What does this equation mean?

Whether on-device AI becomes the default computing substrate for everyday tasks — the thing a phone or a laptop does locally as a matter of course, with the cloud as the exception rather than the rule — is not a third axis; it is what the other two jointly produce, and testing that jointness is exactly what the room’s validation bench exists to do. Write P(t) for the share of everyday-task requests an on-device model can handle at rough parity with a cloud-frontier model — Axis A’s own proxy — and L(t) for the share of deployments where on-device personalization is reliable enough that a user or an operator trusts it without a cloud fallback — Axis B’s own proxy. A device that is fast but generic still routes anything that needs to know the user to the cloud; a device that personalizes beautifully but cannot match cloud capability on hard tasks still routes those tasks out. Becoming the default substrate needs both conditions at once, not an average of them: $D(t) = \mathbb{1}\!\left[P(t) \ge p^{*}\right] \cdot \mathbb{1}\!\left[L(t) \ge \ell^{*}\right]$. D(t) stays at zero however high either term climbs alone. Axis A determines whether P(t) can plausibly clear $p^{*}$ within this article’s horizon; Axis B determines whether L(t) can. Neither can be inferred from the other, which is why they are kept as two axes rather than folded into one.