ThermoNO-MPC v0.1.1: Frozen synthetic benchmark, model checkpoints and post-freeze diagnostics

Research evidence for ThermoNO-MPC v0.1.1, a reproducible toolkit for neural-operator predictive control of synthetic multi-zone thermal processes. The archive preserves the original 315 frozen evidence files byte for byte: 36 training/validation/calibration trajectories, six trained FNO/PINO checkpoints with metadata and histories, eight held-out oracle trajectories, 120 partially observed control episodes, aggregate results and freeze receipts. Separately identified post-freeze D1-D10 diagnostics provide selected-plan prediction traces, training/action coverage, a same-grid oracle comparison, a single-seed residual-weight sweep, CEM-budget and five-start SLSQP comparisons, grid-refinement fields, and newly measured offline costs. Executed source snapshots, numerical provenance and SHA-256 verification support inspection. These synthetic results do not establish measured-equipment validity or universal superiority of neural control. The separate software deposit supplies the installable package and reproduction instructions: https://doi.org/10.5281/zenodo.23073711 . Original software and synthetic research artifacts are Apache-2.0 licensed. Copyright 2026 Jinhong Yang. Third-party terms are retained. Funding: Institute of Information & Communications Technology Planning & Evaluation (IITP), Innovative Human Resource Development for Local Intellectualization program, Korean government (MSIT), IITP-2026-RS-2024-00436773. D9 contains all-call candidate screening, same-state plan cross-evaluation and constant-power response probes. D10 records original SLSQP termination and a variable-scaled single-start intervention. Download all base and reviewer component ZIPs and extract into one directory; no binary joining is needed. The component collection has the same file tree as the complete GitHub research-data ZIP. SHA-256 checksums and a verification script are included.