Currently requests fall back to the network when something is wrong (no provider host, for example).
That was introduced when service workers are not reliable enough, but we can revisit the error reporting logic now.