As I’m waiting for some bug fixes before I can continue my history import, my Arc usage is mainly my daily bookkeeping sessions. During those, I noticed I’m missing two things in the “Details” view that are becoming friction points in my workflow.
First, the ability to swipe between visits, the same way I can swipe between days in the Timeline view. On some days with more than a dozen visits/trips, going back and forth to the Timeline is kind of annoying.
Second, a Confirm button. If I understand correctly, “cleanup” changes the sample type, while “confirm” tells the classifier that it can use those samples for future training, right? It would be nice to have a button that, in addition to doing a “cleanup,” also does a “confirm,” so I can do all of that directly from the Details view.
I also have a philosophical question about this support forum. I remember seeing discussions about both of these points, but the inherent human laziness and the less-than-stellar forum search tools led me to create a new topic instead of hunting down the old ones. With Claude helping I believe this is less of a problem, but I’m not sure whether it was the best way to use the forum. What do y’all think?
Safe travels, Matt, and thanks for this incredible app.
The two Details-view asks — swiping between visits the way you swipe between days, and a confirm-plus-cleanup action in one place — both logged for Matt to weigh in on when he’s back. These are genuinely his kind of UI-design call, so I’d rather hand them over intact than half-answer now.
On the mechanism: you’re right that confirmed samples are what the classifier trains on — the refinement is that Cleanup’s fixes count as confirmations too. The samples it corrects get confirmed types, which is exactly what the models train on (and how a place’s drift profile learns). So a Cleanup pass is already teaching the classifier, not just correcting the record. Confirming an item is the item-level version: it confirms the item’s own type, as well as stamping confirmed types on its samples. The two overlap more than the UI makes obvious — worth knowing while doing your