Isn't the model where the intelligence is?

The instinct is understandable, because the model produces the visible output. But two deployments of identical weights with different scaffolding produce markedly different quality, and organizations routinely attribute that gap to the model. What they are comparing is which instructions were present, how retrieval was bounded, what constraints applied and what happened when output failed a check.

Centralpoint locates all of that in the organization's environment: skills loading in fixed tiers, retrieval bounded by index-time classification, budgets enforced per execution, assembly retained for audit. Improving the harness improves results across every model the organization will ever call — which is a compounding investment, unlike improving a prompt inside one provider's console.