Lesson 04 · Jev 101
Use cases: decision slots Jev is built for
These patterns echo TypeSafe’s public use-case map and cookbooks. We don’t invent customer logos or unpublished wins. If you came from chat UIs, the “instead of prompting” lines are a translation — not the product definition.
Official category buckets come from TypeSafe’s use-case map. Rows tagged Educational are teaching hypotheses, not TypeSafe customer stories. Interactive stages for these slots live on Demo stages. Launch-week platforms and community builds — labeled, with URLs — live on Ecosystem. Chat-native habit pairs sit at the bottom and in the What’s the difference?.
Official task shapes
TypeSafe’s use-case map names ten decision shapes. Pick the shape first; the industry row is just a place to put it.
| Shape | Reach for it when | Examples |
|---|---|---|
| Classification | One known category should win | Intent, topic, department, risk type |
| Detection | A probability that one property is present | Spam, fraud, urgency, jailbreaks |
| Scoring | The answer belongs on an ordered rubric | Severity, relevance, quality, frustration |
| Routing | A category selects the next code path | Tool use, escalation, model routing |
| Search | Find items that match a natural-language query | Semantic search, document discovery |
| Retrieval | A workflow needs the most relevant context | RAG context, evidence lookup |
| Ranking | Items ordered by semantic relevance or quality | Search results, recommendations |
| Verification | An artifact checked for specific failure modes | Citations, policy, tool-call errors |
| ML feature extraction | A classical model needs semantic signals | Purchase intent, churn signals |
| Structured extraction | Known fields recovered from unstructured input | Candidate attributes, order fields |
Official industry brainstorm
These bullets are TypeSafe’s map, shortened. They are idea prompts — not a customer roster.
Search and retrieval
- Score query-to-candidate relevance
- Rerank with pairwise comparisons
- Select context for downstream AI
Scientific discovery
- Screen papers against inclusion criteria
- Label transcripts with predefined themes
- Check whether citations support claims
Model routing
- Choose which LLM gets the prompt
- Classify intent and domain
- Estimate difficulty and risk
LLM guardrails
- Jailbreak / prompt injection
- Policy and sensitive-data checks
- Tool-call and response-quality failures
Semantic code linting
- Team conventions as questions
- Run in CI
- Flag violations for review
Feature extraction
- Probabilistic features from text
- Combine with structured data
- Autoresearch against held-out ground truth
Recruiting
- Score evidence for required competencies
- Match candidates to roles
- Escalate uncertain cases
Lead generation
- ICP fit
- Buyer relevance and purchase intent
- Prioritize and route leads
Customer support
- Issue / product / intent
- Urgency, frustration, churn, refund
- Verify replies against policy
Insurance claims
- FNOL classification
- Complexity, missing info, fraud indicators
- Straight-through vs specialist
Financial crime
- Suspicious narratives and KYC
- Entity match across messy names
- Prioritize alerts
Legal and compliance
- Contracts and filings
- Missing clauses / prohibited claims
- Escalate high-risk finds
E-commerce marketplaces
- Normalize listings
- Prohibited / counterfeit / review abuse
- Rank and route uncertain listings
Moderation and trust
- Company-specific criteria
- Toxicity, spam, fraud, PII, opt-out
- Allow / warn / review / block
Advertising
- Brand safety
- Regulatory claims
- Ad-to-landing-page alignment
Gaming
- Player reports and chat
- Abuse / toxicity
- Frustration, engagement, churn
Risk assessment
- Incident and vendor notes → risk indicators
- Classify risk types
- Features for broader models
Demand forecasting
- Intent and urgency from notes and reviews
- Supply and competitive themes
- Features beside time series
Graphs
- Classify relationships and entity types
- Detect contradictions
- Hierarchical classification
Worked slots (this 101’s wording)
Background automation / smart if-statements
Classify, route, score, and gate inside product paths where a human should not sit in every loop.
Support ticket triage
Instead of prompting ChatGPT to decide which queue owns a ticket, how urgent it is, and whether the customer is furious… Send ticket state plus Choice(queue), Score(urgency), and Noul(frustration or needs_human). Route in code.
Refund eligibility gate
Instead of prompting ChatGPT to decide whether we can auto-refund from a policy plus the message… Ask a Noul on eligibility (and optionally Score abuse risk). Auto-act only above your thresholds.
Spam / phishing quarantine
Instead of prompting ChatGPT to decide if this message is a phish in one chat reply… Compose several Nouls — credentials asked, sender mismatch, reward bait — then combine in code.
Content moderation
Instead of prompting ChatGPT to moderate a community post by chatting with a policy… Choice and Score on toxicity / spam / fraud with severity plus confidence. Escalate the middle band.
Insurance FNOL triage
Instead of prompting ChatGPT to read a first-notice-of-loss narrative and assign a desk… Choice on complexity, Score on missing information and fraud indicators, then route investigators.
KYC / AML alert prioritization
Instead of prompting ChatGPT to rank alerts with a long chat summary… Score suspicious characteristics against your rubric; send high scores to investigators.
Contract clause detection
Instead of prompting ChatGPT to ask ChatGPT whether a contract is “fine”… Noul or Choice per missing clause or prohibited claim. Keep the legal memo for an LLM or a lawyer.
Marketplace listing QA
Instead of prompting ChatGPT to decide if a listing is prohibited or counterfeit in prose… Choice on category issues plus Nouls for counterfeit signals; normalize attributes in code.
Brand safety for ads
Instead of prompting ChatGPT to ask a chat model if this creative is “on brand”… Choice / Score on suitability and prohibited claims. Threshold before the ad serves.
Invoice pay / hold / return
Instead of prompting ChatGPT to have a chat model narrate what to do with an invoice… Workflow-eval scenario: typed action over invoice state. Code executes pay, hold, or return.
Security incident contain / close
Instead of prompting ChatGPT to ask ChatGPT whether to page the on-call… Workflow-eval scenario: decide contain versus close from incident state, then let runbooks fire.
Expense claim approval
Instead of prompting ChatGPT to approve meals, travel, and equipment from a chat essay… Choice against policy criteria; keep numeric checks in code. TypeSafe evals use this as a toy example.
Opt-out / DNC detection
Instead of prompting ChatGPT to have SDR chat decide whether a prospect said stop… Noul on the message. Honor opt-out in code — do not generate a follow-up when P(yes) is high.
Churn risk from ticket text
Instead of prompting ChatGPT to guess churn from a support thread in chat… Score risk from the message plus account state; alert CSMs above a threshold. Educational mapping.
Verify other AI
Judge prompts, drafts, citations, traces, and tool calls — the company highlights verify / guardrail patterns.
LLM output guardrails
Instead of prompting ChatGPT to ask ChatGPT whether its own answer is fine… Take the draft or agent trace as state. Score policy, jailbreak risk, or “needs human.”
Citation support check
Instead of prompting ChatGPT to ask GPT to re-read the source and bless the claim… Choice: does this passage support the claim? Cookbook pattern.
Agent trace review urgency
Instead of prompting ChatGPT to skim a long agent log in a chat window… Workflow-eval observability: score whether a trace needs a human now.
Podcast or report claim check
Instead of prompting ChatGPT to paste a transcript and ask “any problems?”… Noul per atomic claim against the transcript. Educational mapping, not a customer story.
Semantic code lint in CI
Instead of prompting ChatGPT to ask ChatGPT to comment on every PR… Nouls against conventions and guidelines. Fail the check on high-confidence violations.
Shell command safety class
Instead of prompting ChatGPT to let an agent decide if a command is dangerous in prose… Choice: read_only / reversible / irreversible. Educational mapping for agent harnesses.
Make harnesses smarter
Routing, retrieval, tool picking, and feature extraction around LLMs and classical ML.
Model router
Instead of prompting ChatGPT to send every request to a frontier chat model… Score difficulty or risk, then Choice the model. Cheap path versus frontier path lives in code.
Tool / skill picker
Instead of prompting ChatGPT to let an LLM narrate which tool to call… Choice over tool or skill names plus a Noul for “use a tool at all.”
RAG passage filter
Instead of prompting ChatGPT to stuff every retrieved chunk into the prompt… Noul or Score relevance and injection risk before the LLM sees the passage.
Semantic rerank
Instead of prompting ChatGPT to ask a chat model to reorder 200 results as a list… Score query–candidate relevance after BM25; sort in code.
Resume screening
Instead of prompting ChatGPT to write a ChatGPT essay about each candidate… Score fit against job criteria; escalate uncertain cases to a human.
Lead / ICP scoring
Instead of prompting ChatGPT to ask ChatGPT if this account is worth a call… Score industry fit and intent; Noul buyer relevance. Queue the SDR list in code.
Hiring competency evidence
Instead of prompting ChatGPT to grade a work sample with a chat rubric in prose… Score against a predefined competency rubric. Keep the written feedback for an LLM.
Function-calling without LLM prose
Instead of prompting ChatGPT to hope tool-call JSON from a chat model parses… Map natural language onto typed function arguments with Choice over allowed values.
Feature engineering for classical ML
Instead of prompting ChatGPT to hand-label text features for CatBoost overnight… Autoresearch cookbook: Score and Noul features from text, then train a classical model.
Forecast enrichment signals
Instead of prompting ChatGPT to summarize tickets and hope a forecast model understands the essay… Extract intent and urgency features as Score / Noul, then feed numbers to the forecast.
PR risk matrix
Instead of prompting ChatGPT to ask ChatGPT how scary this diff is… Score security and complexity per file. Educational mapping for review routing.
Inbox sponsor inquiry sort
Instead of prompting ChatGPT to triage a form inbox with one mega-prompt… Noul + Choice + Score on the form fields. Educational mapping.
Map-reduce over large corpora
Many independent questions over chunks, then aggregate in code — instead of one giant summary judgment.
Map-reduce judgments over lots of text
Instead of prompting ChatGPT to prompt ChatGPT for one giant summary judgment… Decompose into many independent questions over chunks; aggregate in code (company theme).
Hierarchical patent / product class
Instead of prompting ChatGPT to ask a chat model to invent a taxonomy path… Beam search over Choice probabilities down a predefined tree.
Date-part extraction
Instead of prompting ChatGPT to ask which date is earlier in natural language… Choice over months and days, then assemble and compare in code. Jaggedness: do not do the math in the model.
Structure recovery / autoformat
Instead of prompting ChatGPT to ask an LLM to rewrite a messy blob as JSON… Stitch candidates, then Choice block types and Noul “is this a heading?” Official extraction theme.
Entity alignment / merge decide
Instead of prompting ChatGPT to ask ChatGPT if two records are the same person… Score merge / leave / send-to-curator. Humans take the uncertain band.
Knowledge-graph relation classify
Instead of prompting ChatGPT to extract free-form relationships as sentences… Typed relation Choice over a closed relation set.
Real-time UX / games
System One–shaped calls where the UX needs roughly 100 ms-class decisions. Company demos and claims; do not overclaim game results.
Game / real-time control
Instead of prompting ChatGPT to prompt ChatGPT on every frame (too slow, too pricey)… Action Choice on structured game state. Company demos exist; do not invent win rates.
Wikiracing / high-cardinality links
Instead of prompting ChatGPT to ask a chat model to invent the next article title… Choice among many existing links without inventing new ones. Company demo theme; cardinality up to 255 (launch blog).
Smart-home intent → device action
Instead of prompting ChatGPT to chat with the house until something happens… Evaluate the user request as a Choice over allowed device actions (official demo theme).
If you want Jev inside an agent loop
Official how-to-build: System One is for AI-powered software, not agents. It does not generate code or choose its own next action. Code owns the workflow; Jev answers narrow, structured questions.
- Questions in one request are independent. One answer is not hidden context for another. Need a later option set? Make a second call.
- Keep permissions, argument validation, and side effects in application code. A typed Choice is not an ACL.
- Vercel’s agent-control page says the same in platform words: Jev cannot execute a tool; Gateway
evaluateneeds the AI SDK — compatibility endpoints do not support it.Vercel ↗PRESS - eve’s
approval: auto()helper asks Jevclearvscaution. Failed or invalid reviews pause for a human.eve KB ↗PRESS
If you want Jev to extract or align
Official habit: Jev does not invent a span and does not do calendar or ABV math. Code finds or proposes candidates; Jev picks; code copies or computes.
- Find-then-pick: pre-parsed cookbook — regex over-finds, Choice options are those spans plus
none, code copies verbatim. - Dates: date cookbook — seven Choices read the parts; code builds the
date. Confidence is the weakest used part. - Alignment: entity alignment — Score levels are the outcomes; nearest level wins; no threshold to fit.
If you want Jev to rank a corpus
Official habit: cheap search first. Then one judgment per pair. Do not dump the shortlist into a single state.
- Parallel questions: cookbook — all questions about one document ride together. Batching does not change the answers.
- Rerank: cookbook — one Noul per query–candidate pair; sort by that noul.
If you want Jev next to an LLM
Official habit: Jev judges; the LLM writes. Screen both directions. Filter retrieved passages before they become prompt. Check citations after the LLM attaches them.
- Guardrails: cookbook — Nouls + harm Score on the prompt and on the reply. Code routes pass / review / block / support.
- RAG filter: cookbook — four Nouls per passage; first matching rule wins.
- Citations: cookbook — string-match first; fabricated has no model confidence.
- Find: cookbook — Choice ranks a line;
existsNoul says whether it answers.
If you want Jev to pick UI
Official theme: Jev does not generate HTML, labels, or copy. Community recipes that call this “generative UI” still give Jev a prepared catalog and let code render the choice.
