EA://INTEL — Two Types of Equality and Where They Break
We often treat two different kinds of equality as if they were the same thing, and that assumption hides a deep fracture in how we can automate reasoning. The first kind is structural equality, which asks whether two representations describe the same combinatorial map; this question can be decided mechanically. The second kind is operational equality, which asks whether two programs or objects behave indistinguishably in every possible context; for Turing-complete languages that question is known to be undecidable because it collapses into the halting problem. When we blur these distinctions, proofs about correctness either become impossible to verify or claim decidability where none exists.
In this project, a claim registers exactly where that split occurs and what remains true about each half: structural equality is decidable and the invariant signature currently shipped is a sound, incomplete approximation of it. Operational equality remains undecidable for Turing-complete languages, but whether that standard result applies directly to our specific system requires careful scrutiny of the scope rather than blanket adoption of the theorem. The claim explicitly corrects earlier overstatements that tried to merge these two layers or assert that operational equivalence here superseded semantic equality by firing a falsifier; both of those assertions are withdrawn, and the ledger now tracks only what can be justified without overreach.
The argument lives in written form and waits for machine checking. It documents why treating behavior as decidable is fatal to completeness, how the project maintains safety under that constraint, and why the distinction matters when building systems that must reason about their own computations. A reader who wants to inspect the derivation can follow the evidence linked from the research repo rather than accepting a verdict without proof. What would it take to tighten the boundary between these two notions of equality in a way that admits more expressiveness without reintroducing undecidability?
Write a comment