Cars x CO2 emissions

This is the flagship polyglot example. It stitches public vehicle data to a CO2 concentration series, fans the unified table into Python, R, and built-in stat nodes, then renders an HTML narrative.

Source fixture: packages/core/test/fixtures/experiments/cars-emissions-narrative.

What it Teaches

• SQL ingress can read public JSON directly through DuckDB.
• JavaScript nodes can fetch and normalize external API/CSV data.
• A single feature table can feed Python, R, and terminal stat nodes.
• Python and R nodes can emit diagnostics and return tabular results.
• Stat nodes are terminal report artifacts, not dataframe transforms.

DAG Shape

sql_cars_source ─┐
                 ├─► js_feature_build ─┬─► python_umap_embed
js_co2_fetch ────┘                     ├─► r_efficiency_trends
                                       ├─► t_test_us_vs_japan
                                       └─► anova_cylinders

js_feature_build is the apex dataframe. Everything downstream consumes the same vehicle-plus-CO2 feature table.

Important Nodes

specification_version: "2.1"
nodes:
  - id: sql_cars_source
    kind: sql
    source: scripts/sql_cars_source.sql

  - id: js_co2_fetch
    kind: javascript
    source: scripts/js_co2_fetch.js

  - id: js_feature_build
    kind: javascript
    in:
      cars: sql_cars_source
      co2_yearly: js_co2_fetch
    source: scripts/js_feature_build.js

  - id: python_umap_embed
    kind: python
    in:
      features: js_feature_build
    source: scripts/python_umap.py

  - id: r_efficiency_trends
    kind: r
    in:
      features: js_feature_build
    source: scripts/r_efficiency_trends.R

  - id: t_test_us_vs_japan
    kind: t_test
    inputs: [js_feature_build]
    valueColumn: miles_per_gallon
    groupColumn: origin
    groupA: USA
    groupB: Japan
    equalVariance: false

The SQL source is a DuckDB query:

SELECT
  CAST(Name AS VARCHAR) AS name,
  CAST(Miles_per_Gallon AS DOUBLE) AS miles_per_gallon,
  CAST(Cylinders AS INTEGER) AS cylinders,
  CAST(Horsepower AS DOUBLE) AS horsepower,
  CAST(Year AS VARCHAR) AS year_str,
  CAST(Origin AS VARCHAR) AS origin
FROM read_json_auto(
  'https://raw.githubusercontent.com/vega/vega-datasets/main/data/cars.json'
)
WHERE Miles_per_Gallon IS NOT NULL
  AND Horsepower IS NOT NULL

The JavaScript CO2 fetch node is ingress-style: it has no in: slots and returns a named table output.

export default defineNode({
  in: {},
  out: { co2_by_year: 'table' },
  async run() {
    const response = await fetch(DEFAULT_URL)
    const text = await response.text()
    return { co2_by_year: parseAnnualRows(text) }
  },
})

Run It

Python and R sidecars are required for the full run:

git clone https://github.com/danielsjoo/rime
cd rime

export RIME_PYTHON_BIN=/path/to/python
export RIME_RSCRIPT_BIN=/path/to/Rscript

rime validate packages/core/test/fixtures/experiments/cars-emissions-narrative/pipeline.dag.yaml
rime run packages/core/test/fixtures/experiments/cars-emissions-narrative/pipeline.dag.yaml

The fixture also includes a hand-authored report spec:

rime build \
  packages/core/test/fixtures/experiments/cars-emissions-narrative/pipeline.dag.yaml \
  --report packages/core/test/fixtures/experiments/cars-emissions-narrative/report.yaml

The run writes:

• outputs/js_co2_fetch/co2_by_year.parquet
• outputs/js_feature_build/features.parquet
• outputs/python_umap_embed/default.parquet
• outputs/r_efficiency_trends/default.parquet
• outputs/t_test_us_vs_japan/default.json
• outputs/anova_cylinders/default.json
• outputs/d3_umap_scatter/default.html
• outputs/cars-emissions-narrative-report.html

What To Inspect

• scripts/sql_cars_source.sql for DuckDB URL ingress.
• scripts/js_co2_fetch.js for a JavaScript API/CSV fetch node.
• scripts/python_umap.py for a Python diagnostic transform.
• scripts/r_efficiency_trends.R for an R statistical transform.
• report.yaml to see the compatibility path for hand-authored report prose and ordering.

Requirements

Install the packages expected by the fixture before running the Python/R nodes:

Runtime	Packages
Python	pandas, numpy, pyarrow, matplotlib, scikit-learn, umap-learn
R	jsonlite, arrow, ggplot2, dplyr, boot

For a no-sidecar example, start with DuckDB single source or Single-file pipeline.