All capability theses

A use case to evaluate

Flag records that might describe the same item

With Jev, an agent can compare two descriptions and recommend match, different or review.

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

The next test

This protocol is planned. Its outcome is not yet known.

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.

Read or improve this thesis on GitHub.