The shift toward experimental development under uncertainty requires embracing iterative learning rather than seeking definitive answers prematurely.

So "asymmetric advantage" can be generated by collectively investing in systems that are built to scale, as well as supporting rapid one-off testing of ideas: small, low-cost, and expendable software and models.

Asymmetric advantage, to me, means being able to exert large impact relative to the amount of effort invested. For most, if not all, of us in the environmental sciences, there is no one tool or model to rule them all.

Australia faces compounding strategic risks with a shrinking warning window. In response, a core emphasis is to invest in experimental approaches to accelerating asymmetric advantage through innovation partnerships.

"Robust models don't grow on equations (or data) alone"

Software engineering of the model is essential

Overall an excellent example of research software development and science in action!

APSIM developers rely on automated test suites to ensure changes still meet expected behaviour.

In addition to software tests that apply observed data, we should also incorporate scientific understanding and knowledge. This is similar to property-based testing.

But what if you care about model outputs at lower-than-system levels? Does the model meet expected rules of thumbs?

Individual components are not calibrated. Instead, calibration is done with the final model after components are coupled, with their emergent behaviour assessed. The model is tested at the system level.

Modelling indicates use of tech-enhanced care could have a wide range of benefits including reduced wait-times due to less pressure on the healthcare system

Participatory model development approach helped improve and correct the model to better represent young people's lives experiences.

Trust is the currency of science. Without trust, we will fail to make a difference. Our models must be transparent and robust. We must be honest about uncertainty and limitations.

On trust and distrust in science:

Journal papers alone are insufficient to make impact.

Modelling is not just about algorithms but about people, biases, ..., and how it shapes our decisions. Australia trust scientists compared to other countries, but we are low in impact.

[cont.] Modelling is the bridge between discovery and decision. From the laboratory to the real world to help us make fair and just decisions.

#modsim2025

Rust for data science? How is it?