What green proves
I built the same feature twice, structured differently, sharing almost no code. Both suites went green, thousands of tests each. Both failed adversarial review, four independent lenses apiece, on the same three classes of defect. Two implementations failing in the same ways is not sloppy code. It is evidence about the spec.
The diagram traces it: two green builds, both wrong, the shared failure classes pointing past the code to the design. The move wasn’t a third rework. It was to split the spec, ship the achievable half, and record the rest as unachievable-as-specified with the evidence attached, so a future me doesn’t cheerfully start attempt three.
The whole episode, including what a convenient fake in a test agrees in advance never to see, is in Two green builds, both wrong.