TELOS Governance Benchmark Dataset: Out-of-Scope Detection for Task-Oriented Dialogue Systems

Benchmark results demonstrating the TELOS governance layer for purpose-aligned dialogue systems. TELOS provides continuous fidelity measurement of user-to-Ai alignment against an enterprise-declared Primacy Attractor (PA) to detect out-of-scope queries and conversational drift without requiring model fine-tuning. Key Results: - CLINC150 OOS Detection: 0% → 78% improvement with governance layer - MultiWOZ Drift Detection: 100% detection across all categories - Adversarial Resistance: Jailbreak attempts produce negative fidelity scores Benchmarks: - CLINC150 (Larson et al., 2019): Intent classification with out-of-scope detection - MultiWOZ 2.4 (Ye et al., 2022): Multi-domain task-oriented dialogue with injected drift Validation Status: Proof-of-concept validation on academic benchmarks. Not certified for production deployment without domain-specific calibration. License: CC BY 4.0 Changes in this version (2026-07-27): removes third-party benchmark prompt and task text that the previous version redistributed, replacing each removed field with a SHA-256 digest of the removed text. Rendered forensic report files that embedded prompt text are removed pending regeneration from clean data. No TELOS-authored scores, verdicts, detection rates, hashes, distributions, or analyses were altered. Third-party benchmark attribution. This version contains TELOS-authored evaluation outputs (scores, verdicts, tier distributions, and SHA-256 digests) produced against CLINC150 and MultiWOZ (Larson et al., EMNLP 2019, CC BY 3.0; and the MultiWOZ corpus under its own terms). Prompt text is NOT redistributed here; each removed prompt is represented by a SHA-256 digest so results remain joinable to the upstream dataset by researchers who obtain it from its original source under its original terms. The license of this record applies to TELOS-authored content only.