The Diff as Interrogation
Reading the diff is not a cursory scan for syntax errors. It is an act of interrogation. The Steward asks: What is this code doing? Why is it structured this way? Does this function name accurately describe the operation? Does this dependency belong here, or has the model introduced a coupling that violates the system's modularity?
Each question forces the Steward to activate their internal model of the codebase. Each activation is a retrieval event — the cognitive mechanism that Bjork's research identifies as producing the most durable learning gains. Where the comprehension is strong, the diff reads easily and the verification is fast. Where the comprehension is weak, the diff stalls, and the Steward discovers a gap before the gap becomes a production failure.
Hostility as Care
The word hostility is deliberate. It reframes the practitioner's relationship to AI output from collaborative deference to adversarial scrutiny. The reframing is necessary because the default emotional posture toward generative output is trust — the Seductive Shroud produces a felt sense of competence that suppresses critical evaluation. Structural Hostility names the counter-posture: the assumption that the output is wrong until demonstrated otherwise.
The hostility is not directed at the model. The hostility is directed at the practitioner's own impulse to accept without verifying. The interrogation is the Mind-Meld Tax. The Tax builds the capacity. The capacity is the sovereignty.
Field Note: "The hostile reader is the sovereign reader."