sketch wrote:OK, so, the "penalty" is only when there is a detour. There was no detour in this case.

They said this:

3 road type issue, we have a high penalty when you're on major highway -> exit to ramp -> enter to primary street, to solve it try to change the ramp to primary street or major highway

to mean the small detour prevention?

I thought the road type had to be the same for the highway you are detouring off and back on. I also didn't think primary streets played any role in the small detour prevention. Are saying that if you go MH>R>PS and the major highway and the primary street have the same name/city (or alternate name/city) the small detour prevention kicks in?

Thanks. Didn't see the comments before. And you of course are right, I'm getting my detours mixed up.

So the conclusion is they have retracted their reasoning that this is due road types and simply state that this is due to the fact that it is really faster to avoid the jughandle. Alright, so we do carry on with our current guidelines.

As with most of the intersection related questions, I'm inclined to not to worry too much about adjusting guidance until we see what the long awaited "junction box" may be.

I have the same impression as sketch that the timing of the final turn instruction depends on speed rather than road type.

PesachZ wrote:Keep in mind an AGC set with a shallow angle (stay) going from a highway (mH, MH, FWY) to a non- highway for a right turn, will give an exit instruction. If that seems wrong and will confuse wazers, a dogleg should be used to generate a turn instruction.

While I understand the logic here, I'm not sure that users aren't also confused by early "turn" instructions, which seems to be part of the reason for DwarfLords post.

DwarfLord wrote:

shawndoc wrote:I would like to...remove the bolded section: "There are only a few situations in which connectors are called for in an intersection: When you get an automated report (map problem) that the roads are too far apart due to the intersection being too far from the turn lane." As written, we'd be adding turn lanes to just about any decently sized intersection that uses a mapcat bowtie. I feel the rest of the reasons listed for adding a turn lane cover the times when a turn lane is needed. The MPs about roads being too far apart have too many false positives to be used as a justification by itself to add a turn lane.

Based on the "thanks" this post received, there's apparently consensus that even a junction that throws missing-road MPs does not qualify for an AGCs based on that behavior alone. To me this says that the distance the turn path strays from mapped segments while cutting the corner should not be a factor when deciding to map an AGC. The physical presence of a dedicated ramp-like connector would be a factor! As would the presence of other nearby segments to which the driver turning could be snapped by mistake. But distance all by itself would not be a factor.

I agree based on the properties of the waze experience that we as editors can actually control. But what this doesn't take into account is any reasoning waze has for when the missing road road MPs are triggered. If the MPs are indicating that waze can't properly calculate turn times because the turns are too far from the intersection, then maybe we should give more weight to the MPs. However, this is something the waze staff should let us know. The only things I've heard on the subject from waze staff is to wait for the "junction box."

Except if I've missed a post in another thread, I've notice a set of unexpected changes on this wiki page performed on May 28th.

I would like to draw you attention on it and to get an community approval on these changes that don't seem to be all on the cosmetic facet (there is some rephrasing, but there is also some content addition and new statements)

Having now finished the vast majority of FC in the Santa Clara & San Jose area, I'm faced with a multitude of AGCs that need either to be promoted or deleted. I'd be really grateful for guidance.

Many of these AGCs show as first created by junior editors some time ago. There seems to be a general sense that AGCs are mostly harmless; it appears most of them in this area were kept by default without much thought.

Well, I've come to worry that AGCs are not mostly harmless, for the following reasons:

• Cases where a driver asks for a route while waiting for a light in the slow lane; unbeknowst to the driver he is closer to an AGC than to the main segment, so Waze snaps him to the AGC and gives him a wrong route assuming he's irrevocably committed to the turn.
• Unnecessary AGCs serve as examples to new editors, who may then go on to create more unnecessary AGCs.
• Warnings can be TOO early. Thanks to FC, many roads with AGCs are now minor or major highways. Does Waze give earlier turn warnings on those road types? If so, are we less likely to need AGCs on them? I just saw a UR in which a driver on a 6-lane (combined) mH turned right 161 m before he was supposed to. And there wasn't even an AGC involved.

So, I'd like to assess the AGCs in my area with a fresh eye and not just promote them all willy-nilly! Here are quotes from earlier in this thread, and my take-home message from each:

shawndoc wrote:I would like to...remove the bolded section: "There are only a few situations in which connectors are called for in an intersection: When you get an automated report (map problem) that the roads are too far apart due to the intersection being too far from the turn lane." As written, we'd be adding turn lanes to just about any decently sized intersection that uses a mapcat bowtie. I feel the rest of the reasons listed for adding a turn lane cover the times when a turn lane is needed. The MPs about roads being too far apart have too many false positives to be used as a justification by itself to add a turn lane.

