
3 runs of t/1, compared under tool_name
  agreed for 1 step(s), diverged at step 1

step 1
   count  rate  95% CI     behaviour  then passed          
A  2/3    67%   [21%-94%]  left       2/2  100%  [34%-100%]
B  1/3    33%   [6%-79%]   right      0/1  0%  [0%-79%]    

Intervals are Wilson 95%. With this many runs they are wide; separating 60% from 80% takes about 82 runs per arm.
