Model Architecture Validation 0.1 (MAV 0.1)
A first validation pass on our SIM architecture using heat conduction on a 2D plate, with fast experimental runs and rollout media that make the model behavior easy to inspect.
What MAV 0.1 was meant to test
Model Architecture Validation 0.1 was our first compact experiment to check whether the earliest version of our SIM architecture could learn a simple physical process in a way that was measurable and visually inspectable.
We used heat conduction on a 2D plate as the testbed. It is simple enough to iterate quickly, but structured enough to reveal whether the model is learning variables, conditions, and transitions in a physically meaningful way.
Why this experiment mattered
Early validation work is less about making a grand claim and more about reducing ambiguity. We wanted to know whether the architecture could finish useful runs quickly, whether the outputs looked coherent across time, and whether physics-aware structure was helping rather than just adding complexity.
That is why MAV 0.1 focused on a small, controlled setup. The point was to establish a visible proof that the architecture could move through a physical state progression cleanly enough to justify deeper work.
- 2D heat conduction as the validation environment
- Experiment runtime of roughly 45 seconds
- PINN embedded to understand variables and conditions
- Visual inspection through rollout media instead of a single static frame
What the early result told us
The encouraging part was not just speed, though getting a run done in around 45 seconds compared to a traditional 20 minutes is meaningful. More importantly, the experiment gave us a practical way to inspect how the model behaves over time rather than judging it from one output snapshot.
That makes the rollout media valuable. They show progression, stability, and pattern formation across the rollout. For an early architecture check, that kind of visibility is useful because it keeps the feedback loop grounded in what the system is actually doing.
What comes next
MAV 0.1 is not the finish line. It is a first checkpoint. The role of this validation step is to show that the architecture can support fast experiments, physically structured learning, and observable behavior that can be critiqued and improved.
That is the kind of foundation we want before moving into harder settings. Instead of jumping straight to bigger claims, we are using small validation loops to build confidence in the model class itself.