Kenneth ChengMathematical modeling

Editorial project overview

When a Reduced Ventilation Model Leaves Its Training Regime

Can nominally trained POD and POD-DEIM first pass a frozen nominal trust gate before schedule, source-location, and ventilation shifts are interpreted?

Reproducible studyUpdated 30 Aug 2026
Technical figure from When a Reduced Ventilation Model Leaves Its Training Regime
A reviewed research figure synchronized from the private technical workspace. The original aspect ratio and labels are preserved.

Editorial overview

The literature gate returned REFRAME because POD for indoor fields, ventilation optimization, boundary interpolation, DEIM hyper-reduction, and DEIM error estimation already have direct precedents. The defensible project is a transparent trust-gate audit, not a new POD or DEIM method.

The full-order reference verifies, but both reduced models fail the nominal control. The terminal verdict is STOP_REFERENCE_OR_ROM_INADMISSIBLE. Read When a Reduced Ventilation Model Leaves Its Training Regime for the frozen transport model, training and holdout design, exact nominal failure, non-attributable shift diagnostics, refuted warning and speed gates, two-run reproduction record, visual QA, and the real-building claims that remain outside scope.

Key findings

  • Nominal POD and POD-DEIM field errors are 0.181929901217 and 0.364300559015 against the frozen 0.05 gate, so the phase stops before shift attribution.
  • The 24-point residual warning has Spearman correlation -0.50 against error, below the declared +0.60 gate, and fails the worst-case ranking test.
  • Median accepted-attempt POD-DEIM online speedup is 2.978205301x against a 5x gate; timing excludes offline construction and is environment-specific.

Limitations

  • Because nominal admissibility fails, shifted trajectories are diagnostic only and cannot establish that distribution shift caused additional reliability loss.
  • The five-case sampled-residual result refutes one frozen warning rule, not all residual diagnostics or all POD/DEIM models.
  • The benchmark is a deterministic synthetic passive-scalar surrogate, not a calibrated building, health, infection-risk, controller-safety, or deployment study.

Technical record

Detailed source, calculations, generated figures, and reproduction instructions are maintained in a private technical workspace. Public articles contain only manually reviewed interpretation and approved figures.

Version history

2026-08-30 — Curated overview reviewed against repository evidence.