Out of curiosity I had a look at the link above and could not work out the route you were hoping C.T would create. Bigger clue needed, please.

Sorry about that. The URL didn't move around as I moved the pins around. Looks at this:

https://cycle.travel/map/journey/117601

There exists a road in OSM directly between the two pins. Thanks.

Guessing here because I’m on mobile and it’s difficult to find that way on OSM, but I suspect the missing way is tagged simply “highway=track” with no surface information. In the US, cycle.travel ignores such tracks because they may very well be unrideable. Adding surface or tracktype tags should allow cycle.travel to route along it after the next update. 

Thanks for the reply. That is certainly it. Two thoughts:

1. In the US there are a LOT of dirt roads that are tagged only as highway=track. Ignoring them limits the usability of the system, since many valid routes are omitted, such as the one here. Yes, adding them will create an opposite problem: routes that use nonexisting roads. In that case the user at least has the option to move pins around to bypass a section they know is missing. With the way cycle.travel works right now, I can't even semi-manually fix the route.

2. If we do want to ignore such ways, can we adjust the rendering to indicate this to the user? If you zoom-in on the Death Valley link above, you'll see both routable and non-routable ways drawn in exactly the same way, which isn't helpful.

Thanks!

Yep. It’s a deliberate trade-off – omitting bare highway=track is a little inconvenient but greatly reduces the memory needed for routing (which is 50GB-ish for Europe and North America, and would almost certainly tip over into needing a new server if bare tracks were included).

You can plot a route including the missing section using the “Go direct” feature – put a via marker at each end, then click the first one and choose “direct”. 

Good spot on the cartography – that certainly shouldn’t be the case. (Rideable tracks are meant to be shown as dotted brown.) I’ll try and work out what it’s doing there. 

Thanks for the "direct routing" pointer. I didn't know about it, and it's a decent workaround. One request: as a workaround to this particular problem, it'd be useful to be able to place markers in non-routable areas (currently pins snap to areas on the cycle.travel graph). These pins are ignoring the graph anyway, and it'll allow manually-routed sections to approximate reality.

Thanks!

Definitely. I’ll confess that the part of the routing engine (OSRM) that does that processing isn’t a part I fully understand, so unfortunately it’s not something where I can happily say “oh, I’ll fix that”. But it is something I’d like to implement. 

I had a bit of a virtual explore of that area using the few photo's in google street view. It's beautiful.