Correct the k=20 description of the probe-then-commit bound
#2
by EvalData - opened
The k=20 rows were described as a universal failure. They were never checked: observed regret was only looked up for k in {10,50,100}, so the comparison returned NaN and NaN was counted alongside real violations. Using the observed regret already recorded in the early-predictor sweep, the bound holds 6/6 at k=20 with mean observed regret 0.00331, matching 0.00331/0.00332/0.00332 at k=10/50/100. The genuine failures are Burgers1D and Burgers1D-lownu at every evaluated window. Single-line change; no other rows touched.