i built a codebase classifier with Jev and this might be the solution to overengineered code that agents create what should I test Jev on next?

what should I try to build with Jev next?

Ok, that is ACTUALLY a pretty interesting concept.

Looks interesting, I am currently using for cyclomatic/cognitive complexity, CRAP, etc. So this might be redundant, but fallow takes about 40s on my pretty small codebase so this might be quicker, will have to compare results if you publish.

Test it on a repo after 100+ agent-driven changes. The interesting question isn't whether Jev can spot bad code. It's whether it can find the complexity agents accumulate while each individual change looks reasonable.

Slop scoring text & sections of text as feedback to have the model iteratively improve writing, acting as a more reliable judge.

So kinda like SonarQube?

Damn, this is exactly what I needed to understand Jev. Thank you, man. Brilliant.

Computer use

Test Jev on day trading. Perhaps, for speed's sake, use historical data but anonymize the companies and dates.

done anything w scores? try doing a line break on blocks of text, prepending w the line number, and then asking questions that reference the line number you can make almost like gradients by line (or other block). there’s a p cool demo for this on the devsite i believe!

Send an invite please 😢

Can you make it an MCP, plug it to OpenCode and let it refactor in a loop? Would love to see the results

Very good ideas. Thanks for sharing!

My thing I built it with big pickle

Throw it at a repo after 2 years of features. One where utils is a graveyard, the same API call exists 4 times, and half the types are any. If it can point to real duplicates and explain why they should be merged without trying to rewrite the whole thing, I’m interested.

Ryan - test it on a legacy repo next?

Agents love extra folders for no reason. A cheap pass that flags "this is overbuilt" saves more time than another lint rule.

Good luck testing Jev's limits, Ryan! Can't wait to hear how it handles the AI equivalent of a teenager's first car 🚗💥

This is super cool

בנית מסווג עם Jev ושואל מה הלאה תבדוק אם הוא מתחיל להתנצל ברגע שהקלט נשמע כמו טיקט מלקוח

Test it on the same feature implemented three ways: minimal, abstraction-heavy, and with deliberate dead paths. Expose the concrete signals behind each classification. A developer should be able to identify what to delete and review the score against the code.

verify agentic payment transactions

Let it play Pokémon