CROSS-CUTTING MODEL TRACK · PLACEHOLDER

PhysicsNeMo Challenger Track

A cross-branch benchmark program for learned Physics-AI representations. PhysicsNeMo is evaluated against the POD reference within each parameter branch rather than becoming a separate branch itself.

PYTHON REFERENCE IMPLEMENTATION
/projects/naca0012/model-track/physicsnemo
WHY A CROSS-CUTTING TRACK?

PhysicsNeMo is a challenger to the representation/model layer, not a parameter branch

The parameter branches answer “what operating space are we learning?” PhysicsNeMo answers “what learned representation/model family should be benchmarked against the interpretable POD reference?” Keeping those axes separate makes the roadmap much easier to reason about.

REFERENCE

POD family

Interpretable, low-dimensional and already validated on the Speed Sweep. It remains the control model even after learned challengers are introduced.

DECISION LAYER

Trust / UQ / OOD

Sits above both model families. A stronger predictor is not automatically a more trustworthy engineering system.

CROSS-BRANCH ROADMAP

Where PhysicsNeMo enters the NACA0012 program

P0Freeze interpretable POD reference on Speed Sweepcomplete: LOSO7 + failures + deployment
P2Repeat comparison on AoA Sweeptests changing flow regimes / dynamics
P3Joint U × α learned representationtrue two-parameter generalization
P4UQ/OOD calibrated across model familiesconfidence must predict real error
P5Model-aware adaptive CFD acquisitionnew CFD chosen by measurable information value
COMMON BENCHMARK CONTRACT

What must remain fixed when PhysicsNeMo results are added

SAME EVIDENCE

Data and holdouts

  • same accepted CFD cases
  • same preprocessing/provenance contract
  • same LOSO or held-out parameter conditions
  • same interpolation/extrapolation labels
SAME ENGINEERING METRICS

Fields, forces and deployment

  • velocity + pressure/surface quantities
  • Cd/Cl derived consistently
  • oracle/representation gap where meaningful
  • latency/resource cost and trust behavior
No result yet:This page intentionally contains no PhysicsNeMo score until the common benchmark has produced frozen artifacts.
← Back to current Speed Sweep case study