Many people already give up enough information (by carrying a smartphone) to allow for a simulated world, where their avatar's actions match their real world actions in realtime: phone calls, transactions, inventory, location, physical condition inferred from always-on sensors, connected databases, and natural language processing.

Then a controller could issue useful queries for parallel construction, like "select all avatars who have X in their inventory and who are currently driving in location Y", or how about a bar sting: "select all avatars who purchased more than X drinks at location Y and are driving, but are more than Z distance from their likely destination". Are we to presume that this is not being done? How will society react to the idea of such deep reaching surveillance?

and if it happens that NSA doesn't do it or haven't even yet collected an imagined database, it should be treated as a performance issue and negligence on behalf of the NSA.

Exactly like how all of this should have been imagined pre-Snowden. But of course it was much easier to brush off room 641A as performing simple tcpdump sessions on TV villains.

I call it realistic conspiratorial thinking. Conspiracy thinking does indeed have a lot of positives, namely that of not trusting trusted parties. The problem is it can go whack where people start trusting no one and start erecting totally insane belief structures (truthers/moon deniers).

The question I ask is: Is it technically possible? Is it reasonable for one to do it if it were possible (economically)? Then I just assume that it's done somewhere in the world.

It's technically possible to track everyone, it's reasonable that governments would want this ability, ergo a government will have done it.