Why Vernier exists

Every team ships on top of a pile of assumptions. The checkout charges each card once. A cancelled subscription stays cancelled. One tenant can never read another tenant's records. Almost none of this is written down, and almost none of it is checked. It lives in the heads of the people who wrote it, and it leaves when they do.

The tools we trust do not check it either. Tests verify the cases someone thought to write. Monitors report that requests return quickly and without exceptions. Code review looks at a diff for an hour and moves on. All three are worth having. None of them answers the question that actually matters: is the product still doing what we believe it does?

So the industry runs on a proxy. Green means good. Zero exceptions means correct. Except a request can return 200 and still charge the customer twice. A build can pass while a renewal quietly survives its own cancellation. Zero exceptions does not mean correct behavior.

The unit of correctness is the behavior

Not the test, not the endpoint, not the pull request. A behavior is a claim the software makes about itself. "A refund never exceeds the captured amount." Claims like this can be collected, named and checked. Written down, they are a contract. Left unwritten, they are an outage with a delay on it.

Vernier exists to hold that contract. It reads a codebase, works out what the product actually guarantees, and states each guarantee in plain language for people to confirm, reject or reword. Then it earns verdicts. A behavior is Proven only when a run produced evidence you can open and read. It is Broken only when a failure actually reproduced. When Vernier cannot tell, it says Inconclusive and shows you what is missing. It does not guess, and it does not round up.

Honesty is the product

A correctness tool that exaggerates is worse than no tool at all, because it produces exactly the false confidence it was built to remove. So the rule runs through everything: no claim without evidence, no percentage we have not measured, no state that flatters. If that costs us an impressive-looking number, it was not our number to show.

Software that runs is not software that works. We are building the instrument that measures the difference.