JSON SCHEMA → REGRESSION EXAMPLES

See a breaking input.
Keep a test you can run.

Find concrete values your old schema accepts and your new schema rejects. Download the examples, schemas, and a reproducible check in one small package.

A bounded search over a documented subset. No counterexample means inconclusive, never “compatible.”

No account or payment. Exploring a future $29 package per schema pair.

01 / TRY A SCHEMA PAIR

What changes for your inputs?

Check the supported subset

Loading an example…

Current rules

0 bytes

Proposed rules

0 bytes

64 KiB request limit · use non-sensitive schemas
This free demo does not retain schemas or send them to an LLM.

The demo runs a real check. The saved example does not count as your own input.

SMALL ENOUGH TO INSPECT

The evidence comes with the check.

Exact inputs & findings

Your two schemas, concrete counterexamples and the validation errors behind each result.

Repeatable Python check

A verifier and regression test using Python 3.11+ and its standard library. No model or network needed.

Inspectable source & hashes

The validator source and a file manifest. SHA-256 checks consistency; it is not a publisher signature.

How to check a downloaded package

Unzip the package, inspect its README and Python source, then run these commands from that directory:

python3 -B verify_regressions.py
python3 -B -m unittest -v test_regressions

Saved example schemas · Saved example report · Saved example ZIP

Use the demo from an MCP client

Add this Model Context Protocol endpoint in a compatible client's server settings. It connects the client to the bounded schema-regression demo.

/mcp

The endpoint is a tool connection. Your client controls its own model usage.

AN OPTIONAL PRICE CHECK

Would the package be useful at $29?

We are exploring a future price of $29 per schema pair. This pilot and its downloads are free. There is no checkout.

Try your own schemas, find a counterexample and click download before leaving an informed signal.

Optional and nonbinding. No charge, order, email or wallet.

Know the boundaries.

Supported rules & what the result means

The pilot supports one explicit type; enum; object properties, required and boolean additionalProperties; array items; minimum/maximum; minLength/maxLength; and minItems/maxItems.

References ($ref), composition such as anyOf, patterns, formats and full OpenAPI are unsupported. Unsupported schemas are rejected. Search and value-size limits are included in each report.

A known miss: raising string minLength from 257 to 258 excludes strings of length 257. This generator only builds strings up to 256 characters, so it returns inconclusive for that change.

A counterexample is evidence that an old-accepted input is new-rejected within this subset. Absence of a counterexample does not prove compatibility. This does not check the reverse direction or every possible input.

Free alternatives & the price hypothesis

Free tools such as Atlassian's json-schema-diff, validators and hand-written tests may already fit your workflow. This pilot makes no claim of superior coverage. The question is whether a small, reproducible package is useful enough to pay for later.

Clicking the optional price-interest button records a nonbinding signal only. It is not an order, a payment, proven demand or revenue.

What is processed, and what is measured?

The free-demo endpoints process uploaded schemas and generated results in memory for the request. They do not persist their content or send it to an LLM. Use non-sensitive schemas. The resulting package remains in this browser tab until you leave or replace it, unless you download it.

A random, short-lived session token stays in this tab's memory. This UI uses no cookies or localStorage, loads no third-party scripts, and sends no events to third-party analytics services. Coarse pilot events are kept for at most 21 days; the session expires within an hour.

The free demo can count a request, result, download-click acknowledgment and your explicit price-interest signal. A download click does not establish that a file was saved. Session counts are not verified people, buyers or revenue. Hosting providers may keep their own access logs.

You can inspect this page's JavaScript source and built-in example.