The accuracy of a pseudo doppler directio finder is generaly stated as approximately 5 degrees for a 4 antenna array and around 2.5 degrees of an 8 antenna array.
Of course this depends on a lot of factors:

For a stationary Radio Direction Finder:

How clean is your site? An elevated array with a free line of sight up to the horizon is a dream that will hardly ever be forfilled.
A flat desert or a calm ocean would do the trick.
In that case you may achieve an average error of 2 degrees for a 4 antenna array and around 1 degree for an 8 antenna version.
This is confirmed in real life:
NCI-Folkestone uses one of my dopplers on the very best site I know: Looking down on the sea from a 60m high cliff:
They achieve around 2 degrees of error with the 4 dipole VHF array.
But this is clearly an exception.
Even if the antenna is well above surrounding buildings and trees there will still be a lot of scatter from these obstacles.
These reflections add up to the direct signal and this will increase the overall error of the direction finding system.

For an antenna that just peeks over the roofs of your neighborhood you may expect something like the below table,
made with my 8-antenna V3 conventional doppler direction finder in the 2m band:



For a mobile Radio Direction Finder:

First of all the antenna is generally much lower than surrounding buildings and trees.
This obviously makes things worse.

On the other hand a mobile RDF can move, and if it moves enough in resonable circumstances we can gather enough information to be able to calculate a fair bearing.
In the conventional dopplers we achieve a pretty good result using a very high Q of the digital fiilter.
This makes the RDF slower but it will generally point more or less into the right direction, because a lot of faulty bearings add up and cancel each other to some degree.

In the microcontroller based RDF's we can check the credibility of incoming bearings and ignore the ones that are obviously corrupted by reflections.
This improves the bearing accuracy considerably.

But there will always be a chance that a reflection from a large building is more powerful than the original signal; in that case every Direction Finder will point at that building...

With my RDF40 I drove a route back and forth, tracking a wellknown 424MHz source, in the below picture marked as TARGET.
On three straight roads on that route I took 2 good bearings driving in opposite directions.
Back at home I used Google Earth to project my bearings on the map, and to see how much they were off.





As you can see the overall average error is not bad at all, if you manage to ignore the total chaos during most of the route.
The algorithms in the RDF41/42/43 help you to do just that, though I must say that a bit of training surely helps ;)
  For precision measurements it is much better to use only the few measurements in the best conditions,
rather than polluting the outcome with an abundance of crooked measurements.

Having said that, a mobile RDF will generally be used just to approximately point at the right direction.
Errors of up to 45 degrees don't really matter then, since at the end it will still lead you to the radio source.
But also in this type of use the very steady Long Time Average of the RDF41/42/43 will help you navigate to the radio source with a minimum of attention.
You can focus on the traffic and just now and then take a short peek at the RDF to know where to go.

For those interested in the absolute accuracy of the RDF41 with a UHF mobile doppler array, here's the outcome of a field test performed on december 4 2017:

Environment: lawn, no objects closer than 80m.
Source: 433,2MHz low power beacon, no modulation, @15m distance, 40 cm above ground.
Array: Preamped doppler with antenna spacing =17cm, ground plate = 50x50cm. Placed flat on grass.
Direction Finding Processor: RDF41.

I calibrated at 0 degrees and checked readings @ 90, 180 and 270 degrees.
Then I checked one quadrant in increments of 11,25 degrees.



The average error of these 10 measurements is 2,2 degrees, with a peak error of 4,5 degrees.
These figures are pretty close to figures stated by professional manufacturers.

Real life causes of severe bearing deviations:
As stated earlier, reflections may lead to severe bearing deviations. (multipath signals)
For example if the transmitter is behind a big obstruction or below the surrounding landscape.
Especially when the transmitter has horizontal polarisation these effects can be severe.
In these cases there 's a bad or non-existing direct path between transmitter and RDF.

And in case the hunted signal is horizontally polarized:
This will produce a poor direct signal on the RDF antenna,
but any reflection fromt a bridge or crane may contain enough vertically polarised signal to be seen as the main signal.

Unfortunately, there 's very little we can do to solve this.

Now take a look at the following picture of a city environment:



Just for simplicity, I have drawn an incoming radio "beam" as a fat red line.
(In real life, it is a wavefront of course, but this simplification will help to explain.)
The signal comes in from the right, where obviously the transmitter is located.
For the RDF car, the blue one in the picture, this signal is totally blocked by the buildings on the right.
The higher buildings on the opposite side of the road however reflect and scatter the incoming signal, illustrated by the thinner red lines.
Multiply this idea a thousand times and you will have a good idea of the truth; total chaos!
For the hunting car, the only signals coming through are false, scattered, multiple reflections.
Sometimes even crisp and clear, when the building turns out to be a good reflector, but always coming from a false direction.

There's no algorithm that can solve this, only operator skill will help you make the right decisions.
A rooky in this situation is tempted to throw the doppler out of the window...

Another example in a city environment: Tunnelling:



A signal that is striking a city with avenues running more or less in the same direction will appear to stick to that avenue.
Because that avenue offers a path with reatively low attenuation, signals seem to originate somewhere from the end point of that road.

Once the road leads you out of the city, the true bearing will reveal itself.
The bearing error due to this effect may well run up to 45 degrees.

Note that this effect will be observed only with long and good averaging settings of your RDF.
A fast responding RDF in a city environment will show total chaos.
I can't judge you eyeballs, but mine are not at all capable of "eyeball averaging" such a mess.

So, as a rooky, pick a nice quiet open road and a loud stable signal to improve your skills, and keep your eyeballs on the road.
With the proper settings and a lot of averaging, a quick look at the RDF will tell you all you need to know to take the right turn.
First try to pinpoint what in what part of the city the transmitter must be, before you take a "dive" in the city environment.

Cheers, PA8W.