- A good data request states what the model should learn, not just what files you want.
- Specify the rights you need up front, because rights shape which holders can supply.
- Give a budget range so holders can judge whether a deal is worth preparing for.
- Keep confidential model details out of the request; describe needs at the level a holder needs to answer.
Template
| Field | What to write |
|---|---|
| Goal | What the model should do better after training |
| Modality | Text, audio, video, images, code, trajectories, preferences |
| Domain | Industry, task type, language, region |
| Volume | Minimum useful amount and ideal amount |
| Quality bar | Native authorship, resolution, transcript accuracy, label agreement |
| Metadata | Fields required per record |
| Time period | Date range wanted |
| Exclusions | What must not be included (minors, health data, published content) |
| Rights | Training, fine-tuning, evaluation, commercial use, territory, term |
| Exclusivity | Exclusive, time-limited exclusive or non-exclusive |
| Delivery | Format, transfer method, schedule |
| Budget | Range and payment structure |
| Timeline | When you need a sample and the full set |
Tips
- Describe the outcome you want; holders often have adjacent data you did not think to ask for.
- Ask for a sample size you can actually test.
- State any acceptance tests you will run, such as duplicate rate or personal information scans.
Send your request through our request form.
Questions
Do I need to share my model details?
No. Describe the data need and quality bar. Holders don't need your architecture or roadmap.