Changelog

Versions follow semantic versioning. Each release is on PyPI and has its own DOI on Zenodo. The version you used is recorded in every run.yaml.

0.1.6 (2026-10-18)

• Runs save finished rows to results.json as they go, and on Ctrl+C, so a stopped run can continue with --resume. In Python, pass checkpoint to run or arun.
• Documentation at paperlab.space/docs, with a reference generated from the source.
• Docstrings for every public field, shown in help() and in editors.

0.1.5 (2026-10-18)

• A key with no quota at all for a model, such as a Pro model on Gemini's free tier, now stops that model straight away instead of retrying every row.
• Progress updates carry the row that just finished (Progress.row), so callers can save partial results.

0.1.4 (2026-10-17)

• Errors show the provider's own message instead of LiteLLM's wrapped JSON.
• results.json records the dataset by file name only, so sharing it does not reveal local folder names.

0.1.3 (2026-10-17)

• list_sheets raises DataError for a corrupt Excel file, like load_dataset.
• Zenodo DOI in the citation file and README.

0.1.2 (2026-10-17)

• Author and ORCID in the citation metadata.

0.1.1 (2026-10-17)

• The paperlab-space command is part of the base install. pip install 'paperlab-space[cli]' still works.

0.1.0 (2026-10-17)

First release.

• Loading .xlsx, .csv, and .tsv files, column mapping, validation, and JSONL preview.
• Prompt templates and the YAML run config.
• An async runner with per-model rate limits, retries, resume, and cost and time estimates, for every provider LiteLLM supports.
• Metrics: exact match, accuracy with label extraction, classification F1, and token F1, with tokenisation for scripts written without spaces.
• Exports: Excel, a booktabs LaTeX table, the reproducibility config, and results.json.
• The paperlab-space command: init, validate, estimate, run, and metrics.