- CopilotKit cookbook: one Choice for the next prepared panel, one Score per workspace candidate; CopilotKit + AG-UI render. The app owns schemas, labels, and confirmed actions. OpenAI only handles requests outside those panels. CopilotKit ↗PRESS
- json-render’s experimental composer: Jev selects among catalog candidates and cannot invent missing prose. Uses Gateway
typesafe-ai/jev. APIs are unreleased and may change — pin exact versions. json-render ↗PRESS - LlamaIndex community packages: a fail-closed selector (Choice / Noul) and a fail-open reranker that scores one query–passage pair at a time — the official rerank habit, not a 50-passage dump. llama-index-jev ↗PRESS
- Real-time loops: keep the route and the arithmetic in code; ask Jev only at a branch.jev-plays-pokemon-red does that on PyBoy (bring your own ROM; calibration not published).jev-trade lets Jev pick hold — hold sends no Hyperliquid order. Distinct from jarrodwatts/jev-trader (Monad / Kuru). Live keys send real orders; start dry-run.jev-logtriage maps six answers to suppress / watch / review / notify / page and executes none of them.QuantDinger author tweet: Jev was added post-release as a pre-trade gate in front of the LLM gate — do not invent PASS/REJECT from the GitHub blurb; the README fetch had no durable Jev section.jev-fit.com is a closed idea-scorer; the public demo seeds random numbers with the real API shape.wakegate skips a wakeup only below 0.2 on wake.JevPlaysPokemon hands Jev only the legal FireRed / Showdown menu; an illegal answer stops the fight. Distinct from Pokémon-red (overworld branches) and PlayJev (pixels).milanboers/jev-plays-pokemon is a third Pokémon: Jev has no vision and no conversation history; the harness re-injects recent dialog. Author: opening works; battles are the weak spot.jev-mobile lets Jev pick among prevalidated Android actions — never invent a tap.hush labels a GitHub issue only above your threshold.pi-heed keeps policy in structured rules and asks Jev only what a message meant.DGP lets Jev assess; the app still has to commit.Supercov turns each file into Jev yes/no quality claims.BTK’s SEO study is their own measured pipeline (1,204 pages) — they disclose they operate it.jev-for-chrome lets Jev pick the next click in the tab you already have open — its loop kitjev-dev-kit never lets Jev invent a value.fast-jev-compaction keeps or drops tool text verbatim; it does not rewrite.evaluate MCP is one Go binary for Claude / Codex / pi.pi-jev shadows a four-question gate and fails open.jev-tools puts ten Unix commands on a pipe:
jgrep/jsort/jgate/jroutejudge; they do not write.blakestone jev-mcp keeps the key out of the agent’s context.building-with-jev is a skill that writes the Jev program, not the answer.jev-agent-browser lets a parent agent own the task; Jev only picks the next typed click.Jev Search ranks links and writes no answer.Kelbie/hunch searches code by meaning — not the Ruby gem.jselect packs evidence into a token budget.keltokhy/jsort ranks with pairwise Noul — not jev-toolsjsort.hotchpotch/jev-reranker and shinpr/jev-reranker are two different rerankers (pip 0.2 vs CLI filter 0.5).citation-verifier scores a proven quote; a human still decides.unclutter hides clutter only at both ≥ 0.9.jev-git is agit jevhook — not jev-commit.gargpratyush/jev-router and prismhq/jev-router share a name (CLI wrappers vs LiteLLM).pi-jev-router pins once;Jev-Auto-Router is per-call and a prototype.jcm-router teaches the cache-rewrite cost of switching a main chat.jev-agent-skill-router chooses a skill and never loads it.DiffJury scores a public PR; it cannot merge it.cv-screen keeps the interview line in code so a policy edit is free.email workflow routesinvoice/generaland never sends mail.new-api plugin keeps the native System One wire.pagegrade grades sections; it does not rewrite them.Foreman watches a worker; Jev never picks the next tool.Clean Code Review lets Luna write from Jev findings.JevSlop scores writing quality, not authorship.jev-bfs ranks Wikipedia titles and walks in Python.uehaj semgrep, nlgrep, and jegrep are three different meaning greps.hermes-jev-approvals lets a reducer approve or deny; Jev never executes.Akash jev-scout grounds package names, then Jev ranks.kierandotai/jev-scout is a search MCP, not that cargo CLI.jev-seo is not a free TypeSafe path.sift labels X posts and hides Junk by default.stanley-code lets Jev route; code still decides.pi-typesafe-jev exposes five tools; the key never comes from env.pi-typesafe-router starts off and must not share the session with another auto-router.nexibeo/jev-cookbook is fifteen OpenRouter recipes, not chr-kelly’s.flue-jev-demo routes through Cloudflaretypesafe/jev.omo-jevlike-router is a local clone, not hosted Jev.jev-judgment is a skill that fails open without a key.limpet is a Stop hook in shadow until you set a threshold.jev-pruner trims Bash; it does not compact the session.jlink links records; on macOS the command isjev-link.jev-curate sifts Parquet with a preset.Yoshi is a POC proxy, not a savings claim.fastbrowse lets Jev pick; the 2.9× reliability claim was retracted.NanoJev and Laya are local clones, not hosted Jev.jev-canvas draws only after policy.js says so.pi-quiet-ask registers nothing without a key.jev-browser never lets a password reach a model.Augustus places judgment; it does not prove it.robo-harness uses Jev as evaluate, not chat.pi-fast-jev-compaction is a Pi port of tamaratran, not jev-pruner.jevinci paints locally.jev-superpowers is a wrapper; ₹0 is not free TypeSafe.mastra-jev-moderation blocks at Noul ≥ 0.7 and fails open.huncho holds a page until urgency exits 0.6.hono-jev-router matches HTTP by meaning; it is not auth.Niaz jev-review scores; the agent still writes.kevinpita/pi-jev-context hides older turns unless keep > 0.8.fast-dev-compaction wraps Codex; the author says do not run it.pi-jev-compact writes a new session; the footer % can stay stale.Jev Chat for Twitch is a second column with a $0.50/hour cap.sqlite-jev batches 40 rows; the threshold stays in SQL.jev-browser-use lets Jev click and Codex type.jevkit lints questions before it spends.TheoOliveira/pi-jev is not y0usaf’s shadow gate.jkudish/jev-mcp screens; the agent still decides.pg-jev batches 20 rows, not 40.jevsql stores decisions and refreshes them.jevmail is read-only Gmail.sqlite3-jev is per-row C SQL, not mgaitan’s batch-40 extension.duckdb-jev asks one request per row, sixteen at a time.burnigtm/jev-mcp never invents arguments.jev-eval-mcp measures a threshold before you apply it.skillbox recommends skills on a 0–4 rubric, not probability.n8n-nodes-typesafe-jev is a community node; Yes/No is Noul.jev-tetris shows a keyword table beating Jev — and a shuffle check that still proves the model is steering.JevRouter lets Jev pick; the router still owns permission.BrowserClaw runs a local Jev loop; author speed is not an SLA.jevthoven asks one Choice per musical unit.opencode-jev-orchestrator keeps the parent sticky; Jev only flags.jev-tree walks past the official 255 Choice cap.jev-desktop picks the click; Codex still verifies.jev-architect designs uses; it does not call Jev.winnow hides unread blocks; it does not compact the session.grok-bot-jev classifies before Grok Bot spends a research turn.jev-eval (Shogo) measured 3.2× vs gpt-4o-mini — not the published 193.6×.4esv/jev-eval measured about 5× vs GPT-5.6 Terra — also not 193.6×.jev-skill-suggester recommends a skill; it does not run it.Soupbase lets Jev host 海龟汤 without writing the story.Jev Moderation Bot turns a pardon into the next check’s precedent.jev-go puts anotherjevon PATH.Elixir jev routes inhandle_answerguards.Jevbridge heuristic fallback is not a safety signal.jev-spam-eval shows more email context beating a thinner state.laya-mlx is local Laya, not TypeSafe.jev-local is a drop-in whose default scorer has no intelligence.LitJev is any Qwen, uncalibrated by default.minojev wins in-domain and loses OOD.Von author v2 71.5% vs published Jev 96.6%.Jevinik 30-day higher-price probability is not advice.jev_stock author T+1 45%; a 0.68 is not a 68% win rate.jev-ood-calibration shows an unknowable org rule still getting mean p 0.74 — threshold in code, and not on the confidence field.JevForge and jevbetter are local scorers, not hosted Jev.jev-reflex-autonomy-lab lets a slower planner advise; Jev does not pause.jev-web-analyzer asks ten founder questions — not a SaaS score.jev-palette ranks 77 commands with one Choice; author says run fuzzy first.typesafe-snake goes straight if Jev misses the tick.typesafe-playground shows Jev sensors, not camera pixels.s1_ruby keeps the probability and collapses late.zod-jev fails closed when Jev is unavailable.jev-voice-browser lets Jev pick a span; code types it.Safer with Jev never forwards a review or a block.JevPilot sees tables, not camera pixels — not a real AV.JevScape keeps fishing if the poll is late; the catalog carries the game.perch lints by meaning — not a merge ACL.jev-drone keeps the veto in code and only asks Jev when the scene changed.jev-shell-history ranks; it does not run the command.killmyidea lets code write KILL / FIX / SHIP from eight Scores.semdecide is grep for meaning — exit 3 is uncertain.skillranker can abstain with none-of-these.Canny lets only facts block a “done”.jevlogs keeps uncertain logs eligible for analysis.jevmeter paints probabilities, not a fact-check.Sponsor Skip lets code own the timestamps — notjev-skip.pg_typesafe batches distinct texts at 32, not one HTTP call per row.jev-eval-agent asks a 101-way Choice; done at 0.5 blocks a premature reply — not 4esv or Shogo.SlidePilot lets Jev judge completion; TypeScript is the clicker, and still-explaining is a hard veto.blink scores names; the table % is walker share, not peakedness — not jgrep.tiershift never shows Jev a price; the author 40% is 120 plain prompts, not an agentic SLA.jev-reviewer points at line ids; code copies the quote — not a clinical decision.jevmlx is a local MLX scorer; the cited 67.8% is not the same test.agent-router lets code filter eligibility first; TypeSafe only ranks what is left — not jev-router.vibecheck never shows Jev pixels; OpenAI describes images first and code writes the verdict.typesafe-ai-benchmark is a mixed Qwen vs Jev theater — not the official 193.6×.jev-benchmarks wins two BTZSC slices and leaves Emotion unresolved.jev-search-rerank-eval fuses Jev with embeddings; standalone rerank does not beat bge-m3.jev-korean-benchmark is a 100-question sample: Korean RC is fine; fine-grained meaning is shaky in both languages.zhihz/openjev is a local Qwen3-4B clone — notTheoLeeCJ/openjev and not TypeSafe.Verdict-open-jev is a second Heman10x engine, notopenJev-verdict-2.0.openvons always offers none-of-the-above; pointing an SDK at /v1/systemone does not make it hosted Jev.jev-behavior-study shows city 0/12 and Snake scores that move when you change the framing.daf-jev is an unofficial toolkit; the author ~18× is vs sequential questions, not official 193.6×.jev-experiments lists 21 app directories on GitHub — do not invent a DeepWiki catalog.commit-sentry lets code block a leaked key; Jev only judges the hunk.send-guard underlines spans regex already found; policy never downgrades a hard rule.jev-shell-guard builds the command state in code; a block Choice alone only confirms.turbo-rerank reranks 50 BM25 hits in one request on a 40-query fixture — notjev-search-rerank-eval.inbox-blitz is a Sift UI — notbohutang/sift — and sends seven questions per email; priority stays in code.jev-lint owns parse and thresholds in code; stale answers die when the line changes. Distinct fromperch.agent-assist never writes the reply; Jev only picks a macro id.log-sentinel groups stack traces in code; the 100/100 fixture is synthetic — notjevlogs.ModStream holds the message until judged; self_harm ≥0.5 is care, never a ban.jev-voice-turn keeps audio on device and fires at 0.85 + 250 ms — notjev-voice-browser.jev-ax-pilot drives macOS from the Accessibility tree, not screenshots — notJevPilot.jev-dispatch cleared 275 fictional reports in 180 s; the 40-report accuracy is vs the generator.jev-firehose re-filters stored probabilities; 300 msg/s is this key’s quota — notModStream.jev-instant-search reranks MiniSearch’s top-30 in one request — notturbo-rerank.jev-launcher is archived in the monorepo; live at dabit3/launcher. Time windows stay in code; Enter stays explicit.jev-swarm lets 32 agents share an .io arena; physics stays in code.jev-tower is a radar sim, not