Dataset
Current pack notes, public sample, and September 8 release status.
Schema fields
Type, Product, Started Date, Completed Date, Description, Amount, Fee, Currency, State, Balance.
Current release note
The current public sample is a single shared pack used across the visible rankings. The September 8 update refreshes model coverage while keeping the sample stable.
Next release
Planned additions include more source shapes, more merchant diversity, and clearer separation between public sample rows and scored holdouts.
Gold labels
Scoring scope
All 512 rows are sent to every participant. The 490 completed rows are scored on all six tracks; the 22 reverted rows must be flagged, not categorized, and only feed the reliability column. A private holdout is planned for v1.3.
Client context
Each run ships a one-page client profile (travel per-diem policy, coworking lease, 50% meals rule) and a 90-day vendor history with 38 prior postings. Context-fit rows are the ones where that material changes the correct answer.
Current release note: the public sample is stable; the September 8 refresh updates model coverage and site metadata.