Can This Output Be Safely Written?

A deterministic infrastructure guard that validates payloads before database writes.

✓ ALLOW: Safe to write⚠ ALLOW_WITH_WARNING: Concerns exist✕ BLOCK: Do not write

Built for LLM outputs • AI agents • Batch pipelines • No configuration needed

Load an example:

Request Body

What Does WriteVerdict Do?

WriteVerdict prevents irreversible data corruption by blocking unsafe writes before they reach your database.

Schema Validation

Validates types, formats, required fields, and constraints against your schema.

Logical Consistency

Detects contradictions like end < start, max < min, negative quantities.

Duplicate Detection

Warns about potential duplicates based on recent writes (opt-in).

Simple Integration

One API call before each critical database write. That's it.

// Before writing to your database
const response = await fetch('https://api.writeverdict.com/verdict', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    target: { identifier: 'users', operation: 'insert' },
    payload: generatedData,  // From LLM, agent, or batch
    schema: yourSchema
  })
});

const { verdict } = await response.json();

if (verdict === 'BLOCK') {
  // Handle rejection
} else {
  // Safe to write
  await db.insert(generatedData);
}

Built For

LLM Outputs

Validate JSON generated by GPT, Claude, or other models before persisting. Catch hallucinated fields and schema violations.

AI Agents

Guard autonomous agents with clear verdicts they can parse without custom validation logic.

Batch Pipelines

Validate bulk imports and ETL transformations before commit. Detect issues early instead of debugging failed transactions.

API Gateways

Add a validation layer between external inputs and your database. Block invalid requests before they trigger errors.

Design Principles

✓Deterministic — Same input always produces same verdict
✓Stateless — No learning, no stored state, no drift over time
✓No business logic — Universal patterns only (schema violations, logical contradictions)
✓No modification — Evaluates payloads, never transforms them
✓Explainable — Clear, finite reasons for every verdict (no hidden logic)

What Does the API Response Look Like?

Clean JSON that's easy to parse for humans and machines.

Each response includes a verdict, a list of standardized reasons, a human-readable summary, and technical metadata.

{
  "verdict": "BLOCK",
  "reasons": [
    {
      "code": "schema_violation",
      "level": "BLOCK",
      "description": "Invalid email format",
      "fields": ["email"]
    }
  ],
  "summary": "Payload violates schema constraints.",
  "metadata": {
    "version": "1.0.0",
    "timestamp": "2025-01-15T10:30:00Z",
    "request_id": "req_abc123"
  }
}

Common Questions

Does the API modify my payload?

No. WriteVerdict only evaluates. It never transforms, sanitizes, or modifies your data. This keeps the validation layer simple and predictable: you get a verdict, you decide what to do.

Do I need to configure validation rules for my application?

No. You provide the schema in your API request. WriteVerdict checks against that schema using universal patterns (type validation, format checking, logical consistency). Works out of the box for any data model.

Can I use this with n8n workflows or AI agents?

Yes. The API returns structured JSON with predictable fields (verdict, reasons, summary). Your n8n workflow or AI agent can parse the response and take action without custom logic. Works with MCP agents, LangChain, Zapier, Make, and any tool that can call a REST API.