a real ATC — code predicts conflicts; Jev picks one instruction. Distinct fromjev-drone.judge-sheets fills a column from a header; the author’s 230–339× is vs 4 s/cell, not official 193.6×.Fern writes notes as you talk; 88.3% kind is a 180-utterance fixture.nl-palette re-ranks 66 editor commands per keystroke pause; 30/30 is the author’s 30-case bench.say needs two keys: Jev picks the Mac action; OpenAI transcribes and writes.JevPip is paper-only FX/BTC — no live orders; confidence is not a win rate.exact-paste never invents a value; a miss pastes the whole chunk.jev-analysis shows official ~100 ms is process time; the author’s Oregon e2e includes ~180 ms RTT.thumay9700/jev-plays is a fourth Pokémon Red harness; ~$1 vs $200 is README copy.jev-ecosystem-sim never shows Jev the raw numbers.JevChat builds every word as a Choice — not a ChatGPT replacement.jev-demo-router is 100 generated tickets, not a TypeSafe SLA.jev-demo-guardrails makes the verdict itself a Choice.Hexis is a third Scala door — notzio-typesafe-ai.tryAGI.TypeSafeAI is generated from official OpenAPI — notHawxy/TypeSafeAI.Net.JevTest 553/557 is one labeled run; the Haiku 688 ms is LiteLLM, not live.jev-demo-screener flips interview to reject when you swap the JD — quality scores stay high.jev-demo-moderator deletes at toxicity 2.3 and reviews at 2.0 — notModStream.jev-demo-rag kept 1 of 6 chunks on the author’s refund run.jev-demo-logictest left the belief-bias trap at 32%.Jevify is an open bench, not TypeSafe weights — 0.733 / 0.113 is the author’s Jev 1.13.0 API run.chakuho reads one local token; 52 / 2704 are its caps, not official 255.dsh-jev never throws — a fake key degrades.hipporoute-jev-codex routes Codex only at thread boundaries — notjev-codex-router.Jev Hub is a Chinese aggregator, not jev.page.jevcore stays on an offline mock until you opt in — notdsh-jev.fidecastro/jevify owns no weights — notuspraveen/Jevify.semantic-microscope labels one sentence per call; Envoy 98 ms vs Bengaluru 356 ms.Jevalyn is a Rails layer; Noul certainty is the gem’s.pi-jev-permit blocks when Jev is silent — not pi-verdict.jev-editor-skill kept 21 of 54 questions; counting stays in code.typesafe-jev-dojo runs amber SIM without a key; Stacker is one Choice after code shortlists ~10 — notjev-tetris.NRS Navigator reranks a keyword shortlist and cannot recover a statute keyword omitted — Hit@5 12/18 vs 15/18 is nominated, not lawyer gold.minhgv/jev-mcp never reads the repo; ci-shadow exits 0.arunav25/jev-mcp is one evaluate tool — bare npm jev-mcp is unrelated.mcp_jev stores the key once in ~/.mcp_jev; ping.packs ≥ 13.hj01857655/jev-router asks six questions per ticket; README $0.0578/M is not official $0.042/M.jev-receipt-classifier asks three questions per receipt and holds a category below 0.7.co1smos/jev-demo is planned historical paper-trading — no live orders, notjev-trader.jev-first-look reproduced the official jaggedness examples; SST-2 / AG News numbers stay with the author.TheWayWithin/jev-bench is a 42-claim citation bench — 85.7% is 36/42, notAbdelStark/jev-benchmarks.jev-lab without a key is a keyword demo, not Jev.jev-assist reranks files; skill install does not putjevon PATH.waygatetech/jev-go examplesjev-1— not officialjev-1.13.0.jev-triage sends confidence below 0.6 to a human; auto-close is off.UiPath AML ties 10/12 on 12 synthetic alerts — not a ranking.LogiPulse lets code own the verdict.jev_practice 5-seed zookeeper scores overlap — notjev-tetris.jev-spec asserts numbers in CI; 0.85=85% is not TypeSafe docs.jev-ultralightspeed packs a queue; 32× is pack depth — notjev-ultrafast.jonathanavis96/jev-kit fail-opens; notjevkit.JevGuard “official MCP” is JevGuard’s.jev-terrarium-dinosaurs lets code own physics — notjev-swarm.nak1b/jev-experiments asks 43 flight questions — not dabit3.jqv is stock Qwen3, not TypeSafe weights.awesomejev.cc is a sixth-plus awesome-jev index — badge 356 is the author’s, notawesomejev.com.bo7/jev_test asks two Choices on synthetic mail — noteddydong/JevTest.QA-Classifier-Jev routes Fact / Definition / Reason.fast-decision is explicit-only — it does not silently send chat.rh-guard denies structurally first — companion toAugustus, not a merge.awesome-jev-projects lists 358 as the author’s count.hellogumbo/awesome-jev isawesomejev.com — 691 is the author’s; waitlist copy is stale.jevgate can allow or stay silent; it cannot block — notJevGuard.JevLint is file-level — notmizchi/jev-lint.wellposed lints the request before a closed Choice can look perfect.MacWork reads the Accessibility tree — notjev-ax-pilot.Olti1947/jev-java andgudcks0305/jev-java are two more Java doors — notPremo-Cloud.JevSwiftSDK is a third Swift door.jevals is a local workbench — expected answers stay yours.openlayer-ai/jevals packs evals into one request — TypeSafe backend mock-tested, not live; not dayhaysoos.JevShield is a decorator gate — notJevGuard.Auto-Guard asks five questions before a tool runs — not a sandbox.should-ai-kill-us-all is a toy ruling, not a product case study.AppitStudio/awesome-jev isjevlist.ai — not awesomejev.com.tax-doc-classifier names 261 IRS forms — 0/314 is the author’s; scanned pages need OCR.typesafe-computer-use sends OCR + AX, not pixels — notMacWork.Simple Jev andopenjev-sglang speak a System One-shaped API — they are not TypeSafe weights.SemIf is the OpenJev rename — 84.5% is the author’s 102-row subset.jev-visual scores an image locally — Jev is still not vision.jevon putsjevon PATH — ninth collision.OmniJev/awesome-jev lists 119 as the author’s count.decider is an open Qwen replica — JevBench numbers are the author’s.LLM2Jev andjev-on-a-laptop score local logits — they are not TypeSafe weights.JoshuaSP/open-jev uses DiffusionGemma and never called the Jev API — notdaseinlabs/open-jev.Bespoke Nimble is a 9B LoRA — 90.12% is the author’s holdout.jev-sec-bench is a live injection bench — 96.5% is the author’s.Distill lets Jev pick a model — it cannot approve a tool.TipTour sends local labels, not pixels — notMacWork.dspy-typesafeify is three tickets, not a DSPy replacement.jev-secret-detection is a live Noul — 99/100 is the author’s.wondertwins/jev-benchmark shows FEN chess worse than random.quackd keeps--jevoff — Jev cannot write a joint angle.jev-libero andjev-robot-control are one-recording robot loops — not success rates; Jev is still not vision.RefGarden sends titles, not pixels — the hosted preview has no Jev.SmartMoney-Cub is a read-only journal — 78.43% is a toy suite, not financial advice.jev-playground has no tournament scores yet.thijmenkam/jev-benchmarks 88.2% is four small tasks — notAbdelStark.padflow-jev-evals still has an empty Jev row.JevBench 75.4 is a four-axis geometric mean — cost is $/1,000 decisions, not tokens.qwen-rlcd andLFM2.5-350M-RLCD are inference sketches — neither is TypeSafe RLCD.LFM2.5-2.6B-RLCD token PCD missed its accuracy gate.system-one-mini has five fixed heads — it cannot take your question.harshatheg never defines RLCD.eve-rlcd 0.807 is in-distribution — TypeSafe published no method.Gemma E2B RLCD scores pretrained Gemma — Jev is still not vision.decido without a key is token overlap.jod validates state first — notzod-jev.caiovicentino/jev-shield is notJevShield and is not a security boundary.n8n-nodes-typesafe-ai is notn3ndor.super-jev runs a hard rule before Jev — Jev cannot invent a tool.typesafe-ai-rs is not official — a one-level Score follows Python, not JS.typesafe-client is another unofficial Rust door, nottypesafe-sdk-rs.codeitlikemiley/typesafe-sdk-rust crate 0.1.0 is not the official Python package.Jev4Mellea lets Jev verify while Mellea still generates.typesafe-assist is notHA-Jev — Jev cannot extract free text.judging-with-typesafe cannot check a citation.a0-typesafe-ai uses the official Python SDK — keys never become tool args.typesafeai-go example state is the number 42.nibir1/typesafe-go is notzhirschtritt.n8n-nodes-jev is notn3ndor orDomMonte.jevmetrics does not see dashboards — a keep Noul is not safety.qualm will not compile without anunsurebranch.jev4k is a Kotlin DSL — notufec.jev-lens never blocks Claude;nvim never calls Jev.typesafe_api Noul confidence is library-derived.binnash core never reads env.pydantic-jev-examples is harness hooks — not the pydantic.dev TypeSafeModel door.jev-gates can stop writes;jev-lens cannot.typesafe.zig refuses an HTTPS proxy so the key never travels in plaintext.rashedInt32/jev-mcp is not tumf.jev-reach never writes text.jev-sift is notbohutang/sift.mattneel/typesafe is Hextypesafe— nottypesend.system_one_sdk replaces typesafe_sdk 0.4.1.typesafe_sdk_ex does not retry.jev-php-sdk is not binnash.Ying-Kai jev-browser is notjev-reach.y0usaf/typesafe-cli is a tenth PATHjev.geilt/typesafe-cli istypesafe, notjev.jtsang4/jev-cli is not tumf.hfiguera/typesafe_ai never reads env.jev-playwright-mcp is not a sandbox.jev-ultrafast-mcp is notjev-ultrafast.clarity-judge org TypeSafeAI is not official TypeSafe.typesafeai-dotnet-sdk is the renamed saibimajdi door — Noul has no confidence.valksor GoSystemOneAsmissing answers become zero.valksor PHPresponseModelmust be a trusted class.typesafe-ui labs are fixtures.neo4jev cannot write Cypher.Loki Autorouter is first-task, then sticky.ask-jev-skill is a Hermes tiebreaker — Hermes rules win.jev-recruiter is not vision, and a potential_match is not a hire.rhighs/jev-code is notFrancoisChastel/jev-code.jev-recall 17/18 is the author’s.madeye/pi-jev does not accelerate inference.jev-sentinel fails closed.BorisLeMeec/jev is an eleventh PATHjev— auth isTYPE_SAFE_AI_KEY.jev-dsh-decision is a third DSH door and does not execute the pick.typesafe-skill-router is opt-in and fail-open — notGodsBoy.hermes-jev-plugin is four tools, not askjev.iefnaf/pi-jev isnpm:@alexlikevibe/pi-jev— not y0usaf.actiongate-jev treats Jev as evidence; code owns authority.patdown 0.85 means violation — notpi-warden.omp-jev-compaction keeps at 0.2, not tamaratran 0.5.rajdhakad/jev-router is npmjev-model-router— expected-loss, no fallback.clear-head is a Stop hook, fail-open.git-jev-stage only stages; it never commits.riff is not on PyPI; one Noul per rule.jevex lets Jev pick the action; the writer has no tools.ailerix only acceptsailerix/auto— its local engine is not TypeSafe.construct-auto-classifier is not a sandbox; 0.6 / 0.7 are the author’s.jev-skill-gate never writesoff.jevprune is notjev-pruner.jev-decisions reviews are advice; auto is off.zhangxaochen/dsh-jev is npmdsh-jev— nottr1v3r.jev-workbench keeps the TypeSafe key on this machine; the demo is simulated.pijev is a Pi-based agent, notnpm:pi-jev.todo-jev handlers are examples, not execution.typesafe-migration-guard blocks production DANGER with HTTP 403.JevPromptCoach does not sit on the hot path.jev-pref is a semantic linter — not the twojev-reviewrepos; 0.8 / 0.7 are the author’s.agent-chaperone is not a sandbox and never auto-approves.windows-save-token-jev-setup installs a Codex skill only — installed is not trusted.jev-pr-labeler sizes by conceptual scope; dry-run still pays.jev-spring-boot-starter is not on Maven Central and does not retry.jev-skillful is npm@mrgoonie/skillfuland makes no outcome claim.last-exit is a gate game — 29/90 is the author’s, not a security bench.openpoke-meets-jev falls back to original OpenPoke without a key; its A/B table is not accuracy.rubikjev usesJEV_ENGINE_API_KEYand never shows Jev the move history.typesafe-local is local MLX, not TypeSafe weights — notjevmlx.switchboard is a gateway, not the 101 metaphor; 0.5 / 0.1 are from their table.jev-harness is not a/compactreplacement.jevegis 42/42 is the author’s labeled eval.research_desk is not investment advice; composition is code.magic-jev never acts and never reads the diff.jev-chat is notJevChat— autoregression is the wrong use.system-one-gemma is local Gemma, not TypeSafe weights.pi-agent-foreman is notthruwire/foreman.jev-phishing-bench verdict 62.6% loses to Haiku; a regex already hits 91.8% — the author’s.jev-browse is notjev-browser; Jev never emits selectors.altryne/jevify is a skill, notuspraveen/Jevifyorfidecastro/jevify.jev_local is not TypeSafe weights.jev-autopilot usesTYPESAFE_AI_API_KEY.jev-system-architect says code owns control flow.lucianfialho/jev-model-router never shows Jev the model list; 8/8 is the author’s.KamilPostrozny/pi-fast-jev-compaction is notjoelhooks.lomeshdutta/skill-router is session-start, nottypesafe-skill-router.typesafe-mod AUTO_ANSWER deny renders red.TypeSafe Code Guard line pinpoint is code, not Jev.omni-/ask-jev is Windows