Why this benchmark exists When Anthropic released Opus 4.7, the Every team had no early access — so they ran their vibe check live. To measure how the model handles messy, real-world code, they used the Vibe Slop Benchmark: a special test built around Proof's own history. How it works The setup is simple but demanding: - Input: a frozen snapshot of Proof's production codebase from when it was pure vibe-coded slop — fast, unpolished AI-generated code.- The task: rewrite the codebase from first principles, the way the senior engineer who later cleaned it up actually did.- What it measures: whether a frontier model can find the core invariants and propose a genuinely bold refactor, not just nibble at the edges. What it revealed about Opus 4.7 In the live test, the model got the diagnosis right — it spotted that there was no single authoritative model of who owns the current state of a document, and proposed a nice refactor of a 10,000–20,000-line file. But it showed less courage: it planned the burn-the-ships rewrite, then hesitated and only made incremental changes. That gap between planning and execution is exactly what this benchmark is designed to expose. Why it matters The Vibe Slop Benchmark is a useful proof-point for measuring progress on the messiest, most realistic coding task there is: turning sloppy AI-generated production code into something a senior engineer would actually ship. It pairs well with directives like burn the ships or rewrite from first principles when coaxing bold rewrites out of a model.