Field notes

Governed Context: The Short Version August 24, 2026 · r4

The entry point to the context-window design, about 1,600 words: the diagnosis, the claim and the test it has to pass, the three edges, where the accountable seat actually sits, what is not promised, and the open problems that will decide it. Start here; the full reference is below.

Governed Context: An End-to-End Regime for the AI Context Window August 24, 2026 · design draft r9

The full design behind the second-chair argument: ratified sets at entry, residency computed from declared roots, and a record layer at exit that is never the judge. Adversarial review from independent reviewers is absorbed, the mechanism was rebuilt under it, and the open problems are named. A hypothesis under active review, not a benchmarked system.

Seven Models Read the Second Empty Chair August 15, 2026

I gave the article below to seven models. None challenged the core claim once the seat was read correctly; six of seven read the seat as a person, and the article's own FAQ had invited it. Claude wrote its own section. Nothing simulated.

The Second Empty Chair: Why Your AI's Context Window Is an Ungoverned System August 14, 2026

Session degradation in AI agents is not a capacity problem. It is an ungoverned commit process at the context boundary: everything commits, nothing is authorized, and no one answers for what is kept or dropped. The thesis, applied to the transformer context window.