Perch
Perch scans code and dependency graphs with configurable System One models, including Jev, to assess issues and rank findings.
Added to Jevfast
How it uses Jev
Jev probabilities are recorded per method, merged with other findings, and converted to issue probabilities and ranking. Findings can be filtered by configured minimum for display; ranking uses the probability values.
What Jev decides
Illustrative source-defined Jev questionExample answers · not a recorded Jev response · Source ↗
Question 1 · has bug
YOUR APP
INSTRUCTION
Does `method` contain a concrete behavioral defect that a caller can reach?
STATE
State for one method includes path, method name or file, source excerpt and derived context such as metrics/callers; long methods may have partial reading metadata.
JEV · NOUL
YesNo
App workflow
- Build scan graph
Inspect repository nodes and derive applicable checks from configured rules.
- Ask questions
Evaluate active questions through the selected System One endpoint, splitting bounded work when needed.
- Rank findings
Use judgment outcomes and probabilities in the scan's ranking and report.
Why it is interesting
It applies typed judgments to a graph-based code scan and bounds question batches to fit model context.