Skip to main content

Generate reviewable code in your stack

Anyone's AI can write a mapping. The value is in shipping it safely: DataChord turns your mapping into reviewable code in the language your team runs, with golden-sample tests, an audit-ready mapping spec, and a Docker package — so a senior dev reviews a real diff and runs it in your architecture, not a black box on ours.

By the end of this page you will have:

  • Generated a transform in Python, TypeScript, or Java from one mapping.
  • Downloaded a golden-sample test suite that pins the code to the mapping's semantics.
  • Exported a Markdown mapping specification that replaces the Excel sheet.
  • Built a Docker package (CLI or FastAPI) ready to run.
You'll need
  • A MapCraft project with at least one mapping rule (see the walkthrough).

Step 1 — Generate code in your language​

Open the Codegen panel for the project and click Generate All. Each target renders the same mapping IR deterministically:

TargetOutput
Pythontransform(record) -> dict module (stdlib only)
TypeScriptexport function transform(record) module
JavaTransform class with a transform(Map) method
SQL / JSONata / XSLTwarehouse / document expressions
FastAPIPOST /transform service scaffold

The TypeScript and Java emitters cover the full rule vocabulary — string/number/ date functions, conditionals, lookups, and array iteration/aggregation — with the same semantics as Python (including banker's rounding), so the languages agree value-for-value.

Step 2 — Download golden-sample tests​

With a code target open, click Tests. DataChord computes the expected output for each sample by running the mapping IR through the same evaluator the verifier trusts, then emits a matching suite:

  • Python → pytest (test_transform.py)
  • TypeScript → jest (transform.test.ts)
  • Java → a JUnit scaffold loading the JSON fixtures

Drop the test file next to the generated code and your reviewer has a runnable regression pin, not just a diff to eyeball.

Step 3 — Export the mapping specification​

From the MapCraft toolbar Export dropdown, choose Documentation. You get a Markdown spec rendered from the same mapping definition: source/target schema tables, the field-mapping matrix (with confidence), per-rule detail (natural language + compiled Python/SQL/JSONata), combined rules, and lookup tables. It's deterministic — regenerate it any time the mapping changes instead of hand-editing a spreadsheet.

Step 4 — Build a Docker package​

In the Codegen panel, open the Docker package tab and pick a flavor:

  • Python CLI — a stdlib-only image that reads a JSON record on stdin and writes the transformed record on stdout. Golden tests are bundled.
  • FastAPI service — the generated POST /transform app served by uvicorn.

Download the ZIP, then:

docker build -t my-transform .
echo '{"example":"record"}' | docker run -i --rm my-transform   # CLI flavor

Verify​

  • The generated TypeScript runs under Node and produces the same output as the Python transform for the same input.
  • The downloaded pytest suite passes against the downloaded transform.py.
  • The Docker image builds and runs the sample.

What's next​