Two weeks down, and every sensor said fine.

A precision laser drilling machine started placing features slightly off, but only in one region of the working field, and only now and then. Most parts measured clean, so the escapes reached the next process step before anyone saw a pattern, and finished assemblies were scrapped downstream.

The line pulled the machine straight away. This is the kind of issue with almost no clue attached: intermittent, confined to one region, and invisible to the machine itself.

Every subsystem was checked, and every subsystem passed. Optics in spec. Motion in spec. Power, environment and alignment all normal. No error, no alarm, no trend that moved when the issue did. After two weeks the machine was, by every measurement it could make of itself, perfect. It still couldn't be released.

Cala got the symptom the way you'd tell a colleague: where it happens, how often, and the two weeks of clean checks. Its top hypothesis pointed somewhere nobody had looked. Not at a component, but at a software compensation term that adjusts placement across the working field. With the wrong magnitude in that term, the machine would behave exactly like this: healthy everywhere, wrong in one region, now and then, with every sensor happy, because it was doing exactly what it had been configured to do.

Read the configured value and compare it with the commissioning record. Minutes of work. No teardown, no parts.

The term was about ten times its intended value. Restored, the issue disappeared and the machine requalified. The first hypothesis on the list was the one, reached from first principles on a machine where nothing measured abnormal, after two weeks of everything else.