Dataset

`trialdesignbench.dataset`

Import curated intake JSON into the canonical, versioned dataset format.

Layout written by import_intake:

<dataset_dir>/
  dataset.json          DatasetManifest
  <task_id>/
    task.json           TaskRecord (agent-visible)
    document.md         protocol/SAP text, when available
    rubrics.json        RubricSet (hidden from the agent)

`DatasetError`

Bases: ValueError

Raised when intake data or a dataset directory is invalid.

`attach_document(dataset_dir, task_id, document, *, source=None)`

Add or replace a task's source document and refresh digests.

`check_dataset(dataset_dir)`

Return every problem found. An empty list means the dataset is usable.

`convert_prompts(sub)`

Split intake prompts into agent-visible questions and hidden rubrics.

`find_document(documents_dir, sub)`

Locate <task_id>.md, <trial_id>.md, or <intake stem>.md.

`import_intake(intake_files, out, *, documents_dir=None, username=None, dataset_version=DEFAULT_DATASET_VERSION, publicly_indexed=None, force=False)`

Import intake files into a canonical dataset directory.

`load_dataset(dataset_dir, task_ids=None)`

Load the manifest and (task, rubrics) pairs, optionally a subset.

`load_manifest(dataset_dir)`

Load a dataset.json.

`load_rubrics(path)`

Load a rubrics.json.

`load_task(task_dir)`

Load a task.json.

`read_intake(path)`

Read and minimally validate one curated intake JSON file.

`select_submissions(submissions, username=None)`

Pick one submission per task: latest submittedAt, or the given user's.

`task_id_from_trial_id(trial_id)`

Filesystem- and registry-safe slug for a trial identifier.

10.1200_jco-24-01818 -> 10-1200-jco-24-01818. A leading URL scheme and doi.org/ are dropped so DOI spellings map to the same slug.

`write_manifest(out, *, dataset_version)`

Recompute task digests and write dataset.json.