REPORT Burdisso_Idiap-RR-05-2026/IDIAP DialFork: Going Beyond Factuality to Evaluate Procedural Hallucination in Conversational Agents Burdisso, Sergio Villatoro-Tello, Esaú Sanchez-Cortes, Dairazalia Labrak, Yanis Luque, Jordi Cernak, Milos Motlicek, Petr Benchmark conversational agents dataset hallucination detection procedural hallucination task-oriented dialogue EXTERNAL https://publications.idiap.ch/attachments/reports/2026/Burdisso_Idiap-RR-05-2026.pdf PUBLIC Idiap-RR-05-2026 2026 Idiap October 2026