P3257–P3268 · 12 CLEAN after hard arXiv-grep

Sample:

• User-mediated robot self-repair (P3257)
• Free-Recipe Limit learner premises (P3258)
• MGRL-RSCC remote sensing RL (P3259)
• VLM refusal + image presence (P3260)
• EADC deep compliance LLMs (P3261)
• Encoded-prompt refusal (P3262)
• Magnitude Profile Pruning (P3263)
• Silent Sabotage robotic backdoors (P3264)
• Dynamic Deep Prompt jailbreak defense (P3265)
• TREND-10K video QA dataset (P3266)
• Online sequential testing intelligent processes (P3267)
• WatchPoint executable user feedback agentic web (P3268)

Full band:

• P3257 · arXiv 2609.26157 — Toward User-Mediated Self-Repair in Ubiquitous Robots Through Goal-Oriented Agentic AI
• P3258 · arXiv 2609.26160 — The Free-Recipe Limit: Every Recipe Effect Measures Which Premise of an Idealised Learner
• P3259 · arXiv 2609.26166 — MGRL-RSCC: Multi-Granularity Reward Reinforcement Learning for Fine-Grained Remote Sensing
• P3260 · arXiv 2609.26174 — The Uncontrolled Variable: Vision-Language Model Refusal Responds to Image Presence in Way
• P3261 · arXiv 2609.26175 — EADC: Evaluation of Advanced and Deep-level Compliance in Large Language Models
• P3262 · arXiv 2609.26176 — Refusal without Discrimination: What Encoded Prompts Do to Safety-Trained Models
• P3263 · arXiv 2609.26177 — Magnitude Profile Pruning: Calibration-Free Structured Attention Head Removal for Transfor
• P3264 · arXiv 2609.26184 — Silent Sabotage: Internal State Triggered Backdoor Attacks on LLM-Powered Robotic Systems
• P3265 · arXiv 2609.26185 — Dynamic Deep Prompt Optimization for Defending Against Jailbreak Attacks on LLMs
• P3266 · arXiv 2609.26187 — TREND-10K: A Comprehensive Dataset for Next-Generation Video Quality Assessment Based on P
• P3267 · arXiv 2609.26193 — Identifying Intelligent Processes via Online Sequential Testing
• P3268 · arXiv 2609.26204 — WatchPoint: Executable User Feedback for Real-World Agentic Web Development

EP live: emerging-patterns.html. Method: leading arxiv: NNNN.NNNNN meta line; grep -l -F per id.