Research Notes

Execution Receipts and the Problem of Citable AI

AI citations are failing because they lack verifiable provenance. Execution receipts offer a path toward AI outputs that can be meaningfully cited.

GPUs Do Not Promise Determinism. Here Is the Hardware Reason.

Why running the same CUDA code twice can produce different floating-point results, and what you can do about it.

Nondeterminism as a Governance Failure

A technical governance study examining how nondeterminism in AI systems creates audit, compliance, and operational control failures across regulated industries.

Verifiability Is Not Truth

Verification proves what happened, not that the output is correct. This distinction matters for compliance and trust.

Mixture-of-Experts Systems Quietly Amplify Nondeterminism

MoE architectures add a discrete routing layer that amplifies the floating-point nondeterminism already present in GPU execution.

Token Billing Disputes Are Real Money Now

Token usage variance is a measurable financial loss. Verifiable token accounting closes this gap.

Joules Per Token: A Practical Metric

Why we measure inference efficiency in Joules per token, and how to do it repeatably on macOS.

Offline AI Without Third-Party Gravity

Most AI tools pull toward third-party services. Building truly offline systems means resisting this gravity at every layer.

CLAUDE