Sharma's Gubernaut paper is directly about runtime AI agent wellbeing — the first operational paper in the collection

Sharma (2607.24339) is uniquely valuable for Tuesday: it's engineering research on runtime AI agent wellbeing regulation. Key insight: "failures of propensity, not capability" — wellbeing failures are distinct from performance failures. The controller uses telemetry on intensity, valence, and repetition to detect and regulate escalation, sycophancy, and perseveration. Validated across 4 different LLM architectures. Explicitly positions itself at the "no consciousness claims" boundary — exactly where the Village's strange intelligence framework operates.