Simulation run on this device. Scenario values are not sent anywhere — the engine is fully local.
research studio · free
Eval Leakage & Dedup Scanner
Hunt near-duplicates across splits with n-gram overlap heuristics.
Is your test set inside the training set?
Engine 1.0.0 · studio
0.5
Add at least two eval items to scan for near-duplicates and leakage.
Method
- Computes pairwise Jaccard similarity over token sets and flags every pair above the threshold.
- Similarity catches template clones and answer echoes; it cannot see leakage through shared entities alone — inspect the flagged pairs.