Measuring Structured JSON Reliability Without an LLM Judge

Evaluate extraction and classification outputs with deterministic checks. The useful boundary is a repeatable workflow with named inputs, explicit checks, and a release decision that another contributor can inspect. The examples below are illustrative procedure, not a claim about a particular customer or production result.

Set the operating boundary

Define syntax validity, schema validity, allowed values, completeness, and task correctness. Treat that requirement as a recorded engineering decision rather than an informal cleanup step. Start with a small, versioned input set and keep the original source identifier beside every transformed record. That lets a reviewer trace an unexpected output back to the data, configuration, or evaluation rule that produced it.

The implementation should separate deterministic checks from judgment calls. Deterministic checks can reject missing fields, malformed files, invalid ranges, and inconsistent identifiers. Judgment calls need a documented review rule, a sample of borderline cases, and a reason for keeping or excluding them. Mixing the two creates silent changes that are hard to reproduce later.

Parse outputs without silently repairing malformed JSON

Parse outputs without silently repairing malformed JSON. Treat that requirement as a recorded engineering decision rather than an informal cleanup step. Start with a small, versioned input set and keep the original source identifier beside every transformed record. That lets a reviewer trace an unexpected output back to the data, configuration, or evaluation rule that produced it.

Record the result in a small manifest: source version, transformation version, input digest, output digest, reviewer, and timestamp. A compact manifest is more useful than a long narrative when a later run needs to be compared with this one.

Validate required fields and enum values using standard Python

Validate required fields and enum values using standard Python. Treat that requirement as a recorded engineering decision rather than an informal cleanup step. Start with a small, versioned input set and keep the original source identifier beside every transformed record. That lets a reviewer trace an unexpected output back to the data, configuration, or evaluation rule that produced it.

Report failure categories rather than one blended score

Report failure categories rather than one blended score. Treat that requirement as a recorded engineering decision rather than an informal cleanup step. Start with a small, versioned input set and keep the original source identifier beside every transformed record. That lets a reviewer trace an unexpected output back to the data, configuration, or evaluation rule that produced it.

Add adversarial inputs containing quotes, newlines, and missing fields

Add adversarial inputs containing quotes, newlines, and missing fields. Treat that requirement as a recorded engineering decision rather than an informal cleanup step. Start with a small, versioned input set and keep the original source identifier beside every transformed record. That lets a reviewer trace an unexpected output back to the data, configuration, or evaluation rule that produced it.

When human review remains necessary

Explain when human review remains necessary. Treat that requirement as a recorded engineering decision rather than an informal cleanup step. Start with a small, versioned input set and keep the original source identifier beside every transformed record. That lets a reviewer trace an unexpected output back to the data, configuration, or evaluation rule that produced it.

Practical check

Complete Python evaluator with sample input and output. The following minimal check is intentionally small enough to run before a longer workflow. Expand it only after its output identifies the missing condition or failure category.

import json
import sys
from pathlib import Path

path = Path(sys.argv[1])
valid = 0
with path.open(encoding="utf-8") as handle:
    for line_number, line in enumerate(handle, 1):
        if not line.strip():
            raise SystemExit(f"line {line_number}: blank record")
        json.loads(line)
        valid += 1
print(f"validated {valid} records")

Review and release criteria

A result is ready for the next stage only when the required fields, split boundaries, and evaluation evidence agree. Keep a rejected sample set as well as accepted output; otherwise the same edge case will return as an unexplained regression. Adding an LLM judge as a hidden fallback. The correct response to uncertainty is to label the limitation and add a focused validation case, not to hide it behind a blended score.

Before publishing, verify the data digest, configuration, expected output shape, and the most important rare-case slice. The final registry note should state what changed, what was checked, and what remains outside the adapter's intended scope. That makes the next iteration a comparison instead of a guess.