Based on the "thanks" this post received, there's apparently consensus that even a junction that throws missing-road MPs does not qualify for an AGCs based on that behavior alone. To me this says that the distance the turn path strays from mapped segments while cutting the corner should not be a factor when deciding to map an AGC. The physical presence of a dedicated ramp-like connector would be a factor! As would the presence of other nearby segments to which the driver turning could be snapped by mistake. But distance all by itself would not be a factor.

CBenson wrote:Well personally, I don't find the vast majority of ACGs at cross roads to provide any clarity when I drive. I can figure out to be in the right turn lane for an upcoming turn right instruction and in the left turn lane for an upcoming turn left instruction. So I'd be happy just eliminating most ACGs. On the other hand some users clearly prefer the earlier instructions and all things being equal I prefer more information to less. So I'm not advocating eliminating ACGs where they are separated from the intersection.

This supports the "when in doubt, leave it out" AGC doctrine and suggests that situations requiring early nav instructions for a turn are quite rare.

PhantomSoul wrote:I also personally don't care about getting an instruction to keep/turn/exit/etc. at the point where a solid line begins for a dedicated turn lane. Such an announcement falsely gives the impression that individual lane guidance is available, when it is in fact not. Most of the time, Waze even has to guess which roadway of a dual-roadway highway you're on! However, many people do find this useful, so to reach a better balance between that and avoiding map clutter, I'd be ok with a 50m rule, meaning that if the beginning of a dedicated turn lane, or the AGC itself, is set back from the center of the intersection proper at least 50m (~150 ft), we can include an AGC. Distances of less than 50m are actually trivial to Waze and AGCs entirely within that range amount to just noise on the map.

This suggests that AGCs advanced less than 50 m from an intersection are almost certainly superfluous and should either be advanced further or removed altogether.

sketch wrote:The commitment point (beginning of a dedicated turn lane -- ed.) is a good place to put the "stay left" if the AGC is designed to give a "stay left" followed by a "turn left" for particularly long or physically separated turn lanes or particularly complex intersections.

However, for AGCs designed to give a "turn" instruction at the start point of the AGC (in the case of, say, physical separations), the commitment point is not the optimal location...Putting it too early has the disadvantages of (1) the turn instruction leaving the screen too soon, (2) the next prompt spoken too soon, and (3) inconsistency with timing of non-AGC turns...The "turn" instruction is expected to be based on the actual point where the turn is made, whether you are guided by a physical island or not.

This is saying that an AGC should not be designed (via dogleg etc) to give a "turn" instruction except close to where the driver will actually start turning the wheel. An AGC constructed to give lane guidance warning in advance of the actual turn should give a "stay" instruction instead, but such AGCs are only for dramatic situations ("particularly long or physically separated turn lanes or particularly complex intersections").

sketch wrote:...AGCs aren't generally used for simple single-white-line turn lanes.

If a normal turn lane is...
-- ROAD -- TURN LANE DELINEATED BY SOLID WHITE -- INTERSECTION

And a separated turn lane is...
-- ROAD -- TURN LANE DELINEATED BY SOLID WHITE -- DIVERSION FROM PARALLEL -- INTERSECTION

In the first situation, the "typical intersection", we don't put in an AGC for the turn lane, so the turn instruction is timed from the intersection itself. This works fine in most cases, so typically turn lanes like that aren't drawn in. You're already expecting the turn, so you should already be in that lane anyhow. In the second situation, we put the AGC in at the point where traffic diverges from parallel to the normal lanes. That's the point where, like the intersection itself in the first situation, you actually turn and face a different direction, so it's the point where users would expect to hear the instruction.

This appears to say that AGCs should mark the physical divergence point, not the commitment point. This is contrary to much of the practice I've seen in my region. If it were followed, then combined with the 50-m rule proposed earlier, a lot of AGCs would be ruled out immediately!

PhantomSoul wrote:

russblau wrote:I don't understand why "in 0.2 miles, take a right at Highland Street" is not enough to alert the driver that they should move into the dedicated turn lane.

Because if there's 4 lanes to cross on a 50-mph road to get into that turn lane before the solid white lines starts, the 0.2-mile (roughly, 1000-ft) warning is nowhere near enough time for a driver to get over from the leftmost lane.

Like I said, on a road with only 1 or 2 through lanes, it's not as big of a deal. But on road with 3 or more lanes, particularly ones with heavy traffic, it becomes much more of a problem.

This appears to say that except for physically significant connector lanes, intersections of roads each with up to 4 through lanes both ways are very unlikely ever to need AGCs.

My take-home message from this thread is that AGCs are indeed overused in my area. In most or all cases of 4-through-lane (combined) or smaller highways without significant physical turn structures or commitments they can be removed. Further, AGCs are doing little good at intersections where they are advanced from the intersection by less than 50 m.

