JevMade

Sign in
← Back to experiments

Benchmarks & research

jev_playground

Source screenshot of jev_playground
SOURCE SCREENSHOTFull screenshot ↗

What it does

A record of five attempts to find a worthwhile job for Jev—and the reasons each stayed out of production.

How you can use it

Before writing any code, gather examples of incoming text and note whether each sample should be allowed or flagged. A developer can borrow this testing setup to compare basic keyword matching against an AI model before launching any new feature.

Running the live AI checks requires an access key provided by TypeSafe to connect the software. While the tool scores suspicious commands or instructions, your developer cannot treat these automated warnings as proof of safety, because automated scans still miss real threats.

Maker-reported (not independently measured by JevMade)

Scoreboard of 41 verdict rows: 9 cleared, 17 held, 15 ruled out, 0 promoted, with 82 dead-end ledger entries each carrying a reopen condition (author-reported) · Substituting a well-formed random judge left 254 of 305 tests green (83%) across the vendored third-party suites (author-reported) · Tool-call error rate on allowed commands 3.95%, frozen split 4.01%, over 216k dcg decisions with no API calls (author-reported) · The shipped harm regex scored recall 12/12 and 0/38 false positives on the committed held-out corpus, where live Jev scored 11/12 and a keyword list 5/12 (author-reported)

Primitives
noul, score
Platform
TypeScript
Added
Project created
GitHub stars
1 (snapshot, not live)

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.