Arc App v3.12 is now live! It ticked off a few requested features, including a couple from the previous poll. Now let’s work out what we want the focus to be for v3.13 
What is the Gyroscope import? What kind of data can br imported?
@o1x Gyroscope has a timeline recording feature similar to Arc, so there used to be a lot of demand for an import feature, for people wanting to move from Gyroscope to Arc.
I guess there might be other kinds of data that Gyroscope collects, that might be worth importing into Arc at some stage (assuming Gyroscope has an accessible API). But for now the goal would be simply to support migration from Gyroscope to Arc for recorded timeline data.
I should mention that “An easy way to extract a visit from within another item” refers to this feature request. I also brainstormed some further ideas for it somewhere in a discussion here. I really should find that and link it as well, to stop it getting lost 
Can I add “Add Hiking Activity Type” to the list? 
Sorry, just realized that this is covered by Custom Activity Types… 
Hah. Technically yeah, it could be covered by custom activity types. Though you’ve made a decent case for adding Hiking as a built in type anyway. And I just now found Apple’s SF Symbols icon for “figure.hiking”, which would do nicely as the activity type icon. I’ll get it added shortly!
Custom activity types is my absolute #1 feature I’d love to see; I have activities scattered over the years that I’m patiently dying to finally categorize correctly 
I’d love for the map to zoom to segments while they’re being individually edited, and not just split.
I’d appreciate seeing a trip speed graph when reviewing unconfirmed activities, and not just after they’ve been “confirmed.” A speed profile could help you to rule out certain activities when you can’t well remember how you actually moved in an area that you frequently move through differently.
And I’d love to see the one-time visit (and all naming) option(s) for stationary points within an unconfirmed trip (and not just when selecting from timeline view). I saw you replied to my earlier reply and are already looking to handle it!
I’ve made a changemap suggestion for each of these, assuming that’s the best way to go about it.
Definitely the best. Thanks! I’ve voted for all of them myself
Here’s links for anyone else who wants to vote for these:
Related to one time visits or others found only by address, is there or could there be a way to assign contacts or names? What happens when visiting friends or family?
Also as far as clean up segments perhaps there could be a way to change the circle radius e.g. sometimes a place radius overlaps with other places and one wishes to refine or limit the cleanup. I’m not sure if useful yet maybe there could be some cleanup all past segments for some given radius.
I would love to see a quick way to mark a segment as Stationary from the tap-and-hold pop-up in the timeline. It’s such a common task, marking random 1-2 min segments that turn up while at a location as stationary.
Similarly, it’s often that between two locations there are segments of multiple types - like, car, train, transport, and I need to go and turn them all into train. It would be cool if that pop-up also offered a dynamic shortlist of types based off movement types between the two locations.
And, a bulk edit functionality for the timeline segments! Select a bunch of segments and mark them all X. This would actually be great combined with the above. First take a pass with a bulk edit, then clean up with the tap and hold.
Yeah, I want this too, along with the convert to stationary option in that long press menu. Would be super helpful.
Though I think the catch is I’m not sure if it’s possible. Like, getting a shortlist of activity types for the Path item costs CPU time and energy. It requires firing up a classifier for each item, which also means loading all of the item’s samples into memory. The timeline view instead is constrained to only showing the high level details of each item - the details that were stored in the database the last time they were calculated - which avoids having to load all the samples, fire up classifiers, etc.
Although… presumably that long press menu could load the samples and classifier on demand. I wonder if that’s possible with that particular UIKit widget. Hmm.
This is the kind of thing I typically refuse to do on the grounds that if there’s that much repetitive work to be done, it means there’s something wrong with the underlying system. If you add features that make repetitive tasks easier, it often hides the fact that the underlying system isn’t doing its job properly, and stops the real fix getting done.
One of the “fixes” / improvements coming in the next release is the Trust Factor system having its trust values massively reduced. Like, current iOS versions are being absurdly over confident about reported location data accuracy, frequently reporting things like 10 metre accuracy when the location is 100+ metres from the real location. That’s the root cause of excessive nonsense moving items in the timeline view. So to cut down on that noise, I’ve made the recording engine far less trusting of iOS’s reported accuracies (for locations where there’s Trust Factor data, ie drifting data has been marked as stationary).
I don’t think that fix is adequate to solve the problem, but it’s an improvement.
That’s true, but imo only if the feature purely serves to deal with something broken. But in this case, bulk edit has real use cases, not just turning something into stationary. Car → taxi, metro → train, running → scooter, tram → biking etc. Pretty common to have multiple segments (correctly placed between visits) to be of wrong type.
Heh. In that case I’d still argue that shows something broken. If the classifier is being that indecisive/incorrect about the activity types along the trip then it’s not doing its job well enough 
Though I get your point.
That kind of mess is something I tend to only see happen on “first time” routes, like when it’s the first time I’m in a new city or neighbourhood. In that case we can’t reasonably expect the classifier to get it all right, but ideally it wouldn’t create a messy series of items that worsen the cleanup experience either.
But yeah, this is where I often get to: I don’t think things are working as well as they should, and as such I want to fix the underlying system instead of building a new UI feature to work around that weakness, but also don’t have a good answer for how to improve the underlying system. A bit of a deadlock.
I guess what I’d want to do is look at what the classifier ended up doing, identify the parts that made the timeline messiest, then look for ways to improve specifically those bits. Otherwise we’re accepting that the mess the classifier made is unavoidable and we should just work around it ourselves in the UI. I’d like to set higher standards, even if I can’t always achieve them 
Ah, I think I see what you’re saying. As in, if I travel a route and Arc says “metro”, and I correct it to “train”, next time it should remember that?
I guess that would be ideal. I get that more features is also more opportunities for bugs, and gotta maintain that etc.
That’s definitely how it works! Every confirmation or correction you make is put into the Machine Learning model, and increases the chance the classifier will pick the correct type for that route next time.
The problem is when it fails to learn, making the same repeated mistakes. In those cases there needs to be improvements to the system.
If it makes a mistake once, that’s expected. If it makes the mistake two or three times, it might just be something that’s more complex to learn, which is especially the case in built up city areas. If it continues to make the mistake much longer than that, then… yeah, improvements need to be made, and difficult problems solved in the underlying system.