On 15th Feb 2018 we declared that we had finally resolved the problem of routing through this roundabout. This statement was made after a lot of testing had proven that the Detour Prevention Mechanism was working.

A lot of details needed to be adjusted for the DPM to work. The final adjustment being the renaming of the ramp segment from "A40 (W)" to "To Gloucester and Ross". This change happened on 5th Feb 2018 and once that change went live in the app, all subsequent tests showed that DPM was working. Before that change, DPM was not working.

So I believe that having "A40" in the ramp name again is now causing DPM to fail.

Would you let me prove it, again?

Changing the name to "Gloucester and Ross" (instead of "To Gloucester and Ross") would still allow DPM to work. It may not be standard but it may be necessary in order for DPM to work. DPM cannot work if A40 is in the name because it would match the A40 in the direct route. "A40 > Gloucester" matches "A40 (W)" even though the text is different, the road number part is the same. I don't know how Waze does this I just know it works like that from testing in 2017/2018.

I am requesting we change this back to how it was when it worked. Once we have got it working again we can maybe decide what to do about the non-standard name. I chose "To Gloucester and Ross" back in 2018 because that was the wording on the road sign. Also, at the time, the "To" prefix was being used widely. I agree it's better without the "To".

In terms of testing, ultimately I am using the app to test the DPM. I also use WME Route Checker and WME Route Speeds (MapOMatic Fork). I have found that testing DPM over any distance works but I do indeed check the DPM with long routes and very long routes too.

In terms of requesting an unlock so I can make a change, I see that as a pragmatic way to progress. Otherwise this process is going to take a much longer time. Like everyone else I want to solve this and move on but if others are denying my proposed change will work, without allowing me to do it, then it will not be resolved.

I am sure that DPM is not working here. My tests show that. My app shows that. This UR shows that. Please let me get on with it.

It was "A40" (not "A40 (W)")

Sorry my earlier post, which you quoted, was misleading, it should have said:

The final adjustment being the renaming of the ramp segment from "A40" to "To Gloucester and Ross".

I am seeing a lot more variance than you, Chris. That 14mph speed seems to be the default. It can be like that for hours at a time too. At other times it varies considerably.

If you look at the cross-times through the segment you can see that while the speed is 14mph, the cross-time can vary. Just now it was varying between 39s and 40s (both 14mph). Then it switched to the detour 29s, 34s, 36s, 40s. Then the cut-through 33s, 42s.

It seems to me it is behaving consistently with Waze choosing the faster route, even if there is only 1s difference between the cut-through and the detour.

Compared to the time when I was testing from Nov 2017 to Feb 2018, this behaviour was just the same until we fixed it. Same cross-times. Same switching between cut-through and detour.

Even if it was stuck on 14mph, I don't see that as too relevant to the problem because it is obviously successfully switching between the cut-through and the detour and so the variation is present on at least one of the routes. DPM should work despite the variation (unless it's extreme). The 14mph issue is probably a bit of a red herring.

All this discussion! We could have had it changed already and maybe fixed by now.

I hope we agree that the detour is being activated and that you don't think I am just making it up.

If you would like to try any changes then please go ahead and I will be happy to test.

Anyone have an automatic tester I could use?

Thanks Chris.

That all makes perfect sense and is reasonable but my confidence in the change I made is based on a lot of testing before and after that change. It was like a switch had been turned on. Suddenly it went from DPM failing a lot to DPM not failing at all.

There is always the possibility that I am completely wrong and that some other aspect changed coincidentally at the same time that fixed the issue, but with thousands of tests, day after day, it seemed a very clear and dependable result to me.

I don't see the harm in trying it though. I don't understand the reluctance to at least try it. It could be proved, either way. Last time only a couple of senior editors were interested in this and acknowledged the success. The others probably didn't even know. Subsequently, trying to apply this to other junctions I got the same reluctance.

If I'm right it could explain a lot of off and on issues, and they do crop up a lot. The trouble with most off and ons, they are very hard to test. The conditions for failure are much rarer than at this particular roundabout.