Codex, notask-jev-skill.dgui-hypermem Workers AI fallback is not TypeSafe.instruct-jev is a docs corpus, not weights.jev-engineering is jev-gate, notjev-gates; observe default.az9713/jev-model-router is a Gateway chat, notlucianfialhoorMandrilsquad.jev-organize never moves originals; 61-file numbers are the author’s.jev-predict-skill never runs the target skill.typeful-triage writes nothing back to GitHub; it is notashafizullah/jev-triage.omp-typesafe notes never block.stepwarden starts in audit.agent-gate-loop never merges.agent-handoff-gate never calls the API.check-risk is not on npm; without a key the report is incomplete.Frost recommends a profile and does not launch the model.hermes-jev-router plans only; it is nothermes-jev-plugin.jev-builder-loop mixes Jev with your priors; Jev is not a coder.jev-firewall fails closed; it is notstepwarden.jev-judge never echoes state and never picks a threshold.hermes-jev supervises in the background so ordinary Hermes turns do not wait.thiago-ss/jev-review never edits or merges; it is notdevagrawal09orNiaz.omp-jevens-classifier fails closed and never bypasses the native gate.pi-jev-code keeps one coding agent.twilwa/pi-typesafe starts in advisory; it is notDevMortimer/pi-typesafe.taste-lint uses a Vercel Gateway key; uncalibrated AI rules stay advisory.typesafe-demo-mcp is a thirdtypesafe-mcp; it is notitsmostafa.typesafeai-review never edits source.zcode-jev is live on LLM JSON; the Jev backend is a stub.hermes-jev-skills fails open; it is notkeeltrace/hermes-jev.jot lets Jev choose tools; a local LFM2.5 only drafts.x-scanner judges X posts in the viewport.live-jev drives a 2D car; no key still runs a labelled fallback. Notokinaaudio/live-jev.jev-ego is ego lite, not Chromejev-ultrafast.xtags labels X intent; it is notx-scanner.AskJev lets Claude drive and Jev decide on-page.aside-jev is a 0.2.0 preview.vinilana/jev-browser is a fourthjev-browser.computer-use-jev drives macOS AX; it is nottypesafe-computer-use.kyrylo jev-browse lets agent-browser act; it is not0x7067/jev-browse.jev-mail badges Gmail read-only; it is notjevmail.jev-windows-voice is Windows voice control, notjev-voice-browser.YT Time Saver covers YouTube cards.tontoko/jev-browser is a fifthjev-browser.jev-builder writes the request; it is notjev-builder-loop.jev-frontend-qa lets Jev pick; DONE is not PASS.vmendes jev-shield collapses native ads; it is notcaiovicentino/jev-shield.JevArena is click-only Snake.JevOnly uses no LLM.MCode Jev pack keeps TypeSafe on the policy leg.Otto is desktop computer-use; it is nottypesafe-computer-use.jev-turbo lets a sightmap name actions; it is notjev-ultrafast.ThinkyMiner Winnow judges feeds before you click; it is notGhalebDweikat/winnow.jev-ra creditsjev-ultrafast.Barrunto labels X and HN; it is notxtags.Cerebellum-2B is a local pointer net — its vs-Jev table is the author’s.jevcumber runs feature files; it is notjev-frontend-qa.terryds/jevplayground is a Gateway page; it is notjevplayground.com.psearch returns evidence; it is notjev-bfs.KesavanKing/jev-browser is a sixthjev-browser.Mahmoud/jev-browser is a seventhjev-browser.Page Verdict judges selling vs first-hand; it is notThinkyMiner Winnow.CorieW/JevTest is Playwright QA; it is noteddydong/JevTest.tylergibbs1/sift re-ranks Google; it is notbohutang/sift.Sloppy Jev’s blurs AI prose and ads.jeff is a local/v1/systemone— its vs-Jev table is the author’s. NotAlurith/jeff.kevinbadi/jev-voice is macOS voice; it is notjev-windows-voice.Jev Trades is paper only; it is notjev_stock.invalidate six Nouls then code; memory text is never edited.jev-mail-classifier is IMAP; it is notjev-mail.ilyamk jevMail is Apps Script Choice; it is notfazlerocks/jevmail.tumf/jev-cli is PyPIjev; it is notjtsang4/jev-cli.hev/reranker is Noul rerank; it is notjev-reranker.AI Elo Ranker critique is a closed Choice, not generated prose.jev_antispam_bot deletes Telegram spam only above the author’s 0.90 gate.jev-grug is a toy Choice chat, not a recommended chatbot.AI Provider for Jev is a WordPress plugin; it is not a .org listing.Mintzs/jevify is local Qwen, not TypeSafe; it is notuspraveen/Jevify.SmartSwitch picks one of 10 apps.BTC–Jev Signal never trades; it is notjev_stock.Jev-AV is a community scanner, not a TypeSafe security product.jev-cvss lets Jev pick metrics; code computes the score.JEV Document Classification files locally via Vercel Gateway.jevibe-check labels Bluesky text; it is not in the Web Store.commentcop rates JS/TS comments.jev-audio-beeper beeps Spanish profanity in a POC.JevTicketRouter is ticket triage; it is notjev-triage.DraftPulse scores X drafts.Emoji Jev autocompletes emoji.jev-A-share-trader is research only; it is notjev_stock.jev-resume-analyzer reviews a CV; it is notjev-demo-screener.Mimicry rewrites in your voice; Luna writes, Jev judges.pkg-gate gates npm install scripts.s1s searches a repo with typed judgments, not LLM prose.Scam Shield scores SMS scam signals; it is notjev-shield.Transcript Scorecard scores fictional ACME calls.typesafe-comment lints comments; it is notcommentcop.triage-guard is three pipelines; it is notJevTicketRouter.EmbodiedJev is a MuJoCo workbench; it is notjev-libero.mario-jev plays SMB1 from RAM; it is nottypesafe-mario.Askable arm chains sim primitives; it is notjev-libero.jev-tetris (trungdq88) plays versus Tetris; it is notthelau/jev-tetris.OneVOneJev is a 1v1 quickscope arena.jev-doom-agent picks Doom macros from structured state.JevBall runs 22 independent Choices on a 3D pitch.gg-friggin-ez screens chat toxicity; it is notjev-audio-beeper.Jev Lab (jammaru) is Hundred plus Shogi; it is notandrest04/jev-lab.jev2048 (erhanmeydan) plays 2048.io; it is not a solver.弈瞬 compares gomoku input schemes.jev-play-ping-pong plays RALLY from telemetry.jev2048 (KyleKreuter) is a parallel Choice lab; it is not a solver and noterhanmeydan/jev2048.tsai-civ2 is an experimental Civ II harness; it is notphyous/tsai-sc.TypeSafe Minecraft demo uses direct 250 ms primitives, not pathfinding.Agent-JEV-Tetris precomputes seven legal moves; it is notthelau/jev-tetris ortrungdq88/jev-tetris.BeatJev is a 25-round spam race; MOCK is not Jev.casse-brique-typesafe picks one of 16 paddle zones.Cyber-Breach runs enemy, director, and autopilot loops; the local engine is not Jev.jev-games is a Mario / Kung Fu / Doom lab aftermario-jev; Scripted is not Jev.jev-gomoku (mizchi) is a MoonBit Jev-vs-Jev CLI; it is notXieChengYuan/jev-gomoku.Jev Snake picks one direction per tick; it does not pathfind.jev-t-rex-runner picks jump, duck, or keep_running.jev-tetris (ML-Nerd) picks a verified placement ID; it is notthelau,trungdq88, orYasserbhb.jev.mods is a Minecraft design note, notellistev’s demo.P(DOOM) PROTOCOL puts Jev on p2; it is notjev-doom-agent.pong-jev asks up, down, or stay without sending digits.ps2-ai-agent maps five controller labels; the default ACCELERATE is not Jev.River Oaks picks NPC reactions; it is notRiver Run.River Run picks a lane, fire, and speed.RoverLab defaults to a local baseline; TypeSafe is optional.Shady Town moderates a living-room deduction game.SIEGE gates support-agent tool calls; the public site is a labeled simulation.snake-jev composes nine assessments; it is notiammusham/jev-snake.Terrarium picks one of eleven creature controls.typesafe-3d-chess is AI vs AI by default.typesafe-chess is Jev vs Jev; it is notmalDuffin.jevcache memoizes decisions; local/mock is not Jev.jev-me grills a plan as a design-tree interview.Instinct picks a real shadcn component; no key is a real 500.TypeSafe AI Playground is 110 examples; Enter without a key browses only. Notkavehmz.IntelliPrompter scores each talking point; 1.5 is the author’s.jev-notion fills Notion properties; ≥60% is the author’s.Jev System One lets OpenAI draft after Jev chooses the mode.jev-got lets Luna write; Jev only labels the scene.GPT vs JEV shows text beside three Nouls; presets need no API.guard-jev screens seven hazards; mock is not Jev.harden-jev-decides picks one of eleven stream ideas; it is notailerix.jev_projects hosts eight judgment sites on one domain.jev-cloud-quiz classifies brand-stripped cloud names; the bars are official probabilities.JEV Language is a BYOK substrate; the slices are not a finished language.jev-pii-checker sends scanned text to TypeSafe; it is not local-only.jev-playground (Little-Planet-Labs) is nothegargarcia.ToolGate returns allow, ask_human, or deny.typesafe-arena mirrors official docs and searches them; it is notjevarena.CartShield dispositions checkout fraud.The Council asks 36 questions; scripted demos need no key.HarnessJudge returns ok, retry, escalate, or stop.HireSignal does a resume first pass.human-compiler lets code emit rustc-style diagnostics; thresholds are opinions.Jev Data Analysis is not a live sports product; visiting never starts a paid Jev run.jev-ad-preflight is a README-only public X demo.Jev CSV Workbench keeps the CSV in the browser; 20 / 25 / 3 are the author’s.jev-bun1 runs the official JS SDK sample with Bun.jev-demos asks N maze moves; --mock is BFS, not Jev.jev-dev compares Jev and an LLM on eight Score axes; it is notjev-dev-kit.Jev Paper Judge measures presentation, not correctness.jev-should-i-apply andjev-user-jury are README-only X demos.JevPlay is notjevplayground.com.Job Risk Analyzer falls back to a mock with no key — not Jev.LaneBreak routes support tickets; it is nottoolgate.McpMatch matches a goal to an MCP catalog.Probably places no trades and is not financial advice.PulseLane heuristics without a key are not Jev.Search-Function-Test has no root README; the key never uses VITE_.SpendBrake returns continue, downgrade_model, or stop.SwarmRouter is notjev-swarm.TrustGate is indie media T&S; it is nottoolgate.typesafe-image-diffusion is not a trained diffusion model.razorback16/openjev is not TypeSafe weights and notSemIf.reflex is notjev-reflex-autonomy-lab.Jev Capability Atlas is not a leaderboard.jev-as-a-judge is a small-corpus eval, not a ranking, and notjev-judge.open-alternative-jev is not a Jev reproduction and not an official-SDK drop-in.sgoedecke/system-one is notsystem-one-open.jev-column-race is one recorded run pair, not a ranking.Jev_apps is a catalog, not a reproduction.TypeLLM is SGLang constrained decode — not TypeSafe weights.utk2103/jev-studio is the GitHub leftover of the PyPI kit — same second PATHjev.jevfire is JEV/RLCD-inspired, not TypeSafe weights and nottypesafe-mario.jev-rag-benchmark is a Turkish fixture rerank, not a first-stage retriever.jev-lm is not a product generator and noty0usaf/typesafe-cli.claude-jev is notBorisLeMeec/jev.cobusgreyling/Jev is not the official SDK.realZachi/jevtest is notCorieW/JevTest and noteddydong/JevTest.jev-codes is a fourteenth PATHjev.jev-exploration is the author’s calibration ledger, not a TypeSafe card.MongLong0214/jev-gate is noteugeniughelbur jev-gate and notjev-gates.larguesa/jev-search is notsuperagents-lab/jev-search.jev-starter is not an SDK wrapper.JevGPT is a word-level toy, not a ChatGPT competitor.ORIGIN-CIVILIZATION is not The Sims and not TypeSafe weights.siftr is notbohutang/sift, notkbhuw/jev-sift, and nottylergibbs1/sift.trade-jev is notJev Trades, notjev_stock, and notProbably. Viewer localhost:8765 is alsoLittle Airways.ZeroSweep is an inbox lab on procedural emails, not a ranking.DecisionBridge is an existing-LLM adapter, not a Jev reproduction.hemanth/jev-chess is notqibinlou/jev-chess, nottypesafe-3d-chess, and notTholeG/typesafe-chess.jev-freeform is a character-level toy, notJevGPT and not a ChatGPT competitor.jev-labs is notandrest04/jev-lab and notjammaru/jev-lab — and it is not clinical guidance.jev-routing-experiment is notjev-agent-failure-benchmark.jevchess is nothemanth/jev-chess and notqibinlou/jev-chess.jev-synergy-screening viewer localhost:8765 is also Little Airways and trade-jev.ModelSystem is an independent catalog, not a TypeSafe product.openjev-experiments is a Hugging Face NLI toy, not TypeSafe Jev and notrazorback16/openjev.typesafe-vs-deepseek is notgpt-vs-jev.FinancialPredictionJev is a Colab dump with no saved metrics — notjev-alpha-bench and not trading advice.jev-headline-bench is notjev-alpha-bench.jev-deferred-crispification is a position paper, not TypeSafe docs.jev-pick-and-place-study is not a model ranking.jev-report 50-sample Chinese numbers are not a language-wide claim.jev-shadcn-lint-eval is not shadcn-ui/lint.PocketJev is Qwen3-VL logits, not TypeSafe Jev.system-one-adapter-rust is an LLM adapter, notsystem-one-adapter-python.thaiexam-jev-charts is notkyotsu-ai-bench.shade-arena-jev-monitor 13-transcript numbers are preliminary, not a leaderboard.Typesafe_chess_eval is notTholeG/typesafe-chess and notjevchess.awesome-jev-gallery is notOmniJev/awesome-jev.typesafe-oracles four retractions stay on the page.jev-skill Mode B is agent simulation, not Jev.mizzlelover/jev-hub is notkeepwonder/jev-hub.heyjunpenn/awesome-jev is nothellogumbo/awesome-jev.fable-jev 70–120 ms / 80% / 0% / 84 ms are the author’s.Frank-ZY-Dou/awesome-jev is a robotics/3D catalog, nothellogumbo/awesome-jev.aliaihub/awesome-jev-usecases is notanandi1989.jev-radar 220+ / 218 / 108 are the author’s; keys from maintainer quota.ckaraca/awesome-jev is not hellogumbo.MrJev/awesome-jev 209 is the author’s — not the getting-started guide.LiteLLM JEV classifier 5.43× / 96% / 95.00% are LiteLLM’s, not a TypeSafe SLA.jev-robotics-demo 19.1 s / $0.0006 vs Opus 158.8 s / $0.75 are the author’s one run. NotRoboJEV.Shubham510/typesafe-go is a fourth unofficial Go client.kgonia/typesafe-sdk-java is notPremo-Cloud.jev-physical-ai 300/300 / 0.527 s / 91.3% are the author’s; OpenRouter, not native.Jev-sample 48.3% → 98.3% is the author’s question shape, not a general accuracy.n8n-nodes-typesafe is notn3ndor or DomMonte.nassim-arifette/jevgrep is a third jevgrep — unscopedjevgrepis a different project.savka777/jev-use 0.3–1.5 s/step is the author’s; notshitianfang.jev-gateway fail-open; 712 ms is README key-check copy.SystemOneHarness 5/5 bench rows are the author’s.pZacca/askjev 42/42 is the author’s first eval — notask-jev-skill.philippdubach/pi-jev-router 5/5 $0.1993 is the author’s; notmejiasd3v pin-once.davila7/jev-explained is a playground — ~100 ms is README copy.jcr returns docs; it does not execute. 