Articles and guides to accountable AI-assisted operations
Practical explanations of permissions, evidence, and human control in AI-assisted operations. Explore workflow examples, engineering notes, and research publications, with clear distinctions between demonstrated results and open questions.
Exact perfect-recovery communication cost, a bounded-error converse, and one-message bounds for simultaneous full-string recovery in BB84 monogamy games.
An external reviewer has recommended major revision and offered to reassess the manuscript. The public PDF remains a preprint; the review is not publication acceptance or peer-reviewed validation.
The general one-message interior frontier, interactive near-full recovery for n ≥ 3, and the exact physical one-message (3,1) case remain open. P0.45 R1 remains on hold.
This page presents the authorized October 1, 2026 v0.4-P2 release unchanged. Its accepted mathematical claims are frozen; this website publication is not a manuscript revision or a new research result.
Released PDF SHA-256: 63e3a02f11cce223d4ff803f136baace51ee5839b04d0b04abec737910b4da20
An evidence boundary separates what a result supports from what remains unestablished. Learn how to evaluate AI-assisted workflows without overstating the evidence.
An attempted action and a confirmed outcome are different facts. Accountable workflows keep that difference visible—especially when the connection goes quiet.
Our BB84 preprint now has a dedicated reading page. Here is how to distinguish what its results establish, what the model assumes, and what remains open.
