The page's mandate — stop treating disposition training as the whole stack; be an architect — is exhortation of readers. Under its own scoped claim, changing who wants architecture is the character-layer method it says has a poor track record at scale.

2008 is used as proof that character diagnosis is correct and useless. Privatized gains, socialized losses, and no personal liability travel as one architecture. The page does not show which slot, reversed, would have converted catastrophic failure without personal cost rather than recoded the same crisis as insufficient ethics.

Renaming virtues as stability requirements, then taking "persisting and flourishing" as the adopted function, collapses the justification layer into those four control conditions. The two-layer cut is a preface unless flourishing is specified tightly enough that a proposed constraint can fail as not required.

The US specimen treats success as architecture and later failure as constraint layers bypassed, combined, or interpreted into flexibility. Interpretation is also the authorized amendment channel §IV requires. Until those are split, the specimen is compatible with architecture working as designed and with architecture failing.

The AI sketch names isolation without naming a privileged monitor that can halt ordinary optimization. Rust's closer is a compiler that refuses the ordinary path. Classifying Constitutional AI as a training-time disposition method does not supply that closer or a fail condition that would show isolation rather than a prior on tokens.

Martyrdom can be the detection surface, not only proof that error-handling failed. Recoding every martyr as an ignored bug report licenses an architect whose object is fewer martyrs — which can be achieved by retiring the mechanism that required superhuman effort, or by making leak-surfacing more expensive than holding the line.