User repository on GitHub
AI benchmark on Japan's 2026 Common Test: Jev vs luna-none vs luna-low (static dashboard)
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Practices & Patterns
Kyotsu AI Bench
A static HTML dashboard, four commits deep, scoring Jev against two luna variants on Japan's 2026 Common Test. The repo is results only: CSV tables, one JSON file and two HTML pages, with no scripts to regenerate them.
- Stars
- 1
- Language
- HTML
- Published by
- Community
- Author
- shibadogcap
- Last verified
- Oct 2026
Highlights
- The page's own About defines the scope: Jev versus luna-none versus luna-low on Japan's 2026 Common Test.
- Root is CNAME, cost.csv, daimon.csv, subjects.csv, tracks.csv, data.json, index.html and en.html — published artifacts only.
- No generator script, no requirements file, no README and no license appear in the listing; nothing here is rerunnable.
- The repo holds 4 commits, 1 star and 0 forks, so there is no release history or earlier revision to check numbers against.
Watch out
The About line is the only prose anywhere on the page, and it names three systems without defining what luna-none and luna-low are, which endpoint or model version Jev was called at, or how Common Test items were marked. The eight root files are outputs, not code: there is no script, no data loader and no requirements.txt, so the CSV columns, the scoring rubric and the prompts behind data.json cannot be read here. The Languages and license sections show nothing, no topics are set, and with 4 commits and no tags there is nothing to compare a later run against. Copy reviewed against the repository page on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
The page does not say. The About line is the entire description and it only names 'Jev vs luna-none vs luna-low' as the three things compared on Japan's 2026 Common Test. Nothing on the repository expands the luna names into models, endpoints, dates or versions, and the Languages and license sections are empty, so you cannot tell whether they are hosted APIs, local weights or config variants. Read the comparison as three labels rather than three understood systems.
Not from what is published. The root contains CNAME, cost.csv, daimon.csv, subjects.csv, tracks.csv, data.json, index.html and en.html — all finished output. There is no harness script, no test runner, no requirements file and no README, so there is nothing to execute and no instructions for adding items. To reproduce it you would rebuild the evaluation yourself and guess at both the prompt and the scoring method.
Hard to judge from the page. There is no methodology, no sample size, no indication of whether all three systems saw identical items, and no rubric for partial credit. The repository is 4 commits with no tags and no releases, so earlier numbers cannot be compared against later ones, and data.json ships without a schema. Treat the CSVs as one author's own record rather than an audited result.
A CNAME file usually maps a custom domain, but the page never says which domain it points to and no deployment workflow appears in the listing — there is not even a .github directory in view. Both index.html and en.html exist, which suggests Japanese and English views, but nothing confirms a public URL or that the site is still serving.
The root holds CNAME, cost.csv, daimon.csv, subjects.csv, tracks.csv, data.json, index.html, and en.html. All of them are finished output; there is no script, no data loader, and no requirements file, so nothing here can be regenerated from what is published.
No. The listing shows no README, no license text, no requirements file, and no topics, and the About line naming Jev versus luna-none versus luna-low is the only prose on the page. That leaves both the reuse terms and the meaning of the CSVs unstated.
Barely. It is 4 commits deep with no tags and no releases, and there is no .github directory in view, so there is no earlier revision to compare later numbers against and no CI record. Anything shown reflects a single small run rather than a tracked series.
More Jev resources like this
0GitHub stars
Static chart pages comparing Jev against 110 other language models on Thai standardized exams: cost and speed plotted against accuracy, plus a page on whether its stated confidence predicts correctness.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
vehas/thaiexam-jev-chartsHTML
vehas/thaiexam-jev-chartsHTML24GitHub stars
Unofficial laptop study of parallel constrained decoding: prefill once, share the KV cache across fields, and decide all schema fields in one batched forward pass without writing JSON.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
rorshopping/jev-on-a-laptopPython
rorshopping/jev-on-a-laptopPython16GitHub stars
Vendor-neutral harness that tests whether Jev reranking improves a small RAG stack, using frozen hybrid BM25 plus BGE-M3 candidates and isolating retrieval, rerank, and generation stages.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
erendikmenn/jev-rag-benchmarkPython
erendikmenn/jev-rag-benchmarkPython7GitHub stars
Benchmark harness that tests whether models can forecast federal motion-to-dismiss outcomes from the written record, scored per claim-defendant pair with micro-Brier and clustered intervals.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
johnhughes3/LegalForecastBenchPython
johnhughes3/LegalForecastBenchPython6GitHub stars
Frozen 100-question-per-cell check of TypeSafe Jev on Korean versus English reading, paraphrase, and medical exam items, with recorded responses and uncertainty intervals.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
mahlernim/jev-korean-benchmarkPython
mahlernim/jev-korean-benchmarkPython7GitHub stars
Word-level language model experiment that uses Jev as its output layer, with explicit tokenizer, sampler, repetition mask, stop rules, and chunk verification.
Jev Practices & Patterns#community#jevbest#catalog-import
Community
y0usaf/jev-lmTypeScript
y0usaf/jev-lmTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
Most Jev accuracy problems are state problems. What belongs in a state, what is a pre-made judgment to leave out, and why naming the slice you ask about matters more than the wording.
Jev Practices & Patterns#state#design#primitives
Read article
A Noul returns one number and no separate confidence. How to read the value, pin its boundary with criteria, split compound questions, and set a per-action threshold that matches the cost of being wrong.
Jev Practices & Patterns#noul#primitives#confidence
Read article
A Score is a weighted position, not a ruler. Why the number can land between levels, why levels are judged separately, and how to combine several atomic scores in code instead of one crowded rubric.
Jev Practices & Patterns#score#primitives#scoring
Read article
From the Jev community
Posts from builders shipping with Jev right now.