EVIDENCE-BOUNDED ANSWERS
Focused answers for problem-first searches. Each note separates a practical procedure, the evidence currently available, and the claims that the cited work does not establish.
Focused answer
A stage-consistent comparison protocol for code-space and token-space masked diffusion language models when the discrete codec is lossy.
How should code-space and token-space masked diffusion language modeling be compared?
Read the evidence noteFocused answer
How to prevent unnecessary whole-function rewriting while preserving enough freedom for a generative model to make the requested code change.
How can a generative model modify code without rewriting the entire function?
Read the evidence noteFocused answer
A practical distinction between grammar constraints, type constraints, preservation boundaries, and behavior-level acceptance checks for generated code.
What does constrained code generation guarantee in a software-engineering workflow?
Read the evidence noteFocused answer
How to evaluate recent AI-assisted refactoring methods without confusing a plausible generated patch with verified behavior preservation.
Which methods and evidence matter for AI-assisted behavior-preserving refactoring?
Read the evidence noteFocused answer
Why deterministic sampling is not enough, and how observable protected properties and acceptance checks make code-generation behavior testable.
What makes code generation predictable rather than merely controllable?
Read the evidence noteThey are maintained routes into the two published studies and their nearby methods, not new experiments, product claims, or substitutes for the official papers. Numerical statements remain tied to the exact experimental setting and limitations shown on each page.