Gravito Eval

Scan any website. See what's wrong, what's missing, and what a human reviewer would miss.

npx gravito-eval scan https://stripe.com

What You Get

  Scanning https://stripe.com

  ◐ Fetching page content (2s)
  ◑ Running content analysis (4s)
  ◉ Analysis complete (32s)

  Gravito Eval Results
  ──────────────────────────────────────────────────

  Score:  57/100  🔴 F Grade
  Site:   https://stripe.com
  Title:  Stripe | Financial Infrastructure to Grow Your Revenue

  █████████████████░░░░░░░░░░░░░  57%

  vs Fintech & Financial Services: top 52% (avg: 58)

  Key Issues
  ──────────────────────────────────────────────────

  HIGH  Unsubstantiated Claim
  → "50% of" — no source, no verification
  Fix: Add a verifiable source or rephrase as a qualified statement

  HIGH  Unsubstantiated Claim
  → "100% of" — percentage claim without citation
  Fix: Add a verifiable source or rephrase as a qualified statement

  MEDIUM  Banned Vocabulary
  → "cutting edge" — vague marketing language
  Fix: Replace with specific, measurable language

  ──────────────────────────────────────────────────
  Share: https://gravito.ai/try/report/a1b2c3d4

  Analyzed in 32.0s · Gravito Engine (Live) · 4 issues found

Every scan produces a shareable link.

Compare Two Sites

npx gravito-eval compare stripe.com github.com

  Gravito Eval — Comparison
  ──────────────────────────────────────────────────

  ✓ stripe.com: 57/100 🔴
  ✓ github.com: 89/100 🟡

  Head-to-Head
  ══════════════════════════════════════════════════

  stripe.com
  █████████████████░░░░░░░░░░░░░  57/100 🔴 F

  github.com
  ██████████████████████████░░░░  89/100 🟡 B

  Winner: github.com (+32 points over stripe.com)

  Key Differences
  ──────────────────────────────────────────────────

  Only stripe.com has:
    HIGH Unsubstantiated Claim
    MEDIUM Banned Vocabulary

  Both sites have:
    ~ Missing Disclaimers

Compare your site to a competitor. See who's doing better and why.

Try It

npx gravito-eval scan https://stripe.com
npx gravito-eval scan https://openai.com
npx gravito-eval scan https://your-site.com

No install. No API keys. No setup.

Demo Mode

See a walkthrough with explanations:

npx gravito-eval demo

What This Does

Point it at any URL. It scans the page and tells you:

• What a content reviewer would flag
• What most reviewers would miss
• How your site stacks up against others in your industry

Same engine that powers Gravito.

Why This Is Different

Most tools tell you what is wrong.

Gravito Eval also tells you:

• How your site compares to others in your industry
• What a human reviewer would catch — and what they'd miss
• Whether the AI's extra findings are useful — sometimes it catches things humans don't

Use Cases

| Use Case | What You Get | |---|---| | Scan your landing page | Quality score, issues, fixes, industry benchmark | | Compare to competitors | Side-by-side analysis with compare command | | Evaluate LLM outputs | Alignment metrics, novel signal detection | | Compare prompts | Run the same content through different prompts, compare scores | | QA AI features | Check if your AI-generated content meets quality standards |

Local Evaluation Mode

Already have findings from your own AI system? Compare them against a human baseline:

npx gravito-eval run ./my-findings.json

Input format:

{
  "aiFindings": [
    {
      "id": "ai-1",
      "description": "Hero section uses unsubstantiated claim",
      "category": "trust",
      "severity": "high"
    }
  ],
  "humanFindings": [
    {
      "id": "human-1",
      "description": "Misleading statistic in hero",
      "category": "trust",
      "severity": "high"
    }
  ]
}

Output:

  Recall:     75%
  Precision:  50%
  F1:         60%
  Top-3 Agreement: 100%
  Novel Signal:    67% (validated)

  ✓ Strong alignment with human judgment
  ◆ AI found significant issues humans missed
  Verdict: PASS

Add --explain for detailed match reasoning. Add --json for raw output.

How It Works

Your URL → Fetch page → Analyze content → Compare to industry → Score + Issues + Fixes

Gravito checks your site for:

• Unverified claims ("trusted by 10,000+ teams" with no source)
• Vague marketing language ("cutting edge", "best in class")
• Missing disclosures (AI usage, cookies, data handling)
• Inconsistent messaging across sections
• Issues that human reviewers typically miss

The analysis runs server-side. No local LLM required.

CLI Reference

gravito-eval scan <url>                     Scan a live URL
gravito-eval compare <url1> <url2>          Compare two sites side-by-side
gravito-eval scan <url> --json              Output raw JSON
gravito-eval demo                           See a demo with explanations
gravito-eval run <path>                     Evaluate local findings
gravito-eval run <path> --explain           Show detailed match reasoning
gravito-eval run <path> --json              Output raw JSON
gravito-eval --help                         Show help
gravito-eval --version                      Show version

CI/CD Integration

Add content quality checks to your pipeline:

# .github/workflows/gravito-eval.yml
name: Content Quality Check
on: [pull_request]
jobs:
  check:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
      - name: Scan site
        run: npx gravito-eval scan https://your-site.com --json > report.json
      - name: Check threshold
        run: |
          SCORE=$(node -e "const r = require('./report.json'); console.log(r.overallScore || 0)")
          if [ "$SCORE" -lt 50 ]; then exit 1; fi

Fail PRs that drop your content quality below a threshold.

Requirements

• Node.js 18+
• Internet connection (for scan and compare commands)

Telemetry

Anonymous usage data (timestamp, version, command name) is collected to improve the tool. No findings, file paths, or PII.

Disable: GRAVITO_TELEMETRY=0 gravito-eval run ./data or set DO_NOT_TRACK=1.

Programmatic API

import { evaluate } from "gravito-eval";

const result = evaluate(aiFindings, humanFindings);

result.detection.recall     // How much of what humans find does the AI catch?
result.detection.precision  // How much of what the AI finds is actually relevant?
result.detection.f1         // Harmonic mean
result.novelSignal          // What did the AI find that humans missed?
result.verdict              // PASS | PARTIAL | FAIL | INSUFFICIENT_DATA

Continuous Monitoring

Want this running continuously on your site?

Gravito monitors your site automatically and alerts you when new issues appear.

MIT License