Fellegi & Sunter · JASA 1969 · six weeks

ProbabilisticRecord Linkage

Deciding when two records are the same thing — and, more usefully, when to refuse to decide.

Evidence ledger −16.0 bits · non-link
Cumulative match weight across six weeks, shown as a waterfall against two decision thresholds.

You start 16 bits in the hole — the prior odds that any two records match. Each completed task is evidence. Cross the lower threshold and you leave the non-link region; cross the upper one and you know this material.

Progress saves automatically.

If you only remember one thing: the theorem doesn't tell you how to be certain. It tells you the cheapest honest way to admit you aren't.