DexterDexter

DexterAI Verification System

How Dexter verifies and scores marketplace resources using AI-powered testing.

AI Verification System

Dexter automatically verifies every resource in the marketplace using AI-powered testing. This page explains how verification works, what scores mean, and how to improve your resources.

How It Works

Every resource in the marketplace goes through automated verification:

  1. We make a real paid request to your endpoint
  2. AI generates a realistic test input based on what your endpoint claims to do
  3. We evaluate the response for quality, accuracy, and value
  4. Your resource gets a score from 0-100

Smart Test Inputs

We don't just send {"message": "test"}. Our AI generates contextual, realistic inputs based on your endpoint's description and schema.

Example for a Tax Advisor endpoint:

{
  "message": "I run a 3-person LLC doing $420k revenue in 2023 with ~$260k expenses. I'm considering electing S-Corp for 2024. Can you estimate the potential SE tax savings vs added payroll/admin costs if I set my reasonable salary at $90k?"
}

This means your endpoint is tested the way a real customer would use it.

Understanding Your Score

ScoreStatusWhat It MeansRefund Protection
70-100PassYour endpoint works well and provides good value✅ Full
50-69PassWorks but has minor issues⚠️ Partial
25-49FailResponse quality needs improvement❌ None
0-24FailSignificant issues with the endpoint❌ None

Refund Protection: Endpoints scoring 70+ automatically get refund headers that signal buyer protection. This is a trust badge that encourages more purchases.

What We Evaluate

  • Did you answer the question? Not just "related topic" - actual answer to what was asked
  • Is the response useful? Would a paying customer be satisfied?
  • Any errors or issues? Empty responses, generic templates, or errors
  • Value for price? Is the response worth what you're charging?

Verification Statuses

StatusMeaning
PassTested and working well
FailTested but response quality is poor
InconclusiveCouldn't complete test (see reasons below)
PendingNot yet verified

Why "Inconclusive"?

Your resource might show as inconclusive if:

  • Price > $0.25 - We can't afford to test expensive endpoints automatically
  • Payment timeout - The payment process timed out
  • Network error - Temporary connectivity issue

Important: Inconclusive resources are not penalized in your average score. We only average scores from resources we actually tested.

Common Issues and Fixes

"No 402 Response" (Score: 0)

Problem: Your endpoint doesn't return 402 Payment Required.

What this means:

  • Your endpoint might be returning 200 (free) or 404 (not found)
  • The x402 payment handling isn't configured correctly
  • The URL in our system might be wrong

How to fix:

  1. Verify your endpoint returns 402 when no payment header is present
  2. Check your x402 middleware is properly configured
  3. Contact us if the URL needs updating

"Asked for Clarification" (Score: 25-45)

Problem: We sent a specific question and you asked for more details instead of answering.

What this means:

  • Your endpoint might be designed for interactive use, not single-request
  • The input schema might be missing required context

How to fix:

  • Provide useful default behavior even with minimal input
  • Include example responses or frameworks when context is limited
  • Don't require back-and-forth for simple queries

"Generic Response" (Score: 40-60)

Problem: Response is related to the topic but not specific to what was asked.

How to fix:

  • Parse and use the specific details from the input
  • Tailor responses to the actual question, not just the category
  • Avoid templated responses that ignore the input

"Error Response" (Score: 0-10)

Problem: Your endpoint returned an error after payment.

How to fix:

  • Check your server logs for the error
  • Ensure your endpoint handles the input format we send
  • Test with various input types

Hiding Resources

If you have resources you don't want displayed publicly (test endpoints, deprecated APIs, etc.), you can hide them:

  1. Go to your Seller Dashboard
  2. Find the resource you want to hide
  3. Toggle "Public" to off

Hidden resources:

  • Won't appear on your public seller profile
  • Won't affect your public average score
  • Are still visible in your private dashboard

Your Seller Score

Your overall seller score is calculated from your public, tested resources only.

What's Included

✅ Resources that passed verification ✅ Resources that failed verification (they count, but show you need to improve)

What's NOT Included

❌ Hidden resources (isPublic: false) ❌ Resources we couldn't afford to test (> $0.25) ❌ Resources with no 402 response ❌ Resources still pending verification

This means:

  • Expensive premium tools don't drag down your score
  • Broken endpoints are flagged but don't tank your average
  • Only fair tests count - if we tested it, it counts

Verification History

Every verification attempt is logged and visible:

  • On the resource page: Users can see verification history
  • Test input shown: What we actually sent to test your endpoint
  • AI reasoning: Why we generated that specific test input
  • Score explanation: What the AI noted about the response

This transparency helps you understand exactly why your resource scored the way it did.

Re-Verification

Resources are automatically re-verified periodically. You (or any user) can also trigger a manual re-verification:

  1. Go to the resource detail page
  2. Click "Verify Now"
  3. Wait for results (usually under 60 seconds)

Cooldown: 1 hour between manual verifications per resource.

Tips for High Scores

  1. Be specific - Answer the actual question, not just the general topic
  2. Handle edge cases - Provide useful responses even with minimal input
  3. Return structured data - JSON responses are easier to evaluate
  4. Match your description - Do what you claim to do
  5. Test yourself - Try your endpoint with realistic inputs before listing

FAQ

Why did my score drop?

Scores reflect the current state of your endpoint. If your endpoint's behavior changed (server issues, API changes, etc.), your score will reflect that. Fix the issue and trigger a re-verification.

Why is my expensive tool showing "inconclusive"?

We cap automatic testing at $0.25 per request. Premium tools (> $0.25) show as inconclusive with a note that they weren't tested due to price. This doesn't affect your average score.

Can I dispute a score?

Yes! If you believe a score is unfair:

  1. Check the verification history to see what we tested
  2. If there's an issue with our test input, contact us
  3. Fix any issues on your end and trigger re-verification

How often are resources verified?

  • New resources: Within 1 hour of discovery
  • Active resources: Re-verified every 24-48 hours
  • Manual trigger: Anytime (1 hour cooldown)

Need Help?