Three numbers everyone quoted. None of them survived being checked.
Each case starts with a figure that circulated confidently and meant less than people thought. Open any one for the full write-up and a live model you can push until it breaks.
Everyone knew cost was too high. Nobody could say by how much.
Ringgit of overrun starts a debate. 29 heads of overrun ends one.
Overrun was discussed as a feeling. I built an algorithm that turns a target margin into an allowable cost and the gap into excess headcount, then deployed the dashboard it runs on myself.
Same defect count. Which team has the real problem?
One number, two verdicts. At 7 or more test cases per man-day, 62% reads as a build problem. Below that, a coverage problem.
Exco was making delivery calls on a hunch. Escape rate, read against test coverage, splits that hunch into two different diagnoses, with the bands set by forty historical projects, not opinion.
A fast fix on a bad system still scored well.
Everyone assumed bigger projects carry more defects. They do not: a RM 0.4m project logged 52 while a RM 19m one logged 81.
SLA scoring is a stopwatch: a late fix often counted as on time, and nothing asked whether the system was any good. Four replacement methods were thrown away before defects per million ringgit survived review.