For larger intersections the AGC tradeoff is less clear, especially as Waze has tweaked warning times a bunch over the last year and as a lot of roads that were Primary Streets when the AGCs went in are now Minor and Major Highways. If the "mark the point of physical separation, not driver commitment" and "within 50 m an AGC is superfluous" principles were put into practice, then even some very large highways might not warrant AGCs!

In the absence of additional guidance, it seems the best way forward is to err on the aggressive side when dealing with questionable AGCs in my area. But additional guidance would be most welcome

Thanks for all the thoughts!

Here's an example of cases where, at least based on my understanding of what I've read in this thread, I can't come up with a consistent, persuasive rationale to retain the AGCs:

https://www.waze.com/editor/?env=usa&lo ... 4,61329933

In the lower left is an AGC from Montague Expy to E Trimble Rd. Both of these are pretty big roads, Montague is 4 through lanes in the AGC direction and E Trimble is 3. But, the right turn has only the tiniest of separating islands and there's no way it would be called "ramp-like"; the physical divergence and commit points (they are the same here) are only about 40 m before the center of the intersection.

To the right are some ramp-like AGCs associated with small streets that meet with Montague Expy. Structures like this are very common in the area. The commit point is about 50-60 m before the intersection; the divergence point is only a few meters before the intersection. I do not see why modeling these as ordinary butt-joint intersections would be inadequate.

If I'm missing something please let me know!

(EDIT: I believe it's OK to disregard last-modifier data on the segments mentioned above. Scripts occur frequently in this area that change last-modified-by data to indicate a senior editor for a significant fraction of segments on the map, including even red roads. The editor's handle as last modifier should not automatically be taken to indicate any particular endorsement or attachment to the segments in question.)

Here's an example of an AGC that I need to decide whether to promote to mH; promote and shorten; or remove.
I'm thinking remove, because:

• The distance from the commit point (about even with the gas station entrance) to the center of the intersection is 56 m. PhantomSoul's perspective is that 50 m or less is just noise to Waze, so there is no point in shortening this AGC to match the commit point. If one were to shorten it one may as well delete it.
• The physical separation begins only a few meters from the intersection. If I understand sketch's perspective that is nowhere near adequate justification for an AGC based only on the physical characteristics of the turn lane.
• The origin road is 4 lanes one direction, but the destination road is only 2. It seems unlikely to throw a missing-road MP and we generally are not that concerned about this situation throwing MPs anyway.
• That means the only reason to keep the AGC at all is to advance the turn warning. It is hard to say how necessary that is; my guess is, not very; see the screenshot I attach at the end of this post.
• A complicating factor here is the right turn into the gas station which comes physically EARLIER than the intersection but would be announced LATER than the AGC, due to the AGC's crossover! If the right turn to the gas station doesn't need any AGC for advance warning to drivers, why does the right turn at the intersection need the advance warning? Granted a lot more drivers will head for the intersection than for the gas station.
• One could tie the gas station entrance off the AGC instead of the main road, but then the driver hears "stay right, then turn right" which IMHO would be confusing to a driver heading for the gas station.

So, can someone persuade me that this AGC should be retained?

Before saying anything, take a look at this recent UR trace. This is the case I referred to earlier, where a driver turned right by accident 161 m early, and there was not even any AGC involved.

I updated the figure to add the clarification but although I see the updated image in the preview -- and I even see it in the history at the bottom of the page as I write this! -- it seems to be invisible in the thread (the forum code is finicky about images, isn't it...?). Here it is again.
(EDIT: oooookay, now that I post this the revised image does appear in my earlier post. Rather than remove this instance and potentially break things again I'll leave it. Sorry for the redundancy.)

Another example suggesting that, with the latest app & nav situation, early turn guidance may not be as necessary as it used to be. In this case the driver turned right 427 m too early (!!) and again, no AGC was even involved.

sketch wrote:A user has a duty to look at the damn road when using GPS.

You won't get any argument from me on that count!

The example was not meant to complain that Waze's turn instructions are too early; apologies if it sounded that way. Rather, it was meant to show that creating or retaining an AGC for the sole purpose of advancing (hacking?) the turn alert timing may not be much benefit, at least not with the nav as it is currently tuned. In fact, it suggests that AGCs could potentially be counterproductive in cases where turn alerts are plenty early enough.

All of this arises because parts of my area -- especially parts with large but unsplit roads -- seem to enjoy a "when in doubt, put in an AGC" ethos. New editors largely created the AGCs simply to fill what looked to them like a vacuum on the aerials, but once installed, their presence reinforced the ethos, and thick and fast they came at last, and more, and more, and more.

The completion of FC here seems a good excuse to review all these AGCs with a critical eye, so that's a-what I'm a-doin.