What was already known
Ancheta proved the p=1/2 case. Massey asked whether a time-sharing strategy remains optimal for biased Bernoulli sources.
AI involvement
Wu credits GPT-5.6 Sol with the first proof of the key entropy inequality in an interactive process. Wu simplified the proof, connected it to prior coding work and wrote the paper.
The model supplied the central first entropy proof; Wu’s later simplification and self-contained proof are separately credited.
Importance
Establishes the biased-source linear-encoding rate and resolves Massey’s question.
Limitations
Wu gives a self-contained proof and explanation, but the available source does not document a completed separate check of the retained model-derived entropy argument.
Technical verification
Proof availability and the author’s responsibility statement do not specify a completed check of the original AI argument; default Reported.
Human understanding
Wu explains the coset reduction, zero-coset entropy argument, induction and coding consequence in §§2–4.
Correction and publication history
The entropy lemma and coding consequence form one contribution.
Read this published revision