85% / $0.3700 → $0.1222 are the author’s.jevvy 0.35 / 0.50 are the author’s calibration.oxlint-plugin-jev cutoffs are the author’s.tiab-review-plugin 96.1% / 36.9% / 95.0% are the author’s;TYPE_SAFE_API_KEYis not official.Pi Jev Guard 0.90 / 0.85 are the author’s; notleepokai/jev-guard.jev-ts-repl $0.20/$1.00 is user-supplied, not official $0.042/M.w3cj/jev-chat has no LLM writes.tripwire has no accuracy vs real Jev yet;TYPESAFE_AI_API_KEYis not official.triagedy ~200 ms is README copy.jev-rules 0.6 / 260–520 ms are the author’s; fail-open.compact-adviser 0.90→0.50 is the author’s; notjev-compaction.jev-pong 4.4 dps / 227 ms are the author’s; latency demo, not intelligence.lurk 0–100 is product-side, not a Jev probability.RoboJEV 10/10 / 10/10 / 8/10 are the author’s; notjev-robotics-demo.doc-router 155→87 / $0.3100→$0.1783 are the author’s.live-jev (Ableton) 20 ms / 10 ms are local, not Jev; notvinilana/live-jev.call-coach-ai asks Choice + Score + 7 Nouls in one request.abide 39/15 vs reviewer 10/11 are the author’s; fail-open.macbrow 11 s / 3.7 s / 1.6 s are the author’s; experimental.jevscan-evm 0.7 s / 0.015¢ / 80× are the author’s.crush-monitor affection score is product-side, not a Jev probability.Alurith/jeff is a read-only Go CLI, notlogan-markewich/jeff.ST-jeved 3 s / 2 s waits are the author’s; fail-open.pi-jev-router (sugarforever) 12/13 / 0.6 are the author’s; notmejiasd3vorphilippdubach.krzyzanowskim/TypeSafe is unofficial Swift 0.1.0.ai-backends keeps typesafe off text endpoints.vexjoy-agent/dis the Jev router, not/do.sharp keeps the key in the browser.orchestkitORK_TYPESAFE_API_KEYis not official; 0.8 / 93.5% are the author’s.pi-fabric 0.50 auto-approval is the author’s; Jev stays out of chat-model pickers.agent-dispatcher ships every Jev scope off; Claude-beat-Jev 143/162 vs 158/162 is the author’s.jev-social 63.969 s is the author’s one run.hermes-web-search-plus stays off untilsetup --jev; 0.95 is the author’s.JevSeek / JMPJEV_KETis not official; not fully offline.WrongStack only infersTYPESAFE_API_KEY; Jev stays out of the picker.jevkeep restores scored excerpts after Codex compaction; fail-open.open-jev-approvals 0.70 deny is the author’s; Jev down allows.forma-system1-experiment 0.8–1.6 s / $0.0005 are the author’s.jev-tool-router 0.90 is the author’s; fail to the full list.jev-stage 170–300 ms is the author’s; notgit-jev-stage.jev-browser-pilot 15/15 / 349 ms are the author’s.newsscore 310 / 97.1% are the author’s; waitlist copy is stale.jevlens keeps the key in the browser profile.typesafe-ai-go is another unofficial Go door.beat-jev 850 ms save window is the author’s.filedcom/playjev 8/8 / 0.91 s are the author’s; notOmniJev/PlayJev.decide-mcpTYPESAFE_AI_API_KEYis not official.jev-resume-disqualifier Sub-25ms / 80% / $0.0001 are README copy; notjev-resume-analyzer.Jev-Case 202 cases are the author’s.JEVRACE ten drivers / 30 races are the author’s; sitejevf1.vercel.app.tenet fail 0.8 is the author’s.trading-bot-jev Jev reads; the algorithm decides; notjev-trader.github-star-organizer-jevJEV_API_KEYis not official.J++ default demo never contacts Jev; sitetowow-ai.github.io/jpp.jev-flash-router ~150 ms is the author’s; not a jev-router.picaye/jev-compaction never summarises; 0.5 / 68–98% are the author’s.pi-jev-helm fail-open; 0.75 / 24/24 are the author’s; not a pi-jev-router.typesafe-jev-examples waitlist copy is stale.jev-dspy-lab 0.913 is the author’s; notdspy-typesafeify.kiarina/labsTYPESAFE_AI_API_KEYis not official; 16/16 / 0/43 are the author’s.jev-plugins local default is not Jev.jev-model-router (Pinutss) is not lucian / az9713 / Mandril.JevArena (rolki) Jev fail → BLOCKED; notraihankhan-rk.jev-plays-pokemon (zbloss) is the fourth leftover; notmilanboers.pi-fast-jev-compaction (QuentinDanblon) is the third leftover;JEV_API_KEYis not official; +31.8% / +37.7% are the author’s.WindTunnel Jev decides; Mercury writes; 49/49 / 25/49 are the author’s.jev-review-action recommended is advisory, not a GitHub approval.jevframe one row per request; score is a zero-based expected level.jevcache (kushals256) no key → exact-only; nothyperspaceai/jevcache.jevcheck pinsjev-1.13.0;jev-1.13/jev-1.14are fixture labels.swift-typesafe (ainame) follows official Python 0.7.0; fifth Swift leftover.typesafe-sdk-php (Butochnikov) 0.1.0 tracks official Python 0.6.0, not 0.7.0.cursor-clijev-compaction is Cursor CLI only; never writes hooks.json.dsh-jev-tools no key → inert; jev_gate escalates; 8/8 is not accuracy.jev-guard (Clemens)TYPESAFE_API_URLis not official; notleepokai/jev-guard.how-sure-is-jev is local math; 72-question / max_prob match are the author’s.ProgressGate allowAutomaticHalt default false; 0.42 is the author’s.JevsBistro keys stay in memory; guests are invented.jev-issue-radar 0/3 is retrieval, not Jev.leanest default classifier.dev is not Jev.jevtriage is notjev-pr-labeler.cua (aryaminus) replay has no LLM; Jev is stuck detection only.agentic-harness-cli hosted Jev is opt-in and grants no action authorization.jevassert checks a recording, not a live model.jev-packs 0.979 / ~250× are the author’s.jev-mcp-spring is HTTP MCP, notblakestone-x/jev-mcp orjkudish/jev-mcp.moongate 149053 / 4-of-10 are the author’s.req_llm:booleanis noul; generate_object is not Jev.git-judge-jev waitlist copy is stale; ~$0.0098 is the author’s.typesafe-screening-mcp 0.7 / $0.014 are the author’s.qingliu never auto-blocks; 0.7% is the author’s.jev_vampire_survivors is Linux-only; ~$0.36/hour is the author’s.dspy-jev is notjev-dspy-lab.linkedin-noslop fail-open; 30 examples are synthetic.slop-filter ~4¢/1,000 posts is the author’s.diff-risk-sentinel is a triage aid, not a bug detector.jev-browser-qa is notjev-browser-pilot.shipit-gate “For $0” is marketing.jev-outlook-organizer 80/80/70 are the author’s defaults.survey-qc-with-jev 92.8% PASS is the author’s.dsh-jev-guard is notdsh-jev-tools orjev-guard.work-with-jev demo is labeled presets.DriftLab is educational, not a trading signal.rocky Windows is UNTESTED.jevseo ~2 s / £0.0001 are the author’s.lgtm is not NRS-Navigator.jev-browser-chrome is notjev-ultrafast.jev-adrank +29% is the author’s simulator.jev-subagent-router tests are synthetic.jev-usher leftover is jevusher.openclaw-tool-prefilter fails open.jev-board-game offline stand-in is not Jev accuracy.jev-storyboard-lab 7/1 flagged is one author’s run.semantic-live-caption needs a GPU ASR.moomoo-jev-trader is notjev-trader.jevify (tpellet) is the fifth jevify; leftover is grevi.approval-judge-bridge fails closed.whoa judge corpus is 0.jev-debtgate 0.85 / 0.5 are the author’s.rspamd-jev is shadow only.jevdevice is notjev-android.whatitdid does not judge correctness.JEV-TheFishGame ~300 ms / $0.10/hour are the author’s.jev_playground has 0 promotions and is nothegargarcia/jev-playground.jev-typesafe-real-financial-use-cases is not a trading signal.rag-jev 40/40 / $0.0001 are the author’s.dynamic-context-engine 29/29 is post-selected.jev-crawlers ranks, it does not calibrate.jeffrey is notjeff; Laya is not TypeSafe.jevdash mock 100% clear is not Jev.InfamousCube/JevPilot is notstandardagents/jevpilot.public-browser 20.2 s / $0.0012 are the author’s n=2.clash-jev can ban an account.jev-voice-control ~0.5 s / $0.00002 are the author’s.jevbro lets Jev pick and a small LLM write text.jev-leftpad caps at 10 spaces.actionreflex 89.3 / $0.19 are the author’s.codex-cua-jev needs ChatGPT.app.jev-ads-analysis keeps Jev off until you calibrate.jev-realtime-sdk has no real actuator yet.kaustav1996/reflex is notkshetrajna12/reflex.tegersdorfer-collab/jevkit is notariel-frischer/jevkit.Jev-Checkpoint never performs the action.discoprint asks five questions per song.JevRAG ~1 s / 40 candidates is the author’s.typesafe-api-rs is unofficial.prasanthj/duckdb-jev is notcolliber/duckdb-jev.jevbus readsJEV_KEY, not officialTYPESAFE_API_KEY.loom demo needs no key.comment-opportunity-radar shares :8765 with clash-jev serve.jev-multimodal keeps hosted Jev text-only.typesafe-systemone is nottypesafe-api-rs.tphakala/jev-mcp is only-versionso far.jev-extract turns extractive QA into token start/end Choice.anuran-de/tripwire is notnoelzappy/tripwire.jev-factorio-agent “first game agent” is the author’s.is-jrecord asks Jev instead oftypeof.jev-ai-skill compaction is off by default.jev-table labels a CSV from a pack spec.jevrs is unofficial Rust with WASI.nf-jev is a beta Nextflow plugin.Talya1412/jev-harness is notantoniocoppe/jev-harness.jevi is a shell front end.jev-in-mcp lets Jev pick MCP tools.ririumu/jev-demo shares :8765.spring-ai-typesafe is an unofficial Java SDK.typesafe-ai-php is another PHP SDK.n8n-nodes-jev-classification is a fifth n8n door.jev-wide reranks past one Choice.coffee-under-fire never prefixes a secret withVITE_*.sudoku-jev shares :8080.model-switch fails open on :8317.jevfeed ranks local history.jev-newbie is a bilingual starter.jevtown shares :5191 withjammaru/jev-labShogi.jev-downloads-sorter sorts~/Downloads.open-system-one is a bench, not an open Jev.neurolink adds adecideinference type.home-assistant-typesafe routes Home Assistant intents.x-reply-filter folds X replies.tidy prunes YouTube subscriptions.DataJev steers a data agent.pi-Jev-browser picks browser steps for pi.jevmory grades agent memory quotes.jev-field-trial is a pre-registered trial.jev-browser-wingman picks one Chrome step.svitaTLCO/jev-skill is notwuyoscar/jev-skill.jev-divination-lab uses fictional fixtures only.jev-complete.nvim ranks buffer words.hearth-jev-rental-search reads four rental sites.cartpole-jev holds a pole.sf-early-learning-mcp ranks SF preschools.jev-wrapped is notjevtown.jev-skill-router is nottypesafe-skill-router.slither-jev picks slither.io routes.Jev-LLM-Playground routes support tickets.sael is notStumble/jev-go.jevtok counts Jev tokens.jfind is find plus --like.jev-grand-prix picks an F1 line.jev-2048 is noterhanmeydan/jev2048.pi-jev-effort sets thinking level.neuralint lints a diff.feed-rubric hides X posts.JohnsonRan/pi-jev is noty0usaf/pi-jev.SRTMood labels subtitle mood.playwithjev plays chess.Jev-MongoDB triages care messages.jev-torneo-animales referees a tournament.jev-hierarchy walks a topic tree.hfjev classifies HF rows.jev-hanko sorts CUAD clauses.jev-agent-hooks suggests a skill.zm2231/skill-router is notlomeshdutta/skill-router.clay-jev-people-ranker qualifies Clay founders.jev-linter-action lints files in CI.ask-jev-series weighs TV titles.opencode-jev-compaction prunes tool results.typesafe-curl is curl, not the official skill.jury.nvim is notjev-complete.nvim.hermes-slash-router routes slash tokens.the-claw lifts a night out.awesomejev.dev is notawesomejev.com.yuru-poll turns remarks into fractional votes.poker-theory-knowledge-base holds five Jev poker scenarios.jev-craft runs reflex then