SalesTranscriptQA v0.1.0 — 200-question pilot and CLI
Pre-release
SalesTranscriptQA's first pilot release contains 200 automatically validated questions over 10,829 verbatim synthetic CRMArena-Pro sales calls, with a Python CLI and reproducible generation harness.
- 50 single-call and 50 exactly-two-call questions each for B2B and B2C.
- Full domain corpora, concise answers, exact dialogue/metadata evidence, Parquet and Markdown/YAML exports.
- Final run: 256 candidates, 200 accepted; median answer 13 words.
- Estimated API usage cost: $0.88594 pilot; $0.91960 including discarded development and smoke checks. Not an invoice.
- 16 local tests, live positive/negative CLI judging, full corpus export checks, all 200 reference scoring fixtures, byte-identical corpus regeneration, exact evidence and duplicate gates passed.
Download salestranscriptqa-pilot-v0.1.0.zip and extract it to obtain salestranscriptqa-data/. The CLI uses that directory by default. Install the wheel with uv or use uv tool install git+https://github.com/Endgame-Labs/SalesTranscriptQA.git@v0.1.0.
Data: CC BY-NC 4.0, attributed to Salesforce AI Research / CRMArena-Pro. Independent code: MIT. No human review or calibration was performed; automated agreement is not human-verified correctness. This development pilot is not an untouched test set and does not reproduce CRMArena-Pro's CRM workflow rewards.
Published on Hugging Face, revision d5eae88b4a725b697ab81dc720429477611f1a9c. Anonymous CLI fetch and all file checksums verified. The same data files are included in this archive.