sheg
Codex plugin that runs polls on simulated reader profiles with a System One model, recording typed answers for follow-up studies.
Overview
Sheg helps you explore how different simulated readers respond to text. It runs bounded studies with frozen respondent profiles, typed questions and recorded answers you can inspect and use to shape a follow-up. The README's example: choosing a pull quote by asking varied reader profiles which paragraph best represents an article's central idea, then running a follow-up that shows each chosen paragraph alone and asks what it communicates.
Sheg is packaged as a Codex plugin (Node 24, no local build), with skills for study design and for running polls, queries and follow-ups. It helps shape the study and checks its fit before asking you to approve a run, and hosted inference requires your authorisation and a call limit. Providers include Jev, Cloudflare Clef for image-backed studies, and a separately operated Laya service. The README is explicit that simulated responses do not establish human readership, real-world accuracy, statistical significance or causal lift.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which paragraph best represents the article's central idea (for this reader profile)? | — |
| Choice | What does this paragraph communicate? | — |
How it works
- Design a study: respondent profiles, stimulus, questions and options
- Approve the run with a call limit
- Ask the decision model each question per profile and record answers
- Inspect results and run follow-ups on selected contexts
Strengths
- Structured, recorded studies rather than ad hoc prompting
- Call limits and approval before hosted inference
- Multiple decision-model providers
Caveats
- Simulated readers are not real readers
- Codex-only plugin
Getting started
codex plugin marketplace add HarleyBartles/sheg, restart Codex, then enable Sheg in the Plugins Directory.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Codex plugin (Node 24)
- Has code
- yes
- License
- MIT
- Stars
- 0
Use cases and integrations
How curated lists describe it
Use a system one model to run polls on simulated users
wh000wh000/awesome-jev-live
Frequently asked questions
What does the decision model decide?
Each simulated respondent's typed answer to a poll question about a text.
Do the results reflect real readers?
No; the README says they do not establish human readership or significance.
Which providers can I use?
Jev, Cloudflare Clef or a Laya service.
Related projects
Local-first Windows writing studio with knowledge graphs and author voice, where optional Jev evaluates candidates and editorial decisions.
ProjectbetaTypeScriptAnalytics engine that rates published posts on pre-registered rubrics with Jev and tests ratings against real engagement.
ProjectbetaTypeScriptChrome extension that grades readable page sections on ten rubric scores with Jev via Vercel AI Gateway and gives the page an A-E grade.
ProjectbetaTypeScriptExperimental editor that scores X drafts live on six viral dimensions with one Jev call, with a heuristic fallback.
ProjectprototypeTypeScriptCommunity tool that fetches a SaaS homepage as Markdown and asks Jev ten questions about first-visit clarity, with a leaderboard.
ServicebetaTypeScriptAudience simulator where 10,000 computed personas read your post or listing and Jev decides who sees it and how each reacts.
DemobetaTypeScript