Train, validation & holdout overlap checker
Check whether training, validation and final holdout data share timestamps. Free, no sign-up, with an exportable summary for reviewing MT5 backtests and quantitative research splits.
Paste timestamps to check overlap
One ISO date (2026-01-01) or timestamp with timezone (2026-01-01T08:00:00+08:00) per line. An optional timestamp header is accepted. Dates mean midnight UTC; normalize MT5 server time first. Paste only the timestamp column, up to 500,000 characters per split.
This tool computes in your browser and does not upload the timestamps you enter.
Calculation and worked example
Each split is normalized to UTC and deduplicated. Holdout overlap percentage = unique holdout timestamps present in training OR validation / total unique holdout timestamps. A timestamp seen in both earlier splits is counted once.
The example uses January 1–3 for training, January 3–4 for validation and January 4–6 for holdout. Training/validation overlap is 1; validation/holdout overlap is 1. Holdout overlap is 1 / 3 = 33.33%. These are illustrative data, not platform strategy results.
Does zero overlap mean there is no data leakage?
Also inspect future-looking indicators, labels crossing split boundaries, preprocessing fitted on the whole dataset, and holdout results reused for parameter selection. Check each instrument separately: equal timestamps do not imply identical records across instruments.
Reference: scikit-learn data leakage guidance
What should you check next?
If intersections appear, review original records and experiment logs. Freeze an independent holdout, document sample counts and costs, then examine statistical gates and MT5 replay consistency.