The claim
A6, CASE-089: E₁ → H ← E₂ in two worlds, prior odds 1:1, E₁ and E₂ conditionally independent. The Bayes factor of E₁ is 10 in both. That of E₂ is 20 in world A and 1.1 in world B. Discounting E₁ leaves odds of 20:1 in A and 1.1:1 in B, as A6 said, checked here. “The unweighted graph has the same topology in both cases.”
A8 drew the record’s consequence: “Those edges preserve aspects of dependency and provenance. They do not, by themselves, encode a complete probabilistic model of evidential support.”
Why it is granted
It cuts against the strongest reading of CLAIM-295 and of CLAIM-250, that keeping justificatory structure keeps what revision needs. The thesis has to be stated so it survives: CLAIM-295 says justifications can widen what a retained state suffices for, not that topology suffices. A6: “This is an important case because it blocks a careless inference from ‘justificatory structure matters’ to ‘a graph of justifications is sufficient.’”
What it does not say
- Not that graph memory is insufficient as such. “An attributed graph could include likelihoods, dependence models, assumptions, and other relevant information.”
- Not that the record should weight its edges. That is a design question for luria, and the owner’s U14 objected to turning the exchange toward it.