# Jev Atlas

> Independent, evidence-aware research about Jev, System One models, implementation patterns, projects, claims, and build opportunities.

This index contains 93 records. Prefer official TypeSafe documentation for current API contracts. Use Jev Atlas for ecosystem context, evidence status, case studies, caveats, and authored hypotheses.

## Agent interfaces
- Full context: /llms-full.txt
- Manifest: /api/v1/manifest.json
- Search: /api/v1/search.json?q=QUERY
- MCP: /mcp
- Agent setup: /agent

## Overview
- [What is Jev?](/start): Jev is TypeSafe's System One model for fast, narrow, typed probabilistic judgments that application code can compose into workflows. [Official Documentation]

## Claim
- [Cheap decision calls could make tens or hundreds of semantic judgments per event economical.](/claims/cheap-decision-calls-could-make-tens-or-hundreds-of-semantic-judgments-per-event-economical): The product design follows from published pricing and parallel question primitives, not independent production evidence. [Speculative]
- [Developers have published small Jev integrations and demonstrations.](/claims/developers-have-published-small-jev-integrations-and-demonstrations): 36 retained posts were classified as code or demos; public repositories exist for Home Assistant, MCP, and a MAGI-style experiment. [Demonstrated]
- [Headline benchmark and reliability claims remain insufficiently independently verified.](/claims/headline-benchmark-and-reliability-claims-remain-insufficiently-independently-verified): The collected discussion overwhelmingly relays launch claims; at least one source explicitly labels the figures self-reported. [Plausible]
- [Jev exposes constrained decision primitives for Boolean probability, choice distributions, and ordered scores.](/claims/jev-exposes-constrained-decision-primitives-for-boolean-probability-choice-distributions-and-ordered-scores): The public API documentation defines `noul`, `choice`, and `score` response shapes. [Demonstrated]
- [Routing, classification, verification, and workflow control are the dominant early mental models.](/claims/routing-classification-verification-and-workflow-control-are-the-dominant-early-mental-models): Those categories recur in the collected launch discussion and align with the documented output primitives. [Plausible]
- [The strongest near-term architecture is Jev as a complement and control layer around generative models.](/claims/the-strongest-near-term-architecture-is-jev-as-a-complement-and-control-layer-around-generative-models): Jev produces decisions rather than prose; Vercel and OpenRouter expose it through evaluation-oriented endpoints, and LangChain documents it as a routing and guardrail layer inside its own harness. [Plausible]
- [Typed output removes free-form parsing but does not make wrong decisions impossible.](/claims/typed-output-removes-free-form-parsing-but-does-not-make-wrong-decisions-impossible): The API contract constrains output types; correctness and calibration are separate empirical properties. [Demonstrated]
- [TypeSafe reports a low input-token price and no metered output-token charge for Jev.](/claims/typesafe-reports-a-low-input-token-price-and-no-metered-output-token-charge-for-jev): Published pricing is echoed across launch discussion, but remains mutable vendor pricing. [Vendor Claim]
- [TypeSafe reports Jev as materially faster than LLM workflows on its own evaluations.](/claims/typesafe-reports-jev-as-materially-faster-than-llm-workflows-on-its-own-evaluations): The launch material reports large latency multiples; X discussion mostly repeats those figures. Trade press (The Register) relays the same vendor figures without independent measurement. One report from outside TypeSafe — a Vercel engineer, amplified by Vercel's chief executive — reports a large speedup measured on the reporter's own workload. [Vendor Claim]

