Free AI data tool · Quality
Train/test overlap checker
Quick answer
Paste or upload a training sample and a test sample, choose the fields to match on and an optional group ID, and see every exact overlap.
Training sample
Your sample is processed in this browser. File contents are not uploaded.
Test or evaluation sample
Same limit per file: 10 MiB or 25,000 records. Press Compare above.
Supported input
Two files (CSV, JSON array or JSONL), each up to 10 MiB or 25,000 records.
How it works
Both samples are keyed the same way as the deduplicator: whole record or selected fields, with optional case and whitespace normalization. Every test record whose key also appears in training is reported with the matching training record numbers. Group IDs present in both samples are reported separately, since a shared conversation or source can leak information even when individual records differ.
Limitations
- Only exact matches after the chosen normalization are found; paraphrases are not.
- A clean result covers only the examined samples and does not show a model or benchmark is free of contamination.
Questions
- How do I check for overlap between train and test sets?
- Upload both samples, pick the fields to compare and an optional group ID such as conversation_id, and press Compare. Exact record matches and shared group IDs are listed separately.
- Is my file uploaded?
- No. The file is read and processed in this browser tab, and its contents are not sent to SourceX or anyone else. Closing the tab or pressing Reset clears it.
Learn the format
Related tools
Request fresh evaluation data
Need held-out data that was never in your training set? Describe it and we'll scope licensed business data for evaluation.
Request fresh evaluation data