06Service
Data collection and validation
New data gathered to spec, and existing labels checked and corrected.
When your model needs more data, or your labels can't be trusted, we collect new items to a written spec or audit what you already have. Every change is logged, so you can see exactly what was fixed and why.
- Annotation types
- 4 types
- Delivered as
- Your format, with a change report
- Quality levels
- Attempt → Review → Final Review
- Starts with
- A paid pilot of 500–1,000 items
01Showcase
See it labeled.
What a label audit changes, row by row.
| Item | Label | Audit note |
|---|---|---|
| img_00412 | cat | |
| img_00413 | dog | |
| img_00977 | car | |
| img_01104 | bicycle | |
| img_01528 | person | |
| img_01690 | (empty) |
02Annotation types
What we label.
Each type is defined in your guidelines before the pilot, with examples of the edge cases.
New photos, documents or text samples gathered to your spec, with consent where needed.
A sample or full pass over existing labels, with an error rate by class.
Duplicates, corrupt files and out-of-scope items found and removed.
Human judgments on your model's predictions or responses, for evaluation and fine-tuning.
03What you receive
Labels your pipeline can load as-is.
Delivered as Your format, with a change report. Tell us your schema and we'll match it.
- Labels in the format your pipeline already reads
- A quality report per batch: accuracy on hidden test items against your target
- Every task through Attempt, Review and Final Review
- A 7-day window to accept each batch or raise a dispute
item_id,field,before,after,reason
img_00412,label,cat,dog,misclassified
img_00977,bbox,"[10,12,80,64]","[14,12,92,70]",box too tight
txt_01530,removed,,,duplicate of txt_0152804Use cases
Where teams use it.
- 01
Filling dataset gaps
- 02
Fixing noisy labels
- 03
Evaluating model output
05Quality
Every task passes three levels.
Only tasks approved at Final Review are paid to the annotator and delivered to you.
- Level 1Annotator
Attempt
Labels the item to the project guidelines and label schema.
- Level 2Reviewer
Review
Checks the work, then approves it, corrects it, or sends it back with a reason.
- Level 3Final reviewer
Final Review
Signs the task off. Only finally approved tasks are paid and delivered.
06Other services
Need more than one data type?
Start a data collection and validation pilot. Judge us on your own data.
We label 500–1,000 of your items through all three levels. You approve the results before production begins.
Request a Quote