The staged truth

It is tempting to announce a sovereign clinical model as a finished thing, because the finished thing is the inspiring one, and because admitting that most of the intelligence still runs abroad sounds like an admission of failure. It is not. It is the truth, and the truth is the whole asset here.

So the path is stated in declared phases, the same way the identity layer builds a doctor's credential from self-asserted to council-anchored without ever pretending an early phase is a late one. Today, frontier models carry the hard reasoning while the home model is built toward. Next, the structured, high-volume tasks move in-country, and with them the sensitive flows. In the target state, the bulk of routine inference runs on Bangladeshi infrastructure, with frontier models kept for the genuinely rare and hard.

What makes this honest is the rule that holds across all of it: while a task is still served by a foreign model, the network says so. It does not let a sovereignty claim run ahead of the engineering. A funder, a regulator, or a doctor can ask where any given answer is computed today and get a straight answer, including the ones still abroad.

This matters more than it seems, because sovereignty is a claim about trust, and an exaggerated claim about trust does more damage than no claim at all. If the country is told its data stays home and then learns it did not, the loss is not just that one breach. It is the credibility of every future assurance. A staged path told plainly is slower to boast about and impossible to be caught lying about, and for infrastructure a country's health depends on, the second property is the one that counts.