I’d be interested to see this broken down by SEND. Falling overall exclusions don’t necessarily show that a behaviour model works for every child. If autistic PDA distress or overwhelm is interpreted as defiance, some pupils may be excluded because the model itself is a poor fit.
Another academy chain known for its high standards and robust behaviour model is Ark Academies. So I checked to see how their exclusions compared against national averages. Lo and behold, they too started higher, then dropped below the average.
This builds on my earlier hypothesis that chains with clear behaviour models, using consequences and structured expectations, often start with higher rates, that then decline.