Why Pairs Die Out of Sample

A scan tests 300 pairs at the 5% level and reports 16 raw passes. What does that count show on its own?

• That about 16 real pairs were found
• Roughly what chance alone would produce if none of the pairs were related
• That the 5% level was too strict for this universe

Which death cause can none of the gates catch?

• Fake reversion from stale prints
• A relation that broke near the end of training
• A pair chosen by a test that already saw the test window