jsort

Checked, not fully reviewed. We ran its test suite, and ran the published jev-sort 0.5.1 in Docker against a local stand-in for Jev. The stand-in answered from a hidden score we tagged in each line, so we knew the right order. No Jev calls. The README’s benchmarks, timings and costs are the author’s.

jsort sorts lines, paragraphs, files, or one CSV or JSONL field along a description such as “more urgent”. It shows Jev two texts at a time with one noul question, “Text A ranks higher than text B on this criterion”, and fits a Bradley-Terry scale to the answers. Each request carries only the two texts, cut to --max-chars, and the question. The other CSV columns were never sent. The key comes from the environment or ~/.config/jev/, goes only in the Authorization header, and is not written to the cache.

What we saw. 156 tests passed. It put our 20 lines in the hidden order and answered a rerun from the cache. A saved scale placed new lines, and refused when the stand-in began naming a different model. Bad JSON, missing or out-of-range answers and a redirect were reported as failed comparisons. A 401 or 402 stopped the run. When every third request got a 503, it retried and finished. Two rough edges. With the API down, the budget holds requests back until one has been charged, so comparisons went one at a time: 20 lines took nearly three minutes to fail, against 17 seconds with --budget none. And the error says “gave up after 15s” when it stopped after four attempts.

The HTTP, cache and budget code is in jevkit-runtime, the author’s shared library, also used by jgrep. MIT. 28 stars.

See how it compares with other tools in Best Jev tools, tested hands-on.

Review updated Oct 29, 2026. Numbers quoted from the project are its author's own; we don't publish our own measurements of Jev.