## Project
- [1k Papers](/projects/1k-papers): A live site that classifies 1,018 AI papers against 24 topics, with Jev doing the classification after an LLM has summarised each paper. [Observed]
- [Axon Work integration](/projects/axon-work-integration): The author states that Axon Work uses Jev-style model decisions with a skill/action harness. [Observed]
- [chess-jev](/projects/chess-jev): A chess move scorer with a working FastAPI/uvicorn server, browser UI, and test suite; its primary scoring backends are local fine-tuned models, with Jev wired in as one optional hosted backend. The builder also published a companion local model, ChessJev-MiniLM-v1 (huggingface.co/gopalanj/chessjev-minilm-v1): a MiniLM-L6 encoder with an attention head, full-finetuned on Apple M1 MPS against Stockfish 19 teacher labels. [Observed]
- [Confidence-gated agent control loops](/projects/confidence-gated-agent-control-loops): Use Jev repeatedly to route, verify, and decide whether an agent should continue or escalate. [Authored Hypothesis]
- [Goblin HR](/projects/goblin-hr): A small inspectable demo that evaluates candidates for an impossible mission. [Observed]
- [HA-Jev](/projects/ha-jev): A Home Assistant custom integration exposing Jev evaluations as sensors and automation actions. [Observed]
- [Inbox triage demo](/projects/inbox-triage-demo): An inbox triage demonstration attributed to Riley Brown. Confirmed at second hand only: it appears in the madewithjev.com directory of Jev builds, and the demo itself was not opened for this record. [Observed]
- [is-odd-jev](/projects/is-odd-jev): A published npm package (v1.1.0, zero dependencies) that answers whether a number is odd by calling a Jev Noul primitive and returning its calibrated probability instead of a bare boolean. [Observed]
- [J3vRoute](/projects/j3vroute): A live, working single-page paper-trading demo: it simulates a virtual USDG portfolio traded across three markets on "Robinhood Chain," with fills clearly labelled as simulated (no swap is signed or broadcast). [Observed]
- [Jev Playground](/projects/jev-playground): A live, interactive hosted demo with three preset use cases (Support Router, Compare & Choose, Content Triage) that runs real text through Jev and renders the typed judgments and computed routing. [Observed]
- [jev-router](/projects/jev-router): An npm-published CLI wrapper that launches Claude Code or OpenAI Codex through a loopback proxy, with a documented test suite covering shared policy, both request formats, and decision display. [Observed]
- [jev-ultrafast](/projects/jev-ultrafast): A browser-automation runtime from Browser Use with a published performance report (docs/performance.md). The report documents a 7.07 s Google Flights run made with 17 Jev requests; that timing excludes browser setup and post-run verification, and the run finds flight results rather than booking anything. [Observed]
- [MAGI System on Jev](/projects/magi-system-on-jev): An open-source three-sage voting experiment inspired by Neon Genesis Evangelion. [Observed]
- [ProgressGate](/projects/progressgate): An npm package that detects semantic stagnation in agent tool-calling loops, published with tests, a quickstart, and a live demo build. [Observed]
- [SemIf (formerly OpenJev)](/projects/semif-formerly-openjev): An independent open-weights reproduction of Jev's runtime-defined typed-decision interface, running a locally hosted 4B model (e.g. Qwen3.5-4B) on a consumer GPU with a browser WebGPU demo, committed benchmark runs, and row-level output data. [Observed]
- [system-one-router](/projects/system-one-router): A published npm extension (v0.4.5) for the Pi coding agent that adds a per-turn model router; its README states it is based on and retains attribution to an existing router, yeliu84/pi-model-router. [Observed]
- [TypeSafe Agent Skills](/projects/typesafe-agent-skills): TypeSafe's own published Claude Code plugin and skills.sh package that teaches coding agents to design and wire up Jev workflows. [Observed]
- [typesafe-mcp](/projects/typesafe-mcp): An MCP server exposing TypeSafe evaluation to coding agents. [Observed]

## Pattern
- [Cascade router](/patterns/cascade-router): A cheap decision chooses whether to use rules, a small model, a premium model, a specialist, or a human. [Plausible]
- [Confidence gate](/patterns/confidence-gate): Automation proceeds above a threshold; ambiguous cases go to another check or a person. [Plausible]
- [Decision quorum](/patterns/decision-quorum): Multiple typed judgments are aggregated through voting or confidence-aware stopping. [Plausible]
- [Decision sidecar](/patterns/decision-sidecar): A typed judgment service sits beside ordinary code; code owns effects and safety boundaries. [Plausible]
- [Parallel decision matrix](/patterns/parallel-decision-matrix): One state is evaluated against many named questions, replacing repeated prompt/parse cycles. [Plausible]
- [Probabilistic predicate + deterministic action](/patterns/probabilistic-predicate-deterministic-action): Jev supplies fuzzy predicates while TypeScript, policies, and workflows execute constrained actions. [Plausible]

