The intent-horizon construct (IH-1) and its security model (IH-2) specify what governed agentic systems must protect; the structural account of confident misalignment (IH-3) specifies what failure of that protection looks like mechanically. What the prior papers in this series do not supply is the **operator-facing diagnostic surface** — the instruments by which an operator reads horizon-state in real time and decides whether to intervene, wait, escalate, or re-pin. This paper specifies that surface. The diagnostic surface is named **epistemic weather** for an operationally important reason: horizon conditions are continuous, not binary. The operator does not read a single horizon-status check; the operator reads a *weather report* across multiple boundary dimensions, continuously updating, in which conditions are graded by class rather than reduced to a pass/fail outcome. The metaphor is load-bearing: weather reports are the operator's interface to a complex, evolving, observable-but-noisy phenomenon; horizon state is exactly such a phenomenon and deserves an analogous interface. This paper specifies four weather classes — **clear, drift, fog, storm** — and the instruments that read each one. Clear is the steady-state operating mode: the system is horizon-integral along all four boundary dimensions, intent-CIA-NR properties are intact, no operator intervention is required. Drift is the early-warning condition: at least one horizon boundary is approaching crossing; the system is still horizon-integral but the conditions are evolving in a direction that will produce a crossing absent recalibration. Fog is the operator-side comprehension degradation: the system's self-reports are no longer reliably readable by the operator (the operator's instruments cannot tell whether the system is clear or drifting). Storm is the active boundary crossing: the system is in a confident-misalignment state (per IH-3's structural account) and the architecture must fail-closed to operator intervention.