Are the points being made here muttually exclusive?
Can't we say that a big swing observed in a tool like waltheri is a really interesting observation that suggests that b is a move that we should take a closer look at, but also that we need to do more work because the method has validity issues that stop it being conclusive?
As ever in go, getting closer to the answer seems to require a lot of work.

Personally I don't really like this methodology. I know it's the clasical way of formulating joseki, but it doesn't seem quite relevant/what the bots were built to do. The bots really really like the first few moves in each corner, which means that i think joseki questions posed to bots in this way will get answers that are dominated by trying to take sente at all costs. As a result, I think that studying joseki with bots on an empty board would give you different patterns to real games.
It's easy to do though:
Elf v1 wants to tenuki instead of a or b.
a is a -13.4% mistake
b is a -3.1% mistake
(both on just over 35k playouts)
The reason for this seems to be that in this specific unusual case, a is not sente (white will take an empty corner), but b is. In many real game whole board situations a seems pretty sente though

I sampled 50 recent games featuring a and 44 recent games including b (supposed to be 50 of each, my error). Still small samples, but ritcher data from each if you trust elf. Grp allows me to skip straight to the relevant moves of the relevant games where these paterns happen, estimate the win% with elf before the move, estimate the win% after and output all this into a nice spreadsheet. I also get to look through the alternative moves.

Sample size: 50
(ie. 50 different whole board positions in which a human pro chose to play 'a')
average swing: -1.63%
elf agreed with the pro move: 32/50
elf thought that 'b' was a better move than 'a': 11/50
Win % before Win % after swing
56.66% 53.10% -3.56%
65.79% 50.26% -15.53%
9.82% 8.50% -1.32%
56.28% 55.92% -0.36%
48.00% 46.23% -1.77%
49.11% 46.91% -2.20%
54.92% 54.40% -0.52%
44.15% 44.67% 0.52%
42.38% 42.14% -0.24%
42.63% 40.41% -2.22%
53.89% 53.16% -0.73%
52.30% 52.50% 0.20%
51.27% 51.88% 0.61%
73.36% 70.54% -2.82%
53.69% 53.98% 0.29%
0.22% 0.21% -0.01%
45.53% 42.91% -2.62%
55.43% 55.22% -0.21%
51.40% 52.06% 0.66%
55.98% 56.22% 0.24%
39.47% 40.00% 0.53%
42.74% 42.63% -0.11%
58.98% 58.50% -0.48%
36.19% 36.54% 0.35%
99.61% 99.57% -0.04%
48.43% 45.03% -3.40%
99.67% 99.65% -0.02%
56.32% 56.78% 0.46%
47.39% 45.32% -2.07%
52.67% 52.21% -0.46%
22.91% 19.38% -3.53%
47.36% 46.62% -0.74%
47.21% 45.35% -1.86%
50.80% 43.49% -7.31%
52.02% 51.45% -0.57%
54.15% 52.92% -1.23%
50.52% 49.88% -0.64%
40.40% 32.34% -8.06%
45.24% 45.32% 0.08%
49.80% 49.22% -0.58%
51.06% 49.45% -1.61%
16.50% 15.99% -0.51%
44.66% 44.85% 0.19%
36.08% 29.67% -6.41%
59.21% 54.58% -4.63%
53.77% 54.78% 1.01%
51.84% 48.68% -3.16%
51.28% 50.36% -0.92%
52.64% 52.32% -0.32%
49.69% 46.07% -3.62%

Sample size: 44
(ie. 44 different whole board positions in which a human pro chose to play 'b')
average swing: -2.43%
elf agreed with the pro move: 19/44
elf thought that 'a' was a better move than 'b':18/44
Win % before Win % after swing
99.63% 99.60% -0.03%
37.00% 34.52% -2.48%
37.00% 34.10% -2.90%
38.86% 36.96% -1.90%
58.10% 51.14% -6.96%
36.22% 32.73% -3.49%
14.15% 14.47% 0.32%
82.28% 82.02% -0.26%
72.07% 72.35% 0.28%
64.86% 58.24% -6.62%
61.06% 55.65% -5.41%
49.61% 49.15% -0.46%
40.49% 39.44% -1.05%
50.18% 48.12% -2.06%
44.00% 43.98% -0.02%
62.17% 62.37% 0.20%
50.85% 50.06% -0.79%
55.04% 46.31% -8.73%
63.03% 54.25% -8.78%
39.82% 38.60% -1.22%
47.50% 46.33% -1.17%
50.83% 50.21% -0.62%
49.03% 48.14% -0.89%
65.17% 58.93% -6.24%
40.11% 37.13% -2.98%
51.48% 47.69% -3.79%
52.19% 50.04% -2.15%
57.21% 54.54% -2.67%
60.21% 56.66% -3.55%
52.37% 47.89% -4.48%
53.89% 53.07% -0.82%
61.67% 60.83% -0.84%
29.73% 29.76% 0.03%
63.51% 58.62% -4.89%
88.80% 79.16% -9.64%
52.36% 51.24% -1.12%
64.18% 61.42% -2.76%
55.48% 55.39% -0.09%
52.29% 52.00% -0.29%
70.54% 70.15% -0.39%
52.56% 49.97% -2.59%
46.21% 46.18% -0.03%
56.13% 54.16% -1.97%
68.15% 67.61% -0.54%