Jev then a slow LLM.typesafe-decision-skill does not execute Jev.jev-wechat-live reads WeChat without inject.jev-study is mock notes plus 31 tests.pi-ext-model-router routes 120 labeled prompts.xinwang-nwpu/jev-mobile is notFriedjof/jev-mobile.dsh-jev-verify measures the live API.ego-jev picks a DOM step for ego lite.fewhnhouse/jev-review-action is notfatwang2.jevmoji scores emoji groups 0–3.jev-mailroom labels mail over IMAP :1143.okooo5km/jev is another PATHjev.jev-toolkit serves MCP plus :8788.Jev-Examiner weights moderation questions.yibie/pi-jev-browser is notlaihenyi.scam-checker-with-jev scores 22 questions on SMS.mcp_typesafe is notitsmostafa.jev-fetch keeps page sections via Gateway.jevis picks Flutter test actions.jev-folio-recursive-classifier walks FOLIO with a beam.typesafe-scheduler-diagnostics explains a Pending Pod.Tech-Byte-Frontier/jevgate is notthevibeworks.ibmi-jev is Jev as Db2 for i SQL.ActionJev reviews Actions workflows.typesafe-ai-jev-example is a zero-dep Python 101.jev-town is notgaborishka/jevtown.auto-jev-codex-for-paseo routes a Paseo Codex turn.cloudforge-jev grades a lab writeup.sava-software/typesafe-client is notJedimEmO.traderai uses Jev as an optional pipeline.imsukhe/jev is notjev-ai-skill.obsidian-auto-tagger-jev tags the open note.gut is one-line likely / classify / rate.totally-tim/jev-gate is notMongLong.sahajamit/jev-lens is notrashedInt32.jev-evaluation is a pre-registered eval.ajayk/jev-go-sdk is an unofficial Go client.jevrouter-prompt-tier-extension is notBillionsBobby.k4its1t/jevlens is notknowlet.jev-agent-eval-example scores frozen traces.inbox_classification is notjev-mailroom.jev-acento is a Spanish audit.hermes-and-jev-play-minecraft picks one Mineflayer action.jev-harness-router is notantoniocoppe.x-bookmark-sorter sorts X Premium folders.jev_dart is notjevis.dsh-jev-decide is notdsh-jev-verify.gualican/jev-model-router is notlucianfialho.jev-algorithms sorts and clusters.typesafe_agent_gates gates an unattended agent.themsquared/jev-benchmark is notwondertwins.jev-state is an XState studio.doceval scores house style.pilot-typesafeai-jev routes then guards.system1-system2 escalates on low confidence.jev-crypto-scout judges news, not trades.jevmod is notJev-Moderation-Bot.jev4j is notsava-software.jevlang rewrites.jevto Python.jev-policy-engine is YAML Policy-as-Code.jev-net is notHawxy.typesafe-sdk-ruby (afurm) is notjoshmn.Mrribvar/hermes-jev-plugin is notajensenwaud.dsh-plugin-jev is notdsh-jev-decide.cursor-jev is hooks, not a Cursor model ID.jev-table-tennis picks paddle y.jev-measured logs live cost.jev-injection-bench scores public corpora.profanity-checker is a Worker Noul.zsh-jev-suggest ranks history.jeveryword returns exact spans.dsh-jev-hook is notzhangxaochen/dsh-jev.jev-verifier-skill is OpenCode Zen.system-one-playground is SOS 0.6.demo-symfony-typesafe triages tickets.jev-the-spire plays STS2.jev-flappy-bird picks FLAP or WAIT.openrouter-jev-mcp is OpenRouter only.claude-code-jev-compaction drops spent tool results.jevstudio draws funnels.zurfyx/jev-browser-skill is notjev-ultrafast.playwright-jev picks an actionId.dsh-memory-jev gates memory.jevSweeper scores Minesweeper.jevqa smokes a deployed app.jevlergy is not for people with allergies.explore-typesafe-ai is synthetic FHIR only.jev-fastloop is fail-open.jev-feels is Ruby feels?.prompt-demux dials effort.grok-jev-router is notgrok-bot-jev.jev-judge-mcp is notjev-judge.jev-rust-review reviews Rust.jev-eyes OCRs locally.jev-form-filler does not submit.jev-source-evaluator audits citations.Jev-Mem controls memory.jevfanity is notprofanity-checker.tempo-jev-demo is a local workspace.typesafe-pi judges in Pi.typesafe-jev-gate is fail-closed.quick-decision-assistant is Tauri.jev-audit needs no Jev key.mcp-server-jev is nottypesafe-mcp.typesafe-jev-mcp is notjev-judge-mcp.Jev-chooses-a-LLM is notgrok-bot-jev.obsidian-jev-archiver sorts inbox notes.Jevsus maps true / false.priorbench/jev is a pre-registered eval.jev-vs-llm runs BANKING77.lkclean cleans a LinkedIn feed.jev-writer rates published posts.fotocopiatrice reads Camera amendments.jev-reality-check probes jaggedness.Running-Dolphins/jev-bench is notTheWayWithin/jev-bench.jev-rag-benchmark is noterendikmenn.graphql-classifier lints SDL.jev-lateral-thinking-quiz hosts a quiz.jev-cite-check gates ordinance cites.PageLens ranks page passages.adaptmypage sets intent flags.jev-market-reflex is paper trading, not a bot.HA-SystemOne lives in Home Assistant.realbogart/jev is Haskell, not virolea/jev.Postmark checks a draft.jev-kit is a Gateway client plus bench.stil-lint lints Norwegian style.typesafe-jev-ruby is gem typesafe-jev.virolea/jev is gem jev, not realbogart/jev.fizzbuzz-jev asks Fizz / Buzz.jev-categorization-test compares Ollama vs Jev.boldbug1/jev-triage sorts messages in Go.tidy files Downloads.zammad-jev-dispatcher routes Zammad tickets.jev-calibration-audit audits probabilities.Jev-is-odd asks parity.jev-access-day is a learning lab.jevscope is a local workbench.longjev chunks long state.jev-systemone is an npm client.jevskill shortlists logs.jev-spam-lab reads Outlook Junk.Jev-Rug-Checker screens EVM tokens.jev-voice profiles writing.jev-atlas is a field guide.jev-classification-guide scores GTM titles.everything-about-jev is a bilingual lab.jev-technical-term explains jargon live.judgement is a Go CLI.jevymarket resolves Polymarket briefs.pi-jev-model-router picks a model for pi.jevsome-projects lists proven Jev calls.jev-usecases sketches 17 jobs.typesafe-experiments labs Claude Code / Agy.LocalSafeKidsEngine-Open gates kids traffic.openclaw-plugin-typesafe-ai triages OpenClaw turns.openclaw-typesafe-ai is one explicit decide tool.brain-speeddial-eeg-jev gates a mock order from replayed EEG.jev-decision-gateway gates when an LLM may speak.jev-workbench (Chunky83) is a desktop evidence desk.jev-vphone-cli taps a virtual iPhone.omni-scraper-mcp scores scraped evidence.openclaw-jev-compaction keeps tool calls verbatim.tab-bouncer closes off-task tabs.jev-linkedin scores a job vs a CV.jev-cyrillic-audit measures Russian vs English.AnchorLint checks whether a link means what it says.jev-graph-walk walks a graph with Score.jev-agent-toolkit is a portable Jev skill.jevoracle shows the JSON body.jev-preflop-poker plays preflop fold/call/raise.claude-code-jev-guardrails gates Claude Code tools.jev-field-tests reruns twelve scenarios.Linkedin-Slop-Detector stamps slop posts.mailverdict returns a phishing probability.opencode-intent-gate asks before vague work.jev-codex-bridge picks a Codex pair.omp-jev-extensions gates Oh My Pi done.RCIL frames the questions first.opencode-context-pruner drops stale tool results.jev-reranking reranks TREC passages.omp-jev-web drives one page.umby-jev-stack reviews a repo.jev-enterprise-decision-fabric gates tool calls in .NET.cartas-ciberseguridad-jev scores a cybersecurity CV.typesafe-computer-use-windows picks the next Windows tap.pi-jev-skill-picker ranks Pi skills.LiveSignal labels a live chat comment.jev-lab (Pasblinn) patches jev-router spend.jev-study replays Chinese and game loops.jev-1.13-mini-benchmark scores 34 support-triage cases.pi-jev-session-router pins a Pi model pair.jev-tab-grouper groups Chrome tabs.jev-domain-auctions ranks expiry domains.maestro-jev drives a simulator step.jev-master routes a support ticket.jev-playground-openrouter tries Choice / Score / Noul in the browser.pi-jev-skill-bench compares BM25 to Jev.claude-jev-warden blocks a weak SVG.jev-test (clduab11) gates a web answer.battleship-vs-jev picks the next shot.skillpick loads one skill.minesweeper-jev guesses when logic cannot.omp-greenlight skips a safe prompt.commitjev reads a commit first.pi-jev-approver gates a bash call.omp-jev-watchdog shadows a claim.pi-jev-skill-suggestion loads at most one skill.Metis labels a GitHub issue.jev-feed-filter hides a feed post.jev-router-playground picks a model.jev-x-blocker composes five Nouls.FirstScreen judges a first screen.jev-tic-tac-toe plays noughts and crosses.jobbyjev ranks companies.cc-mod-jev prunes a transcript.jev-pacman turns at a junction.astra-jev-new-vegas picks a typed action.jev-corrective-rag stubs the gates.jev-plays-tetris places a piece.jev-fanout pads one call.jev-lint (vayun) scores a claim pair.HighwayEnv-But-Jev-is-Driving picks a manoeuvre.rag-jev (Emre) filters a passage.typesafe-tetris picks a landing.jev-hands taps one element.FocusTube reveals a video.jev-labeler-action applies a label.jev-prompt-optimization rewrites a Choice.prompt2jev writes a request.jev-vs-luna-phishing triages an email.PlotVeil covers a comment.awesome-jev (AgentBuff) lists a catalog.jev-screen-mcp screens a message.jev-review-mcp gates a patch.jev-skill (abhishek) fits a placement.mcp-server-jev (codeChap) answers a typed tool.jev-decision-skill runs a payload.jev-arcade picks a legal move.JevSharp sends a typed request.jev-dev-harness ranks a file.vibecheck (psyb0t) is not deployable yet.agentic-rl draws a boundary.Jevons selects a skill.hippo-memory reranks a memory.jev-clinic-triage-eval scores a clinic call.jevgo (devbackend) sends a typed request.askgrep scores a chunk.jev-symfony-bundle guards a workflow.astrbot typesafe autoreply decides a reply.evoke picks a reflex.awesome-jev-apps lists a catalog.local-jev (amithgc) serves a local decision.AnyDecisionModel sends a typed request.decision-backend-http talks to Jev or Kev.jev-arm-lab picks a skill.jev-zork picks a command.jev-doombot picks an intent.jev-telegram-channel-router filters a post.jev-table-import-mapper maps a column.jev-is-you picks an action.Jev-Flywheel names a bias.qqbot (Mist-wu) decides a reply.commentlab scores a comment.jev-as-a-judge (sjyangkevin) judges an answer.jev-intent-review checks an intent.lus222-jev-challenge picks a mission action.jev-hub (antTing) pools a key.jevnav picks an element.node-red-contrib-jev-decisions routes a flow.jev-perfume-advisor picks a scent.awesome-jev (BeatAPI) lists a catalog.jev-mcp (Freepik) answers a typed tool.jevernetes triages a log.Jev.UI composes a request.jev-calibrate tunes a question.jev-skill-selection keeps a skill.jev-decision-lab teaches a judgment.jev-doom (olivier-motium) picks a motor.ormus-jev-cua-linux picks a desktop action.jevable-code finds a candidate.japanese-jev-lint flags a sentence.herdr-jev-router picks a harness.jev-router-dashboard replays a route.jev-chat-JARVIS ranks a reply.jev-benchmark (arifulislamat) races a ticket.jev-spire-brain picks a relic.jev-lamplight-house walks a cat.dsh-jev-adapter asks a typed tool.jev-bench (cocodedk) races a label.jev-esp32s3-gateway maps a command.jevagent (Lilithys) picks a skill.Jev_model picks a Codex tier.jevish matches a ticket.jev-agent-control picks a provider.playground-Jev scores a draft.awesome-jev (jtnkminimal) lists a catalog.awesome-jev (robokrunch) lists a measurement.Awesome-jev-use (APA) lists a use.decio gates a diff.clue ranks an issue.Maybe branches on maybe.jev-doc-classification labels a PDF.worth-replying queues a tweet.jev-team-calendar scores a resume.jev-cli (gmaxxxie) asks a gate.naiou answers yes or no.jev-news-cold-start ranks a headline.jev-support-pulse flags an outage.jev-code-review-benchmark checks a rule.chakravyuha-jev picks a move.jev-issue-pulse triages an issue.jev-eval races a label.jev-gamebenchmark picks a drop.jmarket scores a market.your-signal filters a post.jrx holds a command.jev-approval gates a shell.questions asks a typed batch.sawyer-plugins picks a thread.prism fans many questions.argus routes a template.system-one-security-triage scores a case.hermes-jev-north-star gates a finish line.slopcheck flags a tell.JevBird picks a path.jev-tweet-radar scores a post.pr-sieve fails a gate.JevFind finds a window.spend-gate decides pay.jgrep finds a behavior.jevfill fills a field.paper-radar-jev ranks a paper.ego-jev picks a click.jev-context keeps a passage.foreman-jev checks a command.azdaja judges a batch.delegador picks a model.jev-security-scan flags a skill.jev-auto-approve clears a PR.mk-jev-fly-brain picks a move.jevscan scores a token.tokengate cuts a stream.typesafe-go asks from Go.typesafe-ai-ruby asks from Ruby.jev-classifier picks a tool.jev-research routes a task.typesafe-sdk-swift asks from Swift.codex-jev-compaction keeps a pair.pi-jev-compaction drops a pair.fast-jev-compaction-pi keeps evidence.elons-job hides a reply.bluenoise filters a reply.jev-bridge caches a Jev answer.jev-snake picks a move.TrashCompact scores assistant text.ruleworld