jev-rental
Java tool that sorts each claim in a rental listing into verify on site, demand evidence or high-risk pitch, building a viewing checklist.
Overview
jev-rental is one of four apps in jev-suite, a set of Java decision-quality tools built on Jev with a shared kernel. It answers what must be confirmed with a listing agent before a viewing. Per the list summary, every claim in a rental listing is sorted into verify-on-site, demand-evidence or high-risk-pitch buckets, and a pre-viewing checklist is built with code-templated questions.
The shared kernel provides a TypeSafe client with retry handling, a three-way NoulGate (confident pass, confident fail, human review), and a sanitiser against prompt injection in external text. The suite's design rules are that Jev judges and code decides, the system degrades toward caution, evaluation runners refuse mock data, and missing evidence fails closed. The suite README reports for jev-rental v0.1.0 a 50-sample, 178-claim calibration with gated accuracy of 0.910 versus 0.854 raw and 0 of 10 injection flips.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | How should this listing claim be handled before viewing?(inferred) | verify_on_site · demand_evidence · high_risk_pitch |
How it works
- Extract claims from a rental listing
- Sanitise text against injection
- Ask Jev to classify each claim
- Gate uncertain results to review
- Generate a checklist with templated questions
Strengths
- Gated accuracy above raw accuracy in calibration
- Injection resistance tested
- Questions written by code, not the model
Caveats
- Small calibration sample
- Requires JDK 17+, Maven and a TypeSafe key
Getting started
With JDK 17+ and Maven, run mvn package in the jev-suite repository.At a glance
- Type
- Project
- Status
- beta
- Language
- Java
- Runtime
- JVM CLI
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Consumer rental: sorts every claim in a rental listing into verify-on-site / demand-evidence / high-risk-pitch buckets to build a pre-viewing checklist with code-templated questions; 50-sample calibration reports 0.910 gated accuracy and 0/10 injection flips.
yibie/awesome-jev
Frequently asked questions
What does Jev decide?
Which bucket each listing claim belongs to for pre-viewing checks.
How accurate is it?
The suite README reports 0.910 gated accuracy versus 0.854 raw on 178 claims.
Who writes the checklist questions?
Code templates, not the model.
Related projects
Rental-search agent that drives Chrome across Craigslist, Facebook, Redfin and Zillow, with Jev deciding which listings match.
ProjectprototypePythonOpenRouter's catalogue of decision models from several publishers behind its alpha Decisions API, with an agent-skill usage guide.
Agent skillbetaBlog benchmark testing Jev against Mistral Small and Gemini Flash-Lite at validating local event listings, with cost and latency.
BenchmarkresearchOpenRouter provider listing for typesafe/jev-1.13 alongside the moving typesafe/jev-latest alias.
Articleproduction