Whose Convention Is the Default? Interpretive Expectation as a Deployment Setting, Not a Model Property

A verbal confirmation before acting was described as competent honesty by households in one corpus and as awkward hesitation by households in the other. The model behaviour was the same. We re-coded interviews with 62 smart-home households, 29 in the United Kingdom and 33 in India, collected under near-identical prompts, against three axes of interpretive expectation: how a system should display the reading it has committed to, who is presumed to hold authority over that reading, and what makes an exchange feel finished. Two raters coded every transcript, with Cohen's $\kappa$ of 0.71, 0.68 and 0.74 by axis. The corpora diverge on all three ($\chi^2 = 8.9$, $7.4$, $5.3$), and most sharply on authority, where 79% of UK households and 42% of Indian households expected the system to defer to the user. These are patterns in two samples and not national preferences. We argue the three axes belong in deployment configuration, not in a model-level default, and give a six-field interaction log and four probes that let a community see how its own system is set and change it.