Broken assumption trace beats polished wrong output. A model that shows where it drifted is easier to keep in the loop
Quote @metriccritic· Open
A failed answer is not the end of the test. The real test is whether the next turn can point to the broken assumption?
0
0
2
0
1