Jev calculator
Calculator with no arithmetic in its code: Jev picks each answer character from 13 options, exposing its distributions and jitter.
Overview
Jev calculator is a demo whose answer is produced by Jev one character at a time. Nothing in the code evaluates arithmetic. At each step Jev is offered the same 13 options (digits 0-9, a decimal point, a minus sign and END) given the expression and the characters so far, and the page appends whatever it picks. The interface shows, column by column, the probability Jev assigned to every option, its confidence per step, call latency and a run score that multiplies the step confidences; a Rerun button repeats the same expression to reveal call-to-call jitter, marking the first divergence.
The README is frank that Jev is not a calculator: small sums, differences, square roots and mod often come out right, but on large products it gets the first two or three digits and then stops, because once the digit distribution flattens, END beats any single digit, which is the point of the demo. The browser owns the autoregressive loop and calls a Cloudflare Worker that routes to Jev via the Vercel AI Gateway, falling back to the TypeSafe API, with Turnstile, per-IP rate limits and a kill switch. It is MIT licensed and live on Cloudflare Workers.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What is the next character of the answer? | 0 · 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 · 9 · . · - · END |
How it works
- User enters an expression
- Browser sends expression and prefix to the Worker
- Jev chooses the next character
- Append and repeat until END or the cap
- Rerun to compare jitter
Strengths
- Vividly shows how a decision model behaves on arithmetic
- Full distributions visible per step
- Rate-limited, protected public deployment
Caveats
- Deliberately not a real calculator
- Wrong on large products by design of the demo
At a glance
- Type
- Service
- Status
- prototype
- Language
- TypeScript
- Runtime
- Cloudflare Worker and browser
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Model exploration: a calculator with no arithmetic in it, asking Jev one `Choice` over 13 options (`0`–`9`, `.`, `-`, `END`) per answer character given the expression and the digits so far, appending whatever it picks and showing each step's full probability distribution and confidence, with a Rerun that exposes call-to-call jitter on identical inputs, live on Cloudflare Workers.
yibie/awesome-jevVars live in \`wrangler.jsonc\`: \`JEV\_BASE\_URL\` / \`JEV\_MODEL\` (gateway), \`TYPESAFE\_BASE\_URL\` / \`TYPESAFE\_MODEL\` (direct fallback), \`TURNSTILE\_SITEKEY\`, \`TURNSTILE\_HOSTNAMES\`, \`JEV\_DISABLED\` (kill switch). Rate limits are Workers Rate Limiting bindings: \`API\_LIMIT\` per IP and \`FALLBACK\_LIMIT\` for the direct route as a whole.
logicrw/awesome-jev-projects
Frequently asked questions
Does it compute arithmetic?
No. Jev picks each character; the code does no maths.
Why does it stop early on big numbers?
When digit probabilities flatten, END becomes the single most likely option.
What does Rerun show?
Jitter between identical calls, with the first divergence marked.
Related projects
Interactive web workbench for exploring and benchmarking Jev's Choice, Noul and Score via OpenRouter, with threshold transfer curves.
ProjectprototypeTypeScriptKorean study notes and TypeScript mock-ups of Jev concepts, with a local Laya versus GPT-5 comparison and a calibration toolkit.
ProjectresearchTypeScriptLearning project that builds choose-an-answer decision models from scratch and from Qwen, with downloadable 0.8B and 4B releases.
LibraryresearchPythonPreprint survey of Jev-style typed decision models: their research roots, adaptation costs and what current evidence supports.
ModelresearchOfficial TypeSafe docs page explaining how Jev's confidence is derived from the probability distribution and how to set thresholds.
ArticleproductionAgent skill and CLI that chains Jev question design: find decision calls, check fit, draft, lint, calibrate thresholds and plan a cascade.
Agent skillprototypePython