Across 2,170 GitHub projects, Jev drew 1,865 new repos in one week, yet 63% of attention went to routing and interface agents

The study profiles the scale and growth of Jev's public ecosystem: 2,170 verified public Jev projects as of September 22, 2026, of which 1,865 were created in the week after Jev's September 15, 2026 release and 305 were integrations into existing repositories, with those new repositories gaining 43,750 stars in the same week. Prior Jev literature largely reports single-task or single-system applications and evaluations; this work is presented as the first ecosystem-wide census quantifying growth and adoption shape. Built from GitHub repository and code search, deduplicated by repository ID, with each candidate inspected by a GPT-6 Luna Max agent and included only when public code or documentation showed Jev used for a concrete task; key inclusion decisions and domain labels were independently reviewed by a second agent, with disagreements resolved by inspecting original repository materials.

In usage patterns, Jev acts as a reusable decision component: attribute judgment appears in 77% of projects, scoring or ranking in 52%, and action selection in 31%; by interface, Choice appears in 81.0%, Noul in 72.2%, and Score in 45.4% of projects, with 36.8% using all three and 69.7% of projects with an identified purpose using two or more purposes. Moves 'how Jev is used' from anecdotal description to cross-domain distributional statistics, showing interfaces are often combined rather than used alone. Counting rules are explicit: each project is counted once per purpose and per interface, so shares may overlap; results derive from annotation of the 2,170 projects.

Jev's role varies by domain: action selection accounts for 52% of purpose labels in Simulation & Control and 47% in Interface Agents; model and tool selection is 27% in Routing & Automation but only 6% in Software Engineering; content filtering is 21% in Search & Memory and outcome judgment 24% in Safety & Governance; yet attribute judgment remains the leading purpose in six of the eight categories. Shows the same decision model takes on different functions in different workflows rather than serving one fixed judgment type. Based on purpose-composition statistics split by application domain, from which the authors conclude Jev is a reusable judgment component whose role adapts to application inputs and downstream actions.

Project supply and public attention diverge: Routing & Automation has 250 projects averaging 364 stars each and Interface Agents 175 averaging 271, together only 19.6% of projects but 63.0% of all stars, while the largest categories Content & Expert Tasks (387) and Search & Memory (381) average just 31 and 37 stars; Routing & Automation and Simulation & Control have nearly identical project counts (250 vs 252) but average 364 vs 8 stars. Contrasts attention and project volume as two independent dimensions, showing highly uneven visibility. Uses project count as supply and mean stars per project as a demand proxy, each normalized independently; the authors state that stars measure interest in entire repositories and may concentrate in a few prominent projects, so results indicate uneven visibility rather than unmet demand.

This work is aimed at people studying the design and evaluation of general-purpose decision models: it offers a dataset and taxonomy for selecting evaluation settings and for studying how decision models are integrated into larger systems. The conclusions apply to the public GitHub ecosystem as of September 22, 2026, and the authors scope the analysis to public repositories, excluding private repositories and commercial applications. For developers, the seven representative applications in Appendix B (browser action selection, code-review triage, tool-history compaction, tool-call risk gating, per-turn model selection, game control, and tax-form page classification) illustrate integration patterns that turn typed decisions into program actions, with diagrams summarizing source-code paths rather than recorded executions.

The authors list as limitations that the analysis is a snapshot as of September 22, 2026, so project counts, usage patterns, and public attention may shift as the ecosystem evolves, and that the dataset includes only public GitHub repositories, so private repositories and commercial applications are outside scope and the account of Jev adoption is not complete. In addition, stars measure interest in entire repositories and may concentrate in a few prominent projects, so attention findings should be read as visibility differences. This was a full-text reading, but figures are presented in text and table form, so specific graphical details are not expanded in the text.