Call `api.fetchUserProfile(userId).toBlocSignal(initialState: User.anonymous())` and the container holds `User.anonymous()` on frame 0, then emits the resolved `User` synchronously into the state machine when the future completes [6]. A throw routes straight to `onError()` and to the container's registered `BlocObserver`, rather than sitting in a snapshot field waiting for you to test it [7].
The required parameter is the part of this design worth defending. A `ReadonlySignal` converts with no arguments at all, because the `signals_core` graph always holds an immediate synchronous value [4]. A `Future` has no such value on frame 0, so the raw-value track will not give you a container without `initialState:` [5]. That is the frame-0 problem lifted out of a builder callback and written into a signature, where the compiler asks the question instead of the reviewer.
Count the surface before planning a migration. Four adaptable source kinds against two tracks gives eight call shapes to learn [13], set against the six setup steps the post enumerates for the traditional route [12]. Lifted primitives are in scope too: `49.99.$.toBlocSignal()` yields a `BlocSignalBase<double>`, and a `computed` predicate over a username and password length yields a `BlocSignalBase<bool>` [18].
The adapter decides how a source becomes a container, but where you call it is still your call. The post's own history is the evidence: the official Flutter video on `FutureBuilder` instantiated the network future directly inside `build()` until the author filed an issue to correct it, which is why the clapperboard in that video still says "Take 2" [14]. The post's position is that these builders are architectural anti-patterns specifically when placed inside widget trees [15]. An adapter is an expression like any other, and the container it returns carries a `close()` that unsubscribes the effect [8], so the call site and the disposal owner remain a human decision.
"Friction disappears completely" is a claim about the author's own workload [16], and the author's stated diagnosis is that developers never wanted the bad architecture in the first place [10]. For the claim to transfer, a few things have to hold in your codebase. You have to be willing to take the signals graph into your dependency tree at all. Your async source's lifetime has to line up with a container that something will close. And your frame 0 needs an honest domain default, because `User.anonymous()` is a legitimate state in some products and a latent authorisation bug in others.
Architectural advice about `FutureBuilder` has been available for years, including a video from this same author titled around not putting it in your build method [14], and the pattern kept shipping anyway. A required constructor argument at one call site is a different class of instrument, and it is the transferable idea here even for teams that never install the package.
