AI agents

Repairing and validating broken JSON

• A Repair Prompt Template That Names the Failing Field
• A Validator Wrapper That Names the First Failing Field
• Naming the First Failing Field: A Validator Wrapper
• Repairing Invalid JSON from LLM Responses
• Routing a Failed Repair to a Human Queue: Codes, Ownership
• What to Do When a Model Ignores the Schema: Diagnostic Playbook
• When a Model Ignores the Schema: Diagnostic Playbook
• When to Repair a Failed Response and When to Start Over
• A Contract Test for Your Repair Logic: Mocking Model Failure
• A Cost Model for Schema Retries: Tokens, Latency, Repair Prompts
• A Decision Tree for Schema Misses: Fix, Retry or Reject
• A Log Schema for Repair Attempts and Their Cost
• A Repair Attempt Log That Proves What the Model Saw
• A Repair Prompt That Names the Path and Skips the Rewrite
• A Retry Prompt Template That Quotes the Failing Path Back
• Building a Fallback Chain for Structured Outputs
• Estimating Repair Costs Before You Build the Pipeline
• Fallback Strategies for Structured Output Failures
• Feeding Validation Errors Back: A Retry Prompt
• Making the Validator Wrapper Say What to Do Next
• Reading the Stop Reason Before the Schema Check
• Retry Prompts That Quote the Failing Path Back to the Model
• Routing a Failed Repair to a Human Queue
• Trimming a Quoted Fragment Without Losing the Failing Path
• Truncation, Stop Reasons and Max Tokens: A Handling Checklist
• Validating LLM Output in Production: A Small Library
• When a Field Fails Every Attempt: Schema Fix vs Prompt Fix
• When a Repair Fails: Deciding Between a Retry and an Escalation
• Widen, Repair or Enforce: Routing a Schema Miss