If I'm proved wrong in a fair test then I'll be satisfied and I'll shut up about it.

If anyone else is testing and thinking, no, it's working fine, then I suggest you try a restest at critical times: Weekdays 7am-9am, around 1pm and 5pm to 7pm. Something like that. I can guarantee that if you test enough you will see it failing. There may be other times but those come to mind as being times when this junction has failed recently (and in 2017/2018).

I have an automated tester that is now collecting data for the route through this junction.

Wonderful. Thanks Tim.

I have just realised something about the change I have been claiming made all the difference leading to DPM suddenly starting to work. Two segment names were changed not just the one.

The ramp segment had been "A40". The prior segment was "A40 (W)". That "A40 (W)" exactly matched the "A40 (W)" in the second bracketing segment. I changed both these segment names to "To Gloucester and Ross" on 5th Feb 2018. It is highly likely that the change to "A40 (W)" was the change that made the difference, and less likely it was the "A40" one that I have been claiming.

At some point in Feb 2018 the DPM algorithm was changed (according to the wiki). I don't know the date but maybe we can find out. The change to the algorithm may also have had a bearing on the success of DPM at the time, which I was unaware of until now. Or maybe that algorithm change happened later in Feb 2018 after the time I had completed my testing.

The algorithm change altered the check on name discontinuity and road type group discontinuity. They were to be checked only on the last segment of the detour instead of all segments of the detour.

I don't know if the success of DPM was affected by the algorithm change at this A40 roundabout because I had possibly already stopped doing intensive testing by then. It could have been broken again and nobody knew.

Other aspects that seemed to be successful and seemed to improve the DPM success rate included the change to the Road Type Group. This was achieved by changing a segment of the roundabout from Major Highway to Minor Highway on 4th Jan 2018. However, since the algorithm change of Feb 2018, that MH to mH change became irrelevant as now only the last segment was being checked and that one in the middle of the detour was being ignored.

Another aspect which seemed to affect DPM at other junctions was the notion that a ramp doesn't count because it is not in either of the two Road Type Groups. That didn't matter here until the algorithm was changed and the last segment became the key one. That suggests changing this from a ramp to an mH should now be tested.

So I now propose to change the name of the ramp to "Gloucester and Ross" and change the type to mH and then test that to see if the DPM gets switched on again.

I'm still concerned about possible partial string matches so I would still like to change the name to be sure. Later if we find that DPM works, we can try changing the name back to the country standard one to see if that breaks it.

Meanwhile we await the results of Tim's testing which I hope will have already picked up some instances I have seen when the detour was activated. I am confident there will be many more such instances during the week.

Saturday 20th Feb - Tests using WME Route Speeds - A40 Roundabout
19:00 The detour was active on long and short routes
20:31 The detour was active on long and short routes

Understood and thanks Chris.

For anyone wishing to see the effect this issue can have on navigation, please see this video which was recorded on Mon 22nd Feb:

[Video with Sound] DPM Issue - A40/A417 Roundabout

Waze tries re-routing but it's too late as we're already almost completely around the roundabout.

Thank you for the update Chris.

That's great news that DPM is now working again at this junction. Hopefully it's working at the other problem off-and-on-again junctions too.

It's possible that DPM was fixed on 24th Feb by the developers who were investigating DPM problems. That would explain the way it started working again, as if a switch had been turned on.

This junction is a really good indicator of whether DPM is currently working or not. It is very sensitive to variation and when DPM is not working it really shows up the problem, as shown by the automated test results.

If DPM should ever appear to fail elsewhere again, it would be good to check here and see if it is failing here too. It is likely that if DPM is broken again that this junction will quickly show it.

It's good to know that DPM is working at this junction now in it's current configuration. So the issues I raised about ramps, minor highways and partial strings don't apply now. It's also good that the country standard segment names such as "A40 > Gloucester" do not break DPM.

Thanks to all who were involved with setting up and monitoring the testing, raising the issue with the developers and documenting the results. Great job!