ResearchReasoning Theater CoT Performative 80 Percent Token SavingsarXiv·high signalXBlueskyLinkedInCopy linkReasoning models form beliefs earlier than CoT suggests. Performative CoT dominates easy tasks. Probe-guided early stopping saves 80% tokens.SourceSource pagearXiv↳ Follow the threadStack layer / Threat patternAgent Frameworks Detect Dangerous Plan Steps and Then Execute Them Anyway; Fewer Than 20 Lines Closes the GaparXiv 2609.15293Stack layer / Threat pattern787,562 Function Pairs Show AI Code Is Half the Size of Human Code With Different Defect Classes, Not FewerarXiv 2609.12708Stack layer / ContrastAgents Convert Tokens Into Progress Faster Than Independent Sampling at First, Then Fall Below ItarXiv 2609.15309Stack layer / ContrastSplitting Decode by Attention Type Instead of by Operator Buys 31-56% More Tokens Per JoulearXiv 2609.13134Policy dependency / ContrastAn audit of a production agent pipeline found a reported p99 latency of 2,147,483,647 ms, the signed 32-bit maximum, from a lifecycle clamparXiv 2609.12017Stack layer / ContrastDistilled Byte Models Match a Token Model's Accuracy on One-Sixth the Data and Cut Logit Storage to a FiftharXiv 2609.12303Stack layer / Update threadSplitting a CTF Task Into Isolated Sub-Contexts Lets a Local gemma-4 Solve 18.52% of Challenges Standard Agent Loops FailarXiv 2609.12839Threat pattern / Update threadAdversarial Issue Descriptions Make Repair Agents Ship Correct-but-Insecure Patches in 51.7% of CasesarXiv 2609.15963