Field Note

False Green

Why passing checks can still leave the real system broken.

I spent part of building a game looking at green checks that were all telling the truth.

The branch was correct. A file existed. A local check passed. Another component passed its own test. And the integrated game was still broken.

The problem was not that the checks were fraudulent. The problem was that I had confused local correctness with system closure.

A PASS list can be completely accurate and still mislead you

Software is full of producers and receivers. A script creates something. Another system consumes it. An asset exists on disk, but the runtime may not actually load it. A branch contains the intended change, but production may still be running a different authority. A unit can be healthy while the relationship between units is broken.

That creates a dangerous kind of green: every individual statement may be true while the conclusion people naturally infer from the collection—the system works—is false.

The receiving system is where success has to exist

The correction for me was to stop treating “the producer did its job” as equivalent to “the outcome exists.”

If a build process says it created an asset, I care whether the game can actually use the asset. If a worker says it changed a file, I care whether the production hierarchy consumes that file. If a test proves a component works alone, I still need evidence that the integrated system reaches the intended state.

This sounds obvious when written down. It becomes less obvious when a large checklist is green.

The broader lesson

I now treat a green checklist as evidence about the things it actually tested, not as a visual certificate of completion.

The useful question is: Where must success be observable for this work to count?

Sometimes that is the file. Sometimes it is the API consumer. Sometimes it is the production runtime. Sometimes it is a person using the system. Sometimes it is an external source of truth.

The lesson is bounded. One game-integration incident does not prove a universal engineering law. But it gave me a durable diagnostic habit: when everything is green and reality is still wrong, inspect the path from producer to receiver before inventing another local fix.