A human example
Two catalogs list the same lamp under different names, but a second pair differs by voltage.
What the caller supplies
The caller supplies IDs and relevant attributes for one pair.
What happens next
Jev gives a match recommendation. The app checks exact identifiers and preserves conflicting attributes before any merge.
Illustrative example, not a recorded result.
Potential value: medium
Could reduce manual review when records use inconsistent names. False merges make review important.
Evidence confidence: low
TypeSafe has a worked alignment example. We have not reproduced it or evaluated our generic recipe.
The rating describes support for this claim. It is separate from Jev's returned probability. How we assign ratings.
Evidence, including disagreement
- TypeSafe examples for decisions over supplied text. reference. Official examples show how to frame the question; each adaptation needs testing.
The next test
100 record pairs with human labels, near-duplicate variants and missing identifiers.
Compare against
- Exact identifiers
- Normalized string similarity
Measure
- False merges
- Missed matches
- Review volume
- Cost per resolved pair
Decision after the test
Require no increase in false merges over the chosen baseline. Keep uncertain pairs separate.
The report will retain inputs, question versions, every attempt and failure examples. We will update the confidence rating after reviewing the result.
Use a related Engine recipe
Recipes are implementation starting points. Their presence does not mean the protocol above has passed.