## Opportunity
- [Adaptive Game NPC Director](/ideas/adaptive-game-npc-director): A runtime director selecting tactics, attention, difficulty response, and group coordination every tick interval. [Authored Hypothesis]
- [Adaptive Observability Sampler](/ideas/adaptive-observability-sampler): A collector making keep/drop/escalate decisions from trace metadata and partial spans. [Authored Hypothesis]
- [Agent Loop Stopper](/ideas/agent-loop-stopper): A loop controller deciding continue, retry differently, verify, ask, or stop after every step. [Authored Hypothesis]
- [Agent Model Router](/ideas/agent-model-router): A drop-in router that selects model, reasoning effort, and fallback policy per step. [Authored Hypothesis]
- [Agent Tool Firewall](/ideas/agent-tool-firewall): A local gateway that scores every proposed tool call and requires confirmation only when risk is high. [Authored Hypothesis]
- [API Abuse Shape Detector](/ideas/api-abuse-shape-detector): An edge middleware assigning abuse type, severity, and response tier to suspicious request sequences. [Authored Hypothesis]
- [Autonomous QA Scenario Selector](/ideas/autonomous-qa-scenario-selector): A test controller choosing the next action and deciding when a distinct failure has been found. [Authored Hypothesis]
- [Calendar Commitment Guard](/ideas/calendar-commitment-guard): A scheduling guard that classifies each invitation and proposes accept, decline, delegate, or request agenda. [Authored Hypothesis]
- [CI Review Triage](/ideas/ci-review-triage): A GitHub check that assigns review depth, owners, and required test classes to every pull request. [Authored Hypothesis]
- [Context Window Admission Controller](/ideas/context-window-admission-controller): A component scoring every candidate memory, file, or message before context assembly. [Authored Hypothesis]
- [Continuous Form Friction Controller](/ideas/continuous-form-friction-controller): A form engine deciding which question to show next, skip, verify, or escalate. [Authored Hypothesis]
- [Data Pipeline Quarantine Router](/ideas/data-pipeline-quarantine-router): A stream processor choosing accept, normalize, quarantine, retry, or human review per record. [Authored Hypothesis]
- [Decision Regression Harness](/ideas/decision-regression-harness): A test runner recording typed decisions and calibration metrics over versioned scenario suites. [Authored Hypothesis]
- [Dependency Update Gate](/ideas/dependency-update-gate): A bot deciding auto-merge, test-more, human-review, defer, or block for every update. [Authored Hypothesis]
- [Home Automation Intent Layer](/ideas/home-automation-intent-layer): A Home Assistant layer turning state changes into typed comfort, safety, and automation decisions. [Authored Hypothesis]
- [Inbox Micro-Router](/ideas/inbox-micro-router): An email client companion assigning folder, urgency, next-action, and safe automation eligibility. [Authored Hypothesis]
- [LLM Output Escalation Mesh](/ideas/llm-output-escalation-mesh): A mesh of cheap per-claim decisions selecting accept, recheck, regenerate, retrieve, or ask a human. [Authored Hypothesis]
- [Local Activity Auto-Timeline](/ideas/local-activity-auto-timeline): A private desktop timeline deciding whether each app/window event starts, continues, or ends an activity. [Authored Hypothesis]
- [MCP Capability Router](/ideas/mcp-capability-router): A proxy exposing only the small capability subset relevant to the current step. [Authored Hypothesis]
- [Meeting Participation Router](/ideas/meeting-participation-router): A planner assigning attend, async input, optional, or no-action per person and agenda item. [Authored Hypothesis]
- [Micro-Approval SDK](/ideas/micro-approval-sdk): An SDK and embeddable UI for approve/deny/escalate flows backed by typed probabilistic decisions. [Authored Hypothesis]
- [Multi-Agent Vote Calibrator](/ideas/multi-agent-vote-calibrator): A coordinator deciding when votes agree enough, need another specialist, or require human escalation. [Authored Hypothesis]
- [Notification Interruptibility Engine](/ideas/notification-interruptibility-engine): A macOS/iOS layer that decides deliver, batch, summarize-later, or suppress for every event. [Authored Hypothesis]
- [Personal File Attention Index](/ideas/personal-file-attention-index): A local index that scores files, downloads, screenshots, and documents for urgency and project relevance. [Authored Hypothesis]
- [Privacy-Preserving Screen Event Filter](/ideas/privacy-preserving-screen-event-filter): A local preflight layer deciding whether a screen change is useful, sensitive, or ignorable. [Authored Hypothesis]
- [Probabilistic Rules Engine](/ideas/probabilistic-rules-engine): An open-source engine combining Jev judgments with auditable deterministic conditions. [Authored Hypothesis]
- [Semantic Background-Task Scheduler](/ideas/semantic-background-task-scheduler): A scheduler choosing run-now, defer, batch, cancel, or request power/network for each task. [Authored Hypothesis]
- [Semantic Feature-Flag Allocator](/ideas/semantic-feature-flag-allocator): A policy layer selecting safe UI variants based on session intent, friction, and risk. [Authored Hypothesis]
- [Smart Retry Policy](/ideas/smart-retry-policy): A library choosing retry, backoff class, fallback, or dead-letter from structured failure context. [Authored Hypothesis]
- [Support Action Gate](/ideas/support-action-gate): A decision layer approving refunds, credits, escalations, and account changes within explicit limits. [Authored Hypothesis]
- [Webhook Decision Bus](/ideas/webhook-decision-bus): A developer service mapping each event to typed route, priority, and action decisions. [Authored Hypothesis]

