Ambient Conversational Thermal Control: Operational Test Dataset for Closed-Loop Building Simulation

This deposit contains the empirical operational test data and evaluation records supporting research in occupant-centric environmental automation and closed-loop HVAC control. This dataset serves as the companion operational test corpus to the thermal comfort training dataset previously archived on Harvard Dataverse (doi:10.7910/DVN/XKTF2L). Together, these datasets provide the empirical foundation required to conduct closed-loop cyber-physical co-simulations, evaluate noise-rejection algorithms, and analyze mechanical actuator stability under realistic acoustic environments. 1. Research Context and Simulation Scope In intelligent building automation and Occupant-Centric Control (OCC), deploying conversational interfaces introduces severe class imbalance and acoustic interference, where non-actionable ambient conversation vastly outnumbers genuine thermal comfort feedback. This dataset provides the naturalistic conversational stream used to: Execute closed-loop cyber-physical co-simulations and verify room thermodynamic convergence under multi-occupant thermal conflict. Evaluate the noise-rejection and signal-filtering capabilities of adaptive thresholding and low-pass filtering mechanisms under continuous background chatter. Assess actuator short-cycling prevention and equipment safety in mechanical systems driven by probabilistic semantic inputs. 2. Dataset Composition The corpus contains 4,000 sanitized, anonymized conversational turns structured to reflect the operational acoustic profile of a shared indoor environment: Neutral (3,650 utterances, 91.25%): Unscripted conversational background chatter used to evaluate false positive rejection. Hot (175 utterances, 4.375%): Organic expressions of warmth discomfort used to trigger setpoint cooling actions. Cold (175 utterances, 4.375%): Organic expressions of cold discomfort used to trigger setpoint warming actions. 3. File Schema The dataset is provided in comma-separated values (.csv) format with the following schema: id: Unique sequential integer identifier. utterance: Sanitized text string representing an isolated dialogue turn. label: Ground-truth operational category (Hot, Cold, or Neutral). 4. Privacy, Sanitization, and Fair Use Statement This dataset is distributed strictly for non-commercial academic research and simulation evaluation under Fair Use principles. All conversational turns have been isolated, randomized, and completely stripped of personal names, usernames, external URLs, timestamps, and origin metadata to protect privacy.