entity resolution / field study

resolution lab

Compare project records, inspect evidence, and test a bounded decision request.

Simulated heuristic
Explore dataset · Generate pairs first

Generate pairs first to explore this dataset.

Simulated matching rules: not Jev, not benchmark evidence. Scores are illustrative.

synthetic dataset

Pair queue

PairLeft recordRight recordDecision

No matching pairs.

decision instructions

Prompt & fields

Compared fields

Select at least one field. Demo compares field similarity only; instructions affect live Jev runs.

How to choose fields

Address, postcode, city and country identify the site. Description identifies the work and phase. Reference can confirm a planning application; developer and name are supporting evidence. A shared address does not prove two projects are the same.

Try excluding reference to test semantic matching. German Straße/Str. and umlaut transliterations, and UK St/Street, should remain compatible.

run evidence

Metrics

0 matches · 0 different · 0 for review

Completed0/0
Accuracy—
Precision—
Recall—
Coverage0%
Latency—

Accuracy: correct / completed. Precision: correct matches / predicted matches. Recall: correct matches / actual matches among completed pairs. Coverage: decided / completed. Review counts as unresolved.

Input: unknown tokens · Cost: unknown

Request preview & implementation guidance

Only selected record fields are sent. Ground truth, scenario, and browser token are excluded.

{
  "model": "typesafe/jev-1.13",
  "state": [],
  "questions": {}
}

Send at most 20 pairs per request. Treat malformed upstream answers as errors; do not create a fallback decision.