Mapping configuration

A standardizer produces output columns using four mapping modes: simple, steps, expression, and llm.

Simple mode

Copy a value from an input column and optionally apply one transform.

output_column: sku
mode: simple
source:
  alias: orders
  field: product_code
transform:
  type: uppercase

Steps mode

Chain multiple transforms together. The output of each step becomes the input of the next.

output_column: unit_price
mode: steps
steps:
  - type: source_field
    params:
      alias: items
      field: description
    output_as: description
  - type: regex_extract
    params:
      pattern: "\\(\\s*\\d+\\s*X\\s*([A-Z0-9]+)\\s*\\)"
      group: 1
    output_as: ref_number
  - type: lookup
    params:
      dataset: price_list
      match_column: ref_number
      return_column: unit_price

Use output_as to save an intermediate value so later steps can reference it.

Expression mode

Write a formula that references input columns and helper functions.

output_column: line_total
mode: expression
expression: "qty * unit_price"

Expressions support arithmetic, comparisons, field references, and helper functions such as lookup, regex_extract, upper, lower, trim, coalesce, ifnull, and round.

LLM mode

Send a prompt to an LLM for each row and parse the result into the output column. LLM mappings require an llm_provider credential.

output_column: category_name
mode: llm
data_type: string
prompt: "Classify this product into one category: {{products.product_name}}"
credential_id: "<llm-credential-id>"
response_format: value
on_error: null
timeout_seconds: 30
max_tokens: 500
temperature: 0

• credential_id: the LLM provider credential to use.
• response_format: value for a single value, or json for a JSON object.
• on_error: fail, null, or passthrough.
• timeout_seconds, max_tokens, temperature: optional LLM request controls.

Use {{alias.field}} or {{field}} placeholders to include row values in the prompt. The platform sends only the prompt template plus those substitutions; raw file data is not sent outside the transformation context.

Data types

For each output column you can declare a target type: string, integer, decimal, boolean, date, datetime, or time. Declaring a type helps arithmetic and comparisons behave correctly, especially when the input stores numbers or dates as text.

Column order

Output columns appear in the final file in the same order they are defined in the standardizer.