How about adjusting the test code as follows?

--- a/spec/rubocop/cop/lint/circular_argument_reference_spec.rb
+++ b/spec/rubocop/cop/lint/circular_argument_reference_spec.rb
@@ -1,7 +1,7 @@
 # frozen_string_literal: true

-# Run test with Ruby 2.6 because this cop cannot handle invalid syntax in Ruby 2.7+.
-RSpec.describe RuboCop::Cop::Lint::CircularArgumentReference, :config, :ruby26, unsupported_on: :prism do # rubocop:disable Layout/LineLength
+# Run test with Ruby 3.4 because this cop cannot handle invalid syntax between Ruby 2.7 and 3.3.
+RSpec.describe RuboCop::Cop::Lint::CircularArgumentReference, :config, :ruby34, unsupported_on: :parser do # rubocop:disable Layout/LineLength
   describe 'circular argument references in ordinal arguments' do
     context 'when the method contains a circular argument reference' do
       it 'registers an offense' do

And, by merging whitequark/parser#1052, it should eventually become possible to remove unsupported_on: :parser, so I’ll likely work toward that.

Sure, I updated the test, seems reasonable.

And, by merging whitequark/parser#1052, it should eventually become possible to remove unsupported_on: :parser, so I’ll likely work toward that.

Honestly, I wouldn't bother, especially now where basically no user runs 3.4 with parser anymore. I think the better thing to do would be to just automatically exclude Ruby > 3.3 tests on parser. Then you don't have to deal with adding unsupported_on: :parser at all (also for Ruby 3.5+). I'm sure doing it this way is possible with rspec in some way.

What do you think? Currently it would adjust let(:ruby_version) to something reasonable, so the test may coincidentally pass but there's no guarantee like in this case (or more realistically with new Ruby 3.5 syntax)

I openend #14119 for this where it defaults to prism when analyzing 3.4+ instead of 3.5+, same as rubocop-ast

In practice, support for Ruby 3.4 syntax comes from Prism::Translation::Parser. As for whitequark/parser#1052, since it was already an open patch, I just felt that it could be supported if needed. In any case, this PR is a reasonable change, so I think it can be merged. Thanks.