## Evidence
- [Announcement, Technical Explanation, Project, Cost source — relevance 76](/evidence): A cached X post classified as Announcement, Technical Explanation, Project, Cost. Themes: typed, workflow, system one. [Source Evidence]
- [Announcement, Technical Explanation, Project, Performance source — relevance 77](/evidence): A cached X post classified as Announcement, Technical Explanation, Project, Performance. Themes: typed, confidence, classifier. [Source Evidence]
- [Code, Project source — relevance 74](/evidence): A cached X post classified as Code, Project. Themes: system one. [Source Evidence]
- [Demo, Code, Project, Routing, Classification, Cost source — relevance 74](/evidence): A cached X post classified as Demo, Code, Project, Routing, Classification, Cost. Themes: api. [Source Evidence]
- [Project source — relevance 75](/evidence): A cached X post classified as Project. Themes: confidence, system one. [Source Evidence]
- [Project, Routing, Question source — relevance 75](/evidence): A cached X post classified as Project, Routing, Question. Themes: confidence, system one. [Source Evidence]
- [Technical Explanation, Architecture source — relevance 79](/evidence): A cached X post classified as Technical Explanation, Architecture. Themes: architecture, typed, confidence, rlcd, system one, noul. [Source Evidence]
- [Technical Explanation, Architecture, Automation source — relevance 76](/evidence): A cached X post classified as Technical Explanation, Architecture, Automation. Themes: typed, confidence, workflow, agent, system one. [Source Evidence]
- [Technical Explanation, Architecture, Code, Project source — relevance 73](/evidence): A cached X post classified as Technical Explanation, Architecture, Code, Project. Themes: architecture, typed, rlcd, system one. [Source Evidence]
- [Technical Explanation, Architecture, Code, Project, Performance source — relevance 78](/evidence): A cached X post classified as Technical Explanation, Architecture, Code, Project, Performance. Themes: typed, latency. [Source Evidence]
- [Technical Explanation, Architecture, Cost source — relevance 73](/evidence): A cached X post classified as Technical Explanation, Architecture, Cost. Themes: architecture, typed, confidence, rlcd, system one. [Source Evidence]
- [Technical Explanation, Code, Performance source — relevance 75](/evidence): A cached X post classified as Technical Explanation, Code, Performance. Themes: typed, latency, system one. [Source Evidence]
- [Technical Explanation, Code, Project, Routing source — relevance 82](/evidence): A cached X post classified as Technical Explanation, Code, Project, Routing. Themes: api, rlcd, agent, system one. [Source Evidence]
- [Technical Explanation, Demo, Project, Question source — relevance 74](/evidence): A cached X post classified as Technical Explanation, Demo, Project, Question. Themes: typed. [Source Evidence]
- [Technical Explanation, Limitation source — relevance 74](/evidence): A cached X post classified as Technical Explanation, Limitation. Themes: typed, workflow, agent, system one, failure. [Source Evidence]
- [Technical Explanation, Performance source — relevance 74](/evidence): A cached X post classified as Technical Explanation, Performance. Themes: typed, confidence, system one. [Source Evidence]
- [Technical Explanation, Performance, Cost source — relevance 75](/evidence): A cached X post classified as Technical Explanation, Performance, Cost. Themes: typed, probability, confidence, latency, system one. [Source Evidence]
- [Technical Explanation, Project source — relevance 78](/evidence): A cached X post classified as Technical Explanation, Project. Themes: typed, probability, system one. [Source Evidence]
- [Technical Explanation, Project, Performance, Speculation source — relevance 91](/evidence): A cached X post classified as Technical Explanation, Project, Performance, Speculation. Themes: typed, confidence, latency, workflow, system one. [Source Evidence]
- [Technical Explanation, Project, Question source — relevance 76](/evidence): A cached X post classified as Technical Explanation, Project, Question. Themes: typed, probability, router, system one. [Source Evidence]

## Document
- [Architecture Patterns](/research/architecture-patterns): Decision sidecars, cascade routers, confidence gates, quorums, and deterministic action layers. [Authored Hypothesis]
- [Build Ideas](/research/build-ideas): Detailed product hypotheses with MVPs, validation experiments, evidence, and unknowns. [Authored Hypothesis]
- [Build Opportunities](/research/build-opportunities): The highest-leverage experiments ranked for tractability, proof value, and open-source potential. [Authored Hypothesis]
- [Emerging Mental Models](/research/mental-models): Competing interpretations: classifier, router, verifier, policy layer, complement, or replacement. [Authored Hypothesis]
- [Jev Foundations](/research/foundations): Facts, vendor claims, independent observations, opinions, speculation, and unknowns. [Authored Hypothesis]
- [Jev Research Report](/research/report): The complete view: evidence, developer activity, architectures, limitations, and opportunities. [Authored Hypothesis]
- [Research Claims](/research/claims): Important claims with status, evidence, counterarguments, sources, and open questions. [Authored Hypothesis]
- [What People Are Building](/research/projects): Located projects and demonstrations, kept separate from proposals and speculation. [Authored Hypothesis]