User repository on GitHub
Feedback on your paper in seconds.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Jev Paper Judge
Upload a PDF and get an understandability and completeness judgment back in seconds: the browser extracts passages, a Cloudflare Worker judges each one in parallel with typed System One questions and visible rubric arithmetic.
- Stars
- 1
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- JacobLinCool
- Last verified
- Oct 2026
About this project
Upload a paper as a PDF and get feedback in seconds: a structured judgment of two things. Understandability. Can a reader follow it on a first pass? Sentence clarity, defined terms and symbols, signposting, logical flow. Completeness. Does it give readers everything they need? Problem, motivation, contributions, background, reproducible method, experimental setup, interpreted results, limitations, plus mechanical checks such as undefined acronyms and unreferenced figures.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Every scored question was validated by degrading real passages and re-judging them across 649 passages from accepted papers.
- Four questions were dropped for tracking the wrong signal, one of them citation density at rho 0.54 instead of explanations.
- Findings are confidence-routed: Fix now at 0.6 and above, Worth checking between 0.3 and 0.6, otherwise kept in the raw answers.
- Counting stays in code: acronym definitions, figure references, sentence and paragraph length, section coverage.
Watch out
It measures presentation, not correctness: the README states a wrong paper can still be clear and complete. Scores are relative to accepted papers, reported as a percentile among 340 accepted NeurIPS and ICML papers and mapped so the typical accepted paper lands at 65, so they are not an absolute quality scale. The rubric, the fan-out and the API key live on the server, the client never sees which questions are asked, and the endpoint cannot be used as a generic proxy because it accepts a document rather than questions. Counting and figure checks are deterministic code, not model judgments. A passage under 70% prose is marked equation-heavy and its hard-to-follow finding is demoted, and understanding is scored on the main body only: appendix passages can raise a completeness item but never lower one. A 16-page paper costs about 19 upstream requests and 45k input tokens, roughly $0.002 at the current price. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
No, and the README is explicit about it. The two dimensions are understandability, covering sentence clarity, defined terms and symbols, signposting and logical flow, and completeness, covering problem, motivation, contributions, background, reproducible method, setup, interpreted results and limitations. It measures presentation, not correctness, so a wrong paper can still score well.
Each scored question was tested by degrading real passages in the specific way that question is meant to catch, then re-judging, across 60 accepted papers, 649 passages and 30 passages per manipulation, with unchanged and randomly shortened passages as controls. Jev is deterministic, so an unchanged passage scores identically. Shuffling paragraphs dropped logical flow by 48, deleting an acronym expansion cost 34 on that term, and four questions were cut once they proved to track something else.
No. The endpoint accepts a document, not questions, so it cannot be called with arbitrary prompts. The rubric, the fan-out and the API key all live server-side, and the client never sees which questions were asked. The browser does the extraction with pdf.js and one upload is one request to the Worker, which validates and rate-limits before fanning out.
The README puts a 16-page paper at about 19 upstream requests and 45k input tokens, roughly $0.002 at the current Jev price, finishing in two to three seconds because every passage is judged in parallel. Each passage gets a fixed 13 questions plus one per technical term, and paragraphs get a second parallel request, so the loop is short enough to edit a paragraph, re-upload and watch the score move.
Scores are relative to accepted papers: the raw means cluster tightly, so each headline is a percentile among 340 accepted NeurIPS 2024 and ICML 2024 papers, mapped to a display scale where the typical accepted paper is 65, not an absolute quality scale.
A finding that tells you to change a specific term or paragraph is shown under Fix now only when the model's confidence is at least 0.6, under Worth checking between 0.3 and 0.6, and otherwise kept in the raw answers. Counted facts always go to Fix now.
The same atomic answers are combined under three weightings, readers outside the field, readers in the field and specialists, each with its own calibration table. They differ in weights, in which terms must be explained, and in checklist emphasis, and you can switch profiles after judging.
More Jev resources like this
98GitHub stars
Self-hosted Docker service and web UI that packages Laya structured-decision inference behind a compatible System One endpoint.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
1Panel-dev/laya-serverPython
1Panel-dev/laya-serverPython23GitHub stars
Browser drone-fleet lab where TypeSafe Jev steers each drone as fast reflexes and an optional planner offers one-use strategy advice when confidence drops, without pausing control.
Jev Demos & Experiments#community#jevbest#catalog-import
23GitHub stars
Chrome extension that sends each X post to Jev with six typed questions in one call near the viewport, then renders a behavior chip under the post plus a session cost and latency panel.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
oso95/x-scannerTypeScript
oso95/x-scannerTypeScript14GitHub stars
Manifest V3 Chrome extension in TypeScript that sends each LinkedIn post to Jev and collapses bait, self-promo, off-topic, and sponsored posts into a one-line bar with reasons and scores.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
stefw/lkcleanTypeScript
stefw/lkcleanTypeScript7GitHub stars
Demo that turns free-text UI descriptions into real shadcn components using only Jev typed Choice, Noul, and Score answers, with no generated code on the decision path.
Jev Demos & Experiments#community#jevbest#catalog-import
8GitHub stars
Claude-driven browser automation where plain-English chat drives Brave or Chrome, Jev makes on-page decisions, and a guard freezes irreversible clicks.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
ranjan2829/AskJevTypeScript
ranjan2829/AskJevTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.