Node reference

Rime nodes are typed DAG steps. Some shape tables, some produce statistical objects, some package artifacts, and language nodes let you drop into Python, R, JavaScript, or SQL when a built-in is not enough.

How To Read These Pages

The node pages are not all shaped the same way. A statistical node needs interpretation and assumption guidance. A table transform needs review cues about shape, columns, and row counts. A language node needs a slot contract.

Each page keeps the schema facts close to the explanation, then spends its space on the parts that matter for that node: what problem it solves, what the output means, what to inspect in the editor or report, and when to choose a different node.

For transform formulas, start with Expression language. For script-backed custom logic, start with language nodes.

Source And Table Transforms

Node	Use it for	Watch for
source	CSV, JSON, NDJSON, Parquet ingress	inferred types, missing paths, report noise
filter	row-level cohort gates	unexpected row loss
derive	one new feature column	null behavior, unreadable formulas
aggregate	grouped or global metrics	metric aliases, collapsed row counts
select	schema narrowing	accidental column drops
sort	review/report ordering	invisible changes when only row order changes

Combining Tables

Node	Use it for	Watch for
join	enriching a left table from a right table	many-to-many row expansion
pivot	long-to-wide summaries	high-cardinality column explosion
concat	stacking peer tables into one tidy table	schema mode and added group labels

Statistical Nodes

Statistical nodes return object outputs. They are report-friendly terminals and can emit assumption warnings.

Node	Use it for	Warning surface
t_test	two-group mean comparison	small/skewed groups, outliers, high variance ratio
anova	multi-group mean comparison	small/skewed groups, outliers, high variance ratio
mann_whitney_u	rank-based two-group comparison	group validity; node-specific warnings are not emitted yet
chi_square	categorical independence	low expected cell counts
correlation	pairwise numeric association	small n, Pearson/Spearman disagreement
linear_regression	single-predictor OLS	small n, high residual outliers

Composition And Escape Hatches

Node	Use it for
subgraph	wrapping an external DAG behind explicit bindings and outputs
html	packaging authored HTML as a cached report artifact
language nodes	custom Python, R, JavaScript, or SQL logic

Shared Node Fields

Every node has an id, kind, and optional metadata.

metadata:
  label: "Friendly node label"
  group: "feature_engineering"
  report: false
  visual_stats: ["row_count"]
  cache: false

Use metadata.label generously. Labels are what reviewers see on the editor canvas and in report DAGs, so they should explain the intent, not just repeat the node id.