Bob Resolve / Synthetic benchmark
How well does the engine match people on its own?
All figures are measured on synthetic data. Automatic figures count only merges the engine made without a person. Not recorded means the run did not measure it.
Run demo-run
Data: derived from the answer key. Used while building the rules, so not held out.
Model tiers used: none (rules only). Shared IDs on. 3,458 candidate pairs.
MBI and linking policy IDs were available to the matcher. Display masking does not remove that matching signal.
This data is not held out, so these figures are not accuracy on data the engine has never seen.
| Measure | Result |
|---|---|
| Automatic precisionOf the merges the engine made on its own, the share that were right. | 100.0% |
| Automatic recallOf the true same-person pairs, the share the engine merged on its own. | 99.6% |
| Automatic F1One score that balances the two above. | 99.8% |
| Confirmed by a reviewerSuggested merges a person has approved. | 0 |
| Awaiting reviewItems still in the review queue. | 19 |
Hypothetical, not achieved. Recall if every same-person suggestion were confirmed: 99.6%. So far 0 confirmed by a reviewer.