensible
Peak and library comparison	You need local spectral features or reference comparison	Peak table, matched references, similarity scores	Whether peaks/matches agree with sample chemistry and metadata

How to Read a Template Page

Each template page explains:

• expected input data and metadata
• the end-to-end workflow path
• the plots, tables, metrics, and artifacts it should produce
• how a scientist should review the result
• common failure modes before using the output in a report

The repository may contain templates marked wip. They should not be treated as production onboarding paths until their data source, plots, metrics, and story are verified.