Three workflows were built in KF and then deliberately attacked — not demonstrated, tested, with acceptance criteria written down before the work started and a willingness to record what failed.
Each one produced a database you can open and read.
An 8D problem-solving cycle as a reusable type, a candidate-cause library, and one question: when a team learns something on Monday, does it reach the projects that started last year?
Result: yes, and retroactively. A preventive action promoted to a type appeared as an open checklist row on an instance that had been typed before the lesson existed — with nothing written to the instance.
A hazard analysis and risk assessment for an electric power steering system: item definition, malfunctions, operational situations, hazardous events classified S/E/C, ASIL determined, safety goals derived.
Result: 10 of 10 reference vectors reproduced, with no code changes — configuration only. And a HARA and a conventional component FMEA living in the same database, each with its own matrix.
A 65-requirement library and a complete quality management system for a fictional company, then an audit conducted against it — with the auditor working from queries rather than from a guided tour.
Result: 65 requirements verdicted, 24 conform, two major nonconformities. The most interesting finding was not one of the seeded gaps.
Each write-up states what did not work as well as what did. That is deliberate: a case study with no limits in it is marketing, and you would be right to discount it.