jev-seo
Live SEO audit from one homepage URL: 52 rules, Jev page judgments, PDF, XLSX, and Markdown from the same audit.json.
Overview
jev-seo (github.com/AgriciDaniel/jev-seo, MIT, Python 3.10+, README version badge 0.1.1, GitHub Actions CI on main) audits a live site from one homepage URL. It crawls, checks 52 rules tied to Google Search Central, pulls Core Web Vitals through PageSpeed Insights, and asks TypeSafe Jev typed questions about pages and the site. Code scores and ranks every fix. PDF, XLSX, and Markdown render from one audit.json. Run bin/jevseo or the Claude Code skill /jev-seo. Standard mode needs no SEO data subscription. README prices Jev at 0.042 USD per million input tokens, about 0.00015 USD per page, and about a cent per site. Optional --full adds DataForSEO rankings, keywords, and backlinks at about 0.30 USD per site. Hard caps default to 0.25 USD for Jev and 1.00 USD for DataForSEO, checked before each request. Without TYPESAFE_API_KEY the crawl still runs, Jev sections are marked not assessed, and the score is labelled a partial audit. Page cap defaults to 60. Homepage type is set by code and is never asked. Public GitHub counts were 169 stars and 41 forks when this listing was drafted. An example --full audit of claude-seo.md from 2026-09-22 ships under examples/claude-seo.md.
Problem: SEO audits either dump a template checklist or ask a chat model for a vibe score, and neither keeps the probabilities next to the fix list.
Built for: Marketers and developers who want one homepage URL in, a live crawl plus Jev judgments, and the same findings as a client PDF, an XLSX tracker, and Markdown.
First indexed on Jev Directory: 2026-09-29
Creator and team
- Name
- Agrici.Daniel
- Handle
- @AgriciDaniel
How Jev is used
- Role in the product flow
- Per-page and site-level typed judgments on a live crawl, with code owning rules, scores, and report rendering
- Primitives
- ChoiceScoreNoul
- State in
- One System One request per page carries every page question over shared state (page text capped at 6,000 characters). Site questions use homepage text, navigation, and up to 80 page titles. Model jev-latest, resolved to jev-1.13.0 on 2026-09-21 per references/judgments.md.
- Decision out
- Choice, Score, and Noul answers with probabilities kept in the workbook. Decisive band: Choice confidence at least 0.80; Score when at least 0.80 of probability sits on one side of the midpoint; Noul at P(yes) 0.80 or above, or 0.20 or below. Everything else is marked to verify.
- Clone, pip install -r requirements.txt, and copy .env.example (TYPESAFE_API_KEY)
- bin/jevseo doctor checks dependencies and keys without printing values
- bin/jevseo audit URL writes audit.json and digest.md; optional Playwright for JS-only pages
- Jev answers 13 page questions, 5 site questions, and cannibalization pairs (Jaccard overlap at least 0.4, at most 40 pairs)
- bin/jevseo render writes report.pdf, report.xlsx, and report.md; narrative checks refuse unknown action IDs
jev-seo is the live SEO audit: one homepage URL, a crawl, 52 rules, Jev judgments, and three reports from one JSON file. iannuttall-internal-links crawls up to 500 pages and exports internal link pairs as CSV or JSON. docjev classifies and splits documents from natural-language rules. kylejeong-jev-as-judge scores writing on rubric dimensions. kraayenjon-ai-slop-detector flags low-effort pages. Here, code refuses to ask Jev what the crawl already knows: missing meta skips meta_fit, missing H1 skips h1_fit. Page questions cover type, intent, importance, action, helpfulness, specificity, trust, citability, answer-first, next step, title fit, meta fit, and H1 fit. Site questions cover business model, value prop, entity clarity, topical focus, and local-area service. Competing pages become a finding when Noul P(yes) is 0.6 or higher. Content quality is 70 percent importance-weighted helpfulness, specificity, and trust, plus 30 percent rules. Scores rank the work. README says they do not predict rankings or traffic, and that the tool is not a rank tracker or a Search Console replacement. Cite AgriciDaniel/jev-seo README, references/judgments.md, and references/evaluation.md.
Sourced performance claims
- README cost table: Jev at 0.042 USD per million input tokens, about 0.00015 USD per page; --full DataForSEO about 0.30 USD per site. Defaults: --jev-budget 0.25 USD, --dfs-budget 1.00 USD.Source: github.com/AgriciDaniel/jev-seo README What runs where
- references/judgments.md observed on typesafe.ai, 2026-09-21: 14 requests, 47,181 input tokens, 0.0020 USD. About 3,400 input tokens per page.Source: github.com/AgriciDaniel/jev-seo references/judgments.md
- Evaluation dated 2026-09-22: same 59 pages twice, scores moved 0.03 or less on average, confident page types agreed 44/44. Blind second judge: helpfulness and specificity 28/29, opens with the point 23/23, next step 16/16, keyword relevance 27/30. The second judge is another model, not a human panel.Source: github.com/AgriciDaniel/jev-seo README How far to trust it
- Public GitHub repo AgriciDaniel/jev-seo had 169 stars and 41 forks when this listing was drafted. README version badge is 0.1.1.Source: GitHub API 2026-09-29 and README
Features and stack
Features
- bin/jevseo audit, render, rescore, and doctor commands
- Claude Code skill workflow in SKILL.md
- 52 rules across crawl, on-page, content, links, structured data, AI crawlers, performance, and security
- PDF, XLSX action tracker with dropdowns, and Markdown with charts from one audit.json
- Narrative gate: unknown action IDs refused, numbers missing from the audit flagged
- Offline unittest suite with no network and no spend
Stack
- Python 3.10+
- TypeSafe System One (jev-latest)
- Google PageSpeed Insights
- Optional DataForSEO and Playwright
- WeasyPrint (Pango) for PDF
Pricing: MIT. Crawl, rules, and reports run locally. Jev and optional DataForSEO bill to your keys, with README default budgets of 0.25 USD and 1.00 USD.
Links
FAQ
- Does this replace Google Search Console?
- README says it is an evidence tool for one audit. It does not measure traffic, revenue, or rankings over time, and it is not a rank tracker.
- What does a standard run cost?
- README: about a cent in Jev per site at 0.042 USD per million input tokens, roughly 0.00015 USD per page. --full DataForSEO is about 0.30 USD per site. Caps are checked before each call.
- What if I have no TypeSafe key?
- The audit still crawls and scores rules. Jev sections are marked not assessed and the score is labelled a partial audit.
- How is this different from the internal links tool?
- iannuttall-internal-links classifies pages and exports which URLs should link. jev-seo runs a rule crawl, Jev content judgments, and PDF, XLSX, and Markdown reports from the same audit file.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.