This interactive explorer needs its page script, and it did not run. If you are reading this inside the blog, the explorer was embedded in a way that stripped its script: embed ta2_explorer.html as an iframe instead of inlining its markup.
relevant unclear irrelevant golden   looping detrimental suboptimal
How to use this
One click on a run or its name selects it: you are looking at that run, details on the right.
Double-click zooms to that run alone; works on names and the golden line. Double-click empty space or "Back to main view" returns.
Click any step dot for that step's full judgment; golden steps too.
Scroll zooms, drag pans while zoomed; "Reset zoom" restores the view.
Compare judges puts Opus 4.8 and GPT 5.6 Sol side by side, with both reads in the step details.
Hover any control in the toolbar for a one-line reminder of what it does.