turns a category into physics.jevkit-js lints without calling Jev.awesome-jev indexes public projects.jev-shortlist fuses a cached ranking.duckdb-jev judges a row.typesafe-go asks from Go.jev-pilot picks a desktop action.jev-harness specs a kanban loop.polar_llama judges a DataFrame.jev-mcp evaluates from Pi.jevkit asks from Swift.jev-sentinel gates a tool call.awesome-jev-use-cases indexes demos.Jevest filters a finding.opencode-langsearch gates a search hit.jev-xiangqi ranks a legal move.jev-browser-use picks a click.goodall routes a turn.jev-creature-forge compiles a genome.SpecPi advises a harness.oh-my-pi names a judge provider.jev-screening-benchmark screens a title.jev-oasis routes lurk vs generate.jev-minesweeper scores a hidden cell.jev-webui batches a document.jev-radar labels intent.awesome-jev indexes usage.jevgrep filters a line.clash-jev picks a card.jevriel maps a decision.Jev-CustomerService-Demo routes a ticket.jev-gate triages a PR.Footwork arbitrates a click.pytest-jev asserts a claim.jev-foundation-models fills a struct.jev-pre-compaction prunes a tool result.arbitype gates a check.tabjev groups a tab.jev-skill-scout hints a skill.ghost-route ranks a route.lightjev picks a link.Jev-Cognigy-QA-Suite scores a transcript.dsh-fact-check grades evidence.jev-computer answers a gate.jevkit guards a tool call.awesome-jev indexes a catalog.zjev evaluates a request.typesafe-laravel routes a ticket.jev-starter asks a billing message.jev-docs observes official sources.SystemOneDotNet branches on a ticket.dsh-jev routes a tool.guideme matches a department.jevselector keeps a skill.TypeSafe.Jev switches on an enum.jev-go triages a ticket.typesafe-php scores urgency.jev classifies a ticket.jev4jobhunter screens a job.jev-connect-four picks a column.typesafe-laravel assigns a refund.TypeSafe-as-a-Judge routes a task.jev-slot-machine picks a bet.JAST flags a finding.tinystruct-typesafe picks an action.web_attack_detection_jev classifies a request.pi-jev-multi-provider gates a tool.jev asks a batch.jev-triage routes a failure.jev-architect splits a step.jev-usecases triages a ticket.gosys1 scores urgency.jev-cli gates a noul.Slop Mop scores a post.Jev Voice Browser picks a click.Jev.Sdk routes a ticket.jev_blog triages a ticket.jev-mcp asks a classifier.n8n-nodes-jev gates an email.let-jev-speak decodes a word.DroidJev taps an element.Jev demos ranks a film.jev-agent-browser picks a click.jev-mail labels a thread.pi-jev-router keeps a model.Sysone checks a reply.Jeff picks a line.Jev vs Laya scores a task.jev-awesome-skills gates a skill.overseer-judge types a pane.typesafe-mcp returns a verdict.Laya vs Jev steers a snake.typesafe-rust-sdk asks from Rust.mac-voice opens an app.Jeview maps a call.ai-type-safe-platform asks from PHP.jev-mail labels a thread.jev-mcp answers a typed question.İlan Eleme scores a listing.awesome-jev-agent-skills reviews a diff.jev-skill detects intent.jev-harness routes a review.jev-axi asks from a CLI.usher-point picks a harness.jev-patent-disclosure gates a brief.jev-figure-router picks a renderer.dual-process-ai escalates a hard call.awesome-jev lists a cookbook.jevrb asks from Ruby.docfriction scores a heading.openclaw-jev-router narrows a tool family.no-hallucination gates a quote.opnroute-jev classifies a batch.jev-trader votes BUY or WAIT.jev-lab shows a Choice call.Seek re-ranks a file.jev-tradingview-signal scores a chart.r2r-jev admits evidence.Elmridge steers a citizen.doom-war picks a macro.remember-me admits a memory.judgevet asks from Python.tink-route loads a skill.jev-router (hyspacex) binds a session.Model Picker recommends a Codex model.jev-verdict gates a publish.Readwise classifier tags an article.JEV Smart Router picks a Databricks model.node-red-contrib-jev asks from Node-RED.Jev Stories picks a line.Scribe Verify checks a sentence.hunch clicks a control.web-search picks a source.blockly-jev asks from Blockly.jev-crash-course runs a level.jev-trade-cc picks an exit.TypeSafe Keys lists a key.Jev Voice routes a spoken command.omp-classifier asks from OMP.jev-shadow scores a label.Feedlens labels a post.Jev IDS scores a flow.jev-sdk-java asks from Java.moviejev reranks a title.jev-guard lints a guardrail.jev-mcp-server exposes five tools.horse-racing-ai-jev picks a horse.Jev the Janitor votes a note.jev-certify fits a bound.Jevflix picks a film.Smart Paste fills a field.ApplyGuard audits a form.spec-ptc-jev gates an early call.jev-judge asks from a skill.jev-cli asks from Bun.jev-classification-prompting rewrites criteria.Microsoft.Extensions.AI.Jev asks from .NET.jev-safety-benchmark scores a safety case.jev-codex-router picks a Codex model.RFxCheck flags a clause.jev_codex picks a profile.Yapp routes a spoken word.jev-experiments scores a hook.typesafe-jev-opencode asks from OpenCode.awesome-jev lists a project.DocJev splits a packet.Codex Sift picks a lane.jev-triage-cn labels a short text.jevals-data scores a board.jev-deep-dive maps the evidence.awesome-jev-use-cases lists a case.dsh-typesafe-ask asks from DSH.Jev-decision-plugin asks from Codex.dsh-auto-review-jev gates a tool.Commander Oracle scores a card.Book of Answers picks an option.capbroker labels a payload.typesafe-jev-tools asks a band.typesafe-claude-kit asks from Claude.jfrader/jev-mcp asks from MCP.hermes-jev-context-engine scores a search.openjev-mcp asks through OpenJEV.Ask JEV asks a question.jevyoumean picks a subcommand.jev-oncall labels an alert.Mayi gates a tool.agy-jevgate gates a shell.jev-confidence-kit designs a question.use-jev asks from MCP.cpa-plugin-jev asks from CPA.jev-signals labels an event.rerank-bench-jev ranks a passage.jev-go asks from Go.Deslopify labels a post.jev-browser-skill picks a click.jev-mcp-server asks from MCP.jev-gtm labels a prospect.Jev DecisionOps evals a decision.jev-cli asks from PATH.maibot-jev-timing-gate gates a turn.Jev MAGI votes three times.Jev-Integrate plans a turn.jev-skill-router picks a skill.jevbrief briefs state.jev-judge-mcp judges a step.jevskillz checks a claim.jev-claw picks a model.Jev Inbox reorders unread mail.Bagel Review Board rules a bagel.AI Decision Engine routes a reply.S1Rank reranks a passage.Barq picks a click.hermes-jev-tool-gate gates a tool.jev-scheduling-evaluator judges a draft.jev-playground shapes a box.claude-herdr-jev-compaction compacts a turn.jevmem saves a decision.Cube Rule Oracle rules a dish.NaturalCodz guards a string.SwiftJev types a call.tone-your-mind scores a rewrite.mind-your-tone scores a phrase.typesafeai-cli gates a CLI.Relay hears a command.jevkit types a crate.Decision System Forge gates a bundle.ghtriage labels an issue.jev-hearth picks a word.OpenClaw catalog router shortlists a tool.WestAI OpenClaw Jev judges a turn.Jev Yarn picks a line.Customs Authority stamps a sure call.agent-fastpath gates a ship.Auto Mode for Paseo picks a model.jev-site-auditor scores a page.jev_project_context loads a record.kb-jev classifies a note.swarm triages a market.Agent to Trust judges a sample.jev-browser-skill fills a form.jev-mcp drives a tab.JevMatch scores a resume.browser-jev finds a button.Jev Autopilot sets a light.Jev Exam grades a point.Rules Are Lava judges a rule.jevelry triages a ticket.JevApi evaluates a batch.Longform Production Method gates a draft.jev-scout maps a repo.SiteClarity audits a page.Reflex Control gates a step.jev-shadow scores a hazard.Jev Flight Agent picks a click.claude-router picks a tier.Bouncer gates a tool.jev-dom clicks a button.Vampire Survivors picks a move.MedHallu flags a claim.AII codequality scores a file.Jev Replay replays a cut.jev-fanout-bench bills a call.reddit-jev-tool highlights a post.dsh-jev-interceptor shadows a tool.Tavli plays a move.BookmarX sorts a bookmark.jev-accessibility-tree rebuilds a tree.Home Jarvis picks a card.jev-pilot picks a plan.omarchy-beam finds an action.claude-jev-routing picks a tier.jev-model-router picks a model.Jev City picks a move.jev-research-pipeline screens a source.jev-skill picks a tier.stainlu/awesome-jev lists a project.jev-verify checks an identity.VotingAid matches a program.jev-coherence-bench flags a drop.Jev Gatehouse gates a tool.opencode-jev-router picks an effort.Bullshit Detector scores a post.JEV-Arcade plays a game.jev-backend-qa scores a finding.Jev Flow routes a flow.ATHENA judges a tool.firefox-jev-mcp picks a click.jevr ranks a file.jev-opus picks an effort.Doorman drops a pitch.jev-claude-statusline estimates progress.jev-dingtalk sorts a chat.Jev for Gmail ranks a thread.jev-claim-processor routes a claim.rolldown-triager labels an issue.astrbot Jev gate nominates a reply.Sieve scores a feed.jevlint lints a query.TypeSafe C++ SDK types a request.jev-precision-router picks a model.langchain-skill-router loads a skill.Kkdev92.Jev checks a contract.can-jev-bayes compares a bandit.pi-jev-router picks an effort.Slop Alarm flags a habit.jev-browser-use picks a click.Codex Task Router plans a worker.Mina’s Terrarium feels a second.Spark Jev Email Triage shadows a mailbox.Codex Subagent Router picks a profile.Paper Radar screens a listing.Zajszűrő filters a headline.jev-voice-agent hears a clause.jev-why masks a span.pi-jev-core evaluates a state.jev-phone taps a screen.Jevcast matches a request.jev-cli classifies a pipe.jev-x-filter hides a post.JEV CI Pathfinder skips a job.Flick runs a goal.jev-mcp batches a judgment.Tweet Radar scores a timeline.jev-browse checks a flow.Jev as LLM picks a word.jev-reflex pins a chunk.JEV Security Sentinel gates a finding.Jev Draft Advisor picks a Codex model.jev-hermes gates a ticket.zot-jev guards a tool call.Snap Judge compares a distribution.Jev Challenge Lab drafts a question.Jev Agentic Workflows selects a candidate.Jev X Sentiment Analysis reads a tape.jev-judge-experiment checks a CV.awesome-jev-robustness lists a robustness test.jev-in-the-wild lists a wild build.awesome-typesafe-jev lists a field guide.awesome-jev-essentials lists a shortlist.typesafe-sdk-go calls a typed API.omp-jev-decision judges a log line.JEV Switch hands a turn.jev-model-router-for-claude picks a tier.Jev Game Theory Arena plays a mixed strategy.JEV Model Navigator routes a PR.jev-cli (plurp911) asks from a shell.typesafe-playground compares inbox backends.jev-harness (cyanheads) runs a dataset.jevmcp screens spec drift.cog-typesafe binds a Cog.jev-pi-model-router picks a pi model.dsh-agent-kit checks an agent draft.jev-figma-design-plugin picks a Figma op.jev-paste fills a field.jev-security-prioritization ranks a finding.Should I Jev? says whether Jev fits.jev-decision-mcp exposes jev_decide.Jev Computer Use picks a Windows action.Let Jev Decide answers yes or no.jev-skills asks from a shell.gmail-classifier labels mail.Jev Arena picks a Pokémon move.dsh-skill-suggest suggests one skill.a2h-onus verifies a claim.crush-monitor-with-jev reads a chat.Jev from Java routes support.jevex indexes a table.jev-testing scores a chat.copilot-jev routes a task.jevkit locks a policy.跃然 StylePop picks a style.Reply Guy Repellent hides a reply.Dertek routes a coding turn.jevcompat checks a server.jev-songwriter picks a note.Jev The Spire 2 picks a move.jev-vs-llms probes a lean.jev-internal-links picks a link.stock-lite flags a split.jev-poc tours a shape.Shapeshift picks a card.ReflexGate triages a webhook.jev-cc picks a Claude tier.claude-jev-funnel keeps the uncertain band.Jev vs Laya Tetris picks a placement.go-jev asks from a pipe.JevPDF lights a line.Jev steer or queue holds a mid-turn message.jevsearch re-ranks a hit.jev-pi-orchestrator picks a worker tier.jev-skills (eran-broder) injects a skill.JEV Inbox Router picks a folder.Sensibility gates a change.jev-enforce quotes a broken rule.jev-claude-code picks a model once.Jev Intent Triage routes a ticket.SwiftJev asks from Swift.TypeSafe.Jev .NET asks from C#.jev-skillbench flags a skill.Jeverative UI picks a layout.jcol codes a column.macos-computer-use-kit guards a click.好感度分析 labels a chat.Jev-cu picks the next AX click.TypeSafe AI Playground runs a lab.awesome-jev (JohnDotOwl) lists a project.awesome-jev (MumuTW) picks a short list.TypeSafe AI Implementation Skill finds a seam.agent-workflow-typesafe-ai advises a route.kojev asks from Kotlin.pydecide chains a backend.Nightwatch triages a failed job.jev-swap shadows a decision call.taurus-jev-sdk-go asks from Go.jev-kmp asks from Kotlin.vigia scores a chat line.jev-sap-commerce holds a review.jev-triage (CMaintz) labels an issue.typesafe-ai-php-client asks from PHP.pi-jev-todo-audit checks board drift.jev-brotato picks a legal move.jev-reviewer-navigator suggests a reviewer.adk-go-typesafe hooks an ADK agent.Reclaim holds a cache dir.jev-flaky-detective labels a red test.guideme-python branches on a verdict.jev-ql filters a row.ask-jev (Aether) exposes MCP tools.Slop Fold folds a post.jev-test-intelligence picks a test group.jev-resource-rightsizer recommends a size.guideme-typescript branches on a verdict.Jong aims a paddle.pi-ask-jeff asks Jeff.jev-loop (King4s) routes a build.typesafe-jev-dotnet-sdk (mehdihadeli) calls System One.JevNet wraps a ticket.adk-go-jev-demo routes an incident.jev-cu-jp picks a control.jev-graphify names a cluster.polyjev ranks a market.yuru-come lifts a comment.jev-plays (mansicer) picks a Craftax move.AD Sentinel blurs an ad.Tabwerk finds a tab.jev-datafusion scores a row.jev-triage-benchmark (SENZO) measures a question.Invisible CAPTCHA judges a session.jev-crush labels a chat.jev-edge gates a request.xjevboost picks a column.typesafe-mcp (nirvana124) exposes MCP.minesweeper-thinking-lab picks a cell.bloxmith-jev judges a block.jev-native-mcp ranks a corpus.dsh-system1 decides in Cordis.jevqa (sairohith) drives a browser.jev-issue-router (buckmoon) picks a model.jev-mcp-middleware gates a tool.jev-lab (tanayvasishtha) audits a claim.claude-code-jev-smart-router picks a Claude model.jev-skill-selection (redreamality) keeps a skill.jev-webmcp-extension fills a tool.jev-cli (shaharia) asks from a shell.jev-code-context-router picks a symbol.Codex Jev Router picks a Codex tier.jev-browser-skill (hqman) picks a click.typesafeai-sdk-rust-community calls from Rust.typesafe-sdk-elixir calls from Elixir.pi-jev-router (cskwork) picks a Pi model.jev-model-router (lastlad) picks a LiteLLM route.jev-model-router (satviksinha) picks a Claude tier.jev-router (reallygood83) picks a role.pi-jev-router (toorop) shadows a Pi turn.TypeSafe.AI.Sdk calls from .NET.jev-lab (BrendanH18) runs a workbench.jev-llm-router picks a Codex pair.typesafe-sdk-go (blazskufca) calls from Go.typesafe-sdk-go (dwisiswant0) rejects redirects.typesafe-sdk-go (guchengod) ships a PATH CLI.jev-mail-router labels Gmail.jev-ai-sdk-form-router routes a form.pi-jev-task-router picks a Pi tier.claude-jev gates an edit.jev-recipes calls a recipe.jev_model_routers picks a model.clouter ranks a visual model.typesafe-sdk-rust (volker48) calls from Rust.jev-apply selects a saved fact.typesafe-ai-sdk (aoprisan) calls from Rust.typesafe-sdk-scala calls from Scala.hermes-switchyard picks a skill.jev-chatgpt-visibility reads a ChatGPT answer.factory-dispatch routes an incident.jev-webmcp picks a WebMCP tool.jev-cannibalization picks a page.astra-jev-harness selects context.pi-model-router picks a Pi model.jev-delegate ranks a search hit.dsh-model-router picks a Harness model.skill-picker ranks a skill.JevRev routes a query.routr reads a brief.DrissionPage-Jev-Browser-Harness-Skill picks a control.dsh-jev-prune keeps a result.dagos classifies a node.context-router scores a file.partway picks a command.gatekeeper picks a skill.stingray catches a half-done turn.sizeup picks a model size.similarity-ts-jev scores a pair.tdd-gate blames a failing test.jev-playgroud asks a typed question.jev-ai-project labels an alert.typesafe-tutorial walks a primitive.jev-inbox-queue queues a thread.vellum picks a block.toolgate vetoes a tool call.agent-seek ranks a snippet.jev-toto scores a number.jev-demo picks a control.dsh-context-curator keeps a result.convoai-jev-vad ends a turn.jev-skill picks a skill.jev-gtm-cookbook scores a title.ego-jev picks a click.jev-rubiks decides to speak.dsh-completion-supervisor labels a claim.dsh-jev-plugin batches a judgement.inbasket-triage routes a message.jev-showcase walks a primitive.tb-spam-guard flags a header.play-with-jev picks a move.jev-paper-screener keeps a paper.pi-jev-route steers a prompt.flick taps a control.jev-cua fires a spoken command.jev-approach-control clears an arrival.jev-web-agent clicks a link.langchain-jev-tutorial pages on-call.esketcher picks a paint.jev-guard holds a tool call.typesafe-rust calls Jev from Rust.jev-explained walks a typed answer.jev-hermes gates a shell command.typesafe-jev-benchmarking labels an inbox.luna-occulta-poker plays a street.trading_bot vetoes a trap.Prompt-Router keeps a cache warm.jev-router preloads three skills.ai-dj picks the next move.jev-calib repeats a judgment.jevkit gates a lead.jev-ai-use-cases triages a ticket.jev-mcp exposes a typed tool.jev-mcp forwards evaluate.llm-typesafe asks Jev from LLM.JevRPS throws rock.rust-bevy-jev-ants sends an ant.jev-relevance keeps a chunk.typesafe-ai-go asks from Go.obsidian-smart-tagger tags a note.tally-android scores a comment.bili-hardcore-jev picks an answer.jev-calculator emits a digit.jevai asks from Rust.chat2jev rewrites a chat.Jev-Anti-Netabare flags a spoiler.dsh-jev scores a turn.jev-tmmluplus-eval picks A through D.jev-search-cli ranks a hit.opencode-jev-guard judges a shell.comms-check flags a sentence.jev-playground tries a prompt.veil hides a reply.file-guide picks a folder.learn-jev-end-to-end runs a notebook.jev-wingman scores a chat.jev-skills words a question.jev-harness reviews a proposal.jev-lab filters a turn.Rime-Capsule-AI reranks a candidate.dsh-teacher-consult asks if a teacher is worth it.pairsort ranks a pair.jev-rerank-bench reranks a shortlist.typesafe-sdk-rust calls System One.pi-jev-router picks an agent.ego-jev picks a click.pvz-jev picks a plant.building-with-typesafe-jev designs a question.paperless-jev labels an inbox.jev-quant picks a trade.jev-codex-plugin reviews a claim.jevzf filters a line.jev-ios-bridge checks a screen.jev-ticket-triage triages a ticket.n8n-nodes-jev-ai routes a branch.jev-gmail-filter labels a thread.jev-desktop picks a click.opencode-auto-jev picks a model.jev asks one question.jevfilter matches a topic.jev-java asks typed questions.ralph-jev judges a loop.mastra-inbound-router scores an email.lisa reviews a diff.GeoDeploy picks a region.jev-ragcheck flags a sentence.jev-driver types a decision.unwedge flags a stall.jev-test-impact picks a test.jev-req-gate gates a requirement.jev-effort picks an effort.jev-expert reviews an integration.jev-pi-token-reduction trims a chunk.pi-typesafe-approve gates a command.jev-frontier-bench scores a decision.pi-shift-router picks a tier.jev-search-intent-classifier tags a query.semcull classifies a log.baton nudges a handoff.honeytongue judges a pitch.jev-game-engine picks a goal.pi-jev keeps a block.scrutus scores a comment.ha-jev-conversation picks a device.jermes ranks a skill.better-cheaper-llm grades a passage.jevort picks an effort tier.jev-workflow fuzzes a contract.computah picks a control.Jev-Frame schedules a judgment.jev-fast-judgements filters a pile.jev-ecc gates a tool.jev-riffs ranks a motif.jevcut picks a cut.jev-tool-lint-eval flags a sentence.jev-lifecycle shadows a path.jev-harness gates a command.JEV_ADK routes a step.jev-review scores a hunk.jev-watch notices a loop.jev-2048-experiment picks a move.opencode-jev-plugin trims a turn.jev-rust asks a question.jev-controller suggests a next step.jev-harness advises a host.voicevox-jev-proxy picks a reading.jevextract classifies a span.quicksilver shortlists a pile.jev-hub indexes a source.jev-router picks a tier.jev-browse clicks a path.system-one-connector answers a branch.jev-client calls a route.jev-fabric owns a process.x-jev-gate gates a draft.jeverifier checks a rule.jev-worker serves a decision.jev-model-routing picks a tier.dsh-jev-router sets an effort.sooth checks a claim.PRESS
Anti-patterns (keep ChatGPT)
- Write the customer email
- Refactor this module
- Brainstorm product names
- Explain quantum computing like I’m 5
- Draft a multi-step plan from a blank page
- Generate image, audio, or video understanding
All 36 habit pairs
Pedagogical mappings from the corpus. Not A/B results.Educational = writer hypothesis for Luke-style teaching.
| # | ChatGPT / LLM habit | Jev / System One pattern | |
|---|---|---|---|
| 1 | Prompt: reply yes/no if this ticket is urgent. | Ask a Noul on ticket state, then threshold on the probability. | Theme |
| 2 | Prompt: classify billing / tech / other as JSON. | Ask a Choice with your label criteria, then branch on the choice plus confidence. | Theme |
| 3 | Prompt: rate severity 1–5 and explain. | Ask a Score on your severity levels. If you need an explanation, call an LLM separately. | Edu |
| 4 | Hope JSON.parse works on the reply. | Read typed answers. TypeSafe claims there is no prose-parse step. | Theme |
| 5 | Ask ChatGPT “are you sure?” | Use the returned confidence or probability distribution as the control signal. | Theme |
| 6 | Write one mega-prompt that decides everything. | Ask many atomic questions in one call; compose the policy in code. | Theme |
| 7 | Ask for chain-of-thought, then extract the decision. | Skip generation. Ask the decision directly. | Theme |
| 8 | Fine-tune a classifier whenever labels change. | Change Choice criteria per request. TypeSafe does not offer customer LoRA. | Theme |
| 9 | Route only with embeddings similarity. | Optionally add a semantic Choice or Score rerank on candidates (cookbook theme). | Theme |
| 10 | Let an LLM agent choose the next tool in prose. | Ask a Choice over tool names, then gate on confidence. | Theme |
| 11 | Run a second GPT call to check for prompt injection. | Ask cheaper Noul / Score guardrail questions (company framing). | Theme |
| 12 | Ask the model to count keywords. | Count in code. Use a Noul per item only when the question is semantic. | Theme |
| 13 | Ask which date is earlier. | Extract parts as Choice questions, then compare in code. | Theme |
| 14 | Generate a refund email and decide the amount in one reply. | Let Jev decide eligibility. Let an LLM write the email. | Edu |
| 15 | Moderate community posts in a chat window. | Ask Choice / Score questions with policy criteria and a confidence gate. | Theme |
| 16 | Ask for a JSON confidence field. | Use native probabilities. Do not invent a confidence schema in chat. | Theme |
| 17 | Retry the call when the JSON is invalid. | TypeSafe claims schema-constrained answers cannot type-error. | Theme |
| 18 | Paste the entire CRM history into the prompt. | Filter state. Send only the fields the decision needs. | Theme |
| 19 | One question per API round-trip. | Use speculative fan-out: batch questions in one request. | Theme |
| 20 | Ask an LLM to rank 200 links by writing a list. | Score or choose among candidates; sort in code. | Theme |
| 21 | Ask the model to invent categories. | Predefine the options. Include an “other” label if you need an escape hatch. | Theme |
| 22 | Use chat inside a real-time game loop. | Typed decisions at a ~100 ms company-claimed band for System One–shaped work. | Theme |
| 23 | Verify a citation by asking GPT to re-read the passage. | Ask a citation-support Choice over the passage (cookbook pattern). | Theme |
| 24 | Route models with hand-written rules only. | Ask Jev to Score difficulty or risk, then pick the model in code. | Theme |
| 25 | Screen resumes with a ChatGPT essay. | Score against job criteria; escalate when confidence is low. | Theme |
| 26 | Detect churn risk in a free-text reply. | Score or Noul on the message plus account state. | Edu |
| 27 | Ask “what should we do?” as an open-ended question. | Offer a Choice over the actions your product is allowed to take. | Theme |
| 28 | Parse invoice fields via LLM JSON. | Generate candidates with regex or an LLM, then let Choice select. | Theme |
| 29 | LLM-lint every PR with a comment. | Run semantic lint Nouls or Scores in CI. | Theme |
| 30 | Hide rage-bait with a GPT classify call. | Ask a per-item Noul or Choice in the feed pipeline. | Edu |
| 31 | Summarize first, then label the topic. | Label on the full or filtered state with a Choice. Summaries stay with an LLM. | Edu |
| 32 | Ask whether two hex colors are similar. | Convert colors in code. Ask Jev only semantic naming questions. | Theme |
| 33 | Write a double-negative prompt: “is it not false that…” | Keep wording literal. Rewrite as a positive atomic question. | Theme |
| 34 | Trust a single chat “yes” for a wire transfer. | Set a high-stakes threshold and require a human confirm. | Theme |
| 35 | Let an agent while-loop until it feels done. | Code owns the loop. Jev only answers the judgments. | Theme |
| 36 | Ask Jev to write the customer reply. | Don’t. Use a generative model. Jev is not trained to generate text. | Theme |