Train Val Leakage

TODO:

• [ ] Add page-specific notes for data-governance/data-quality/train-val-leakage.md after inspecting the real project files.
• [ ] Cross-link Train Val Leakage to the most relevant code, data, testing, and safety pages.
• [ ] Define the rule this page enforces for data or labels.
• [ ] List files, folders, and scripts affected by the rule.
• [ ] Add the validation command or inspection step for the rule.
• [ ] Add what stale or duplicate data looks like.
• [ ] Add recovery steps if the rule was broken.
• [ ] Add public-release checks for privacy and traceability.
• [ ] List dataset tag, image count, label count, source, and purpose.
• [ ] Add how to verify filenames, annotations, and correction labels match.
• [ ] Add the full training command with every flag.
• [ ] Document default value, unit, and reason for each training flag.
• [ ] Add how to verify samples per epoch, roots, corrections, and source weights were used correctly.
• [ ] Add the exact source path, artifact path, or hardware component name.
• [ ] Add the command or procedure needed to reproduce the result.