Conclusion

Everything above follows from one decision: put the entire stack on a single mathematical foundation.

That sounds like an aesthetic preference. It is not. When the proof system, the database, the privacy layer, and the networking layer all speak the same arithmetic, things become available that a stack assembled from borrowed parts simply cannot reach: vector arithmetic shared across layers, one challenge domain instead of several, batching by packing many values into one, and moving between the privacy path and the proving path without converting anything on the way.

A stack of borrowed components spends its performance at the borders. There are no borders here.

One operation governs both halves of the system.

Feed it random challenges and it binds a proof — the mechanism behind soundness. Feed it structured values and it expands correlated randomness — the mechanism behind private computation. One operator, one analysis, two disciplines.

That is why ring0 is not six good components that happen to ship together. Every component is a consequence of that observation, which is also why they compose: they were never separate to begin with.

Most security claims in this industry are arguments in a PDF, reviewed by people who are tired.

The formal layer checks algebraic statements and their explicit hypotheses. Protocol reductions connect those statements to authenticated state, valid execution, unpredictable challenges, and the declared adversary model. Implementation tests exercise a different obligation: whether the software follows that contract.

A local theorem, a passed test, and a measured operation are distinct kinds of evidence. None is a substitute for the complete security budget or for consensus finality.

The numbers, in one place.

None of them came from tuning. Each is what happened after replacing a component whose design predated the problem it was being asked to solve.