perf triage classification of PR 89611 was "probably changed"; why not definite?

[pnkfelix]

Looking at the performance report linked here:

Benchmark & Profile	Scenario	% Change	Significance Factor ?
encoding opt	full	-0.80%?	2.21x
encoding opt	incr-full	-0.72%?	2.36x
coercions check	incr-patched: println	-0.64%	4.82x
helloworld debug	incr-full	-0.53%?	2.07x
helloworld debug	incr-unchanged	-0.52%	4.79x
coercions check	incr-unchanged	-0.51%	7.92x
helloworld debug	full	-0.45%?	1.83x
coercions opt	incr-unchanged	-0.43%	5.67x
ripgrep opt	incr-unchanged	-0.42%	2.73x
cranelift-codegen opt	incr-unchanged	0.38%?	1.81x
coercions debug	incr-unchanged	-0.36%	3.87x
issue-58319 debug	incr-unchanged	-0.34%	2.09x
token-stream-stress debug	full	-0.31%	1.74x
token-stream-stress debug	incr-unchanged	-0.31%	2.25x
coercions doc	full	-0.25%	2.89x
unicode_normalization check	incr-unchanged	-0.24%	3.79x
coercions check	full	-0.24%	3.79x
issue-58319 debug	incr-full	-0.24%	2.00x
tuple-stress debug	incr-unchanged	-0.24%	1.94x
helloworld opt	incr-patched: println	-0.21%	2.55x
helloworld opt	incr-unchanged	-0.20%	2.19x
coercions check	incr-full	-0.20%	5.63x

why isn't this classified as a definite improvement in the performance triage script?

Yes, a lot of them have question marks on them (though I'm not sure the performance triage script includes the question marks in its weighting at this point, based on what I saw elsewhere), but even if you remove the question mark cases, you're still left with e.g.

Benchmark & Profile	Scenario	% Change	Significance Factor ?
coercions check	incr-patched: println	-0.64%	4.82x
helloworld debug	incr-unchanged	-0.52%	4.79x
coercions check	incr-unchanged	-0.51%	7.92x

which seems like it should have been enough for being classified as a performance win, not a "probably" ?