Text

collectra.Text holds text produced by OCR, language models, or text-processing tasks. For example:

transcription:
  type: collectra.Text
  id: transcription-001
  parents: page-001
  data: "Collected in Melbourne."

Commands

evaluate PREDICTED_FOLDER GOLD_FOLDER is the only Text command. It uses difflib.SequenceMatcher similarity, not character-error rate or word-error rate. The shared evaluation options control matching and CSV output.

collectra --pipeline pipeline.yaml artefact transcription evaluate predictions gold --output text-metrics.csv
collectra --pipeline pipeline.yaml artefact transcription evaluate --help

There is no Text extract CLI command. Text is stored in results.yaml and can be inspected in the GUI or accessed in Python.

File-backed text

Use path instead of data to store text in a separate UTF-8 file:

transcription:
  type: collectra.Text
  id: transcription-001
  path: texts/transcription.txt

Relative paths are resolved from the results directory. The file must exist. Providing both data and path raises an error. Calling the Text object returns the loaded contents; serialization preserves path and omits data. The GUI writes edits to the referenced file. Legacy file references in data remain supported, but new records should use path.