@JulianWolf I'm not current on the literature anymore, but you could try looking up Devroye.

Vectorized results are not always identical with the mapped function:

foo = Subdivide[0., 2., 1100];
Sin@foo - (Sin /@ foo) // Abs // Max
Sin@foo - (Function[x, Sin[x], Listable]@foo) // Abs // Max

(* 5.55112*10^-17
    5.55112*10^-17  *)

@C.E. Thanks, @C.E.; that also worked for me. How did you know that People was the correct unit for the Population quantity — is that defined somewhere? (This doesn't work for me if you substitute the string "People" with "Robots", for example.)

WTF, there's a completely separate implementation of Export for Datasets...

Block[{Dataset`ImportExport`PackagePrivate`jsonStr = Identity},
 Export["foo.tsv", myDataset, "TSV"];
 ]

It doesn't even support Export options... so frustrating

The Dataset::exportargs is not even a relevant error message: Specific export elements for format "" are not currently supported.

anyway thanks @b3m2a1 for showing me PrintDefinitions

@JohnFeminella It isn't documented, but it is usually not difficult to find out experimentally. In this case, you could evaluate

EntityValue[
 Entity["City", {"LosAngeles", "California", "UnitedStates"}],
 "Population"
 ]

You'll see that it returns an answer with the unit "People", and then you know that this is what you have to use.

@hftf Export is actually a whole collection of sub-implementations, determined by the Formats. You register and import / export functions for file types. See FileNames["Export.m", $InstallationDirectory, \[Infinity]].

I can't find the answer know but someone, I'm guessing @WReach discusses how it's a standard representation for many different types of structured data. I'm guessing there are extra assumptions / internal knowledge they could throw in to improve the export.

@b3m2a1 I understand the sub-implementations by format, that makes sense. But it seems like the Dataset Export function should transform its data slightly, then defer to the previously existing Export functions for TSV/CSV/JSON/etc., not have a parallel, different implementation.

Also just received a sad response from support, regarding my inquiry about whether my feedback was being received: I found the last Documentation Feedback email in our records from this email address was in March, 2016. It is true that Documentation Feedback and Predictive Interface Feedback have lower priority that regular customer reports. It DOES get followed up. I do not know why Feedback from you would not arrive here. We have been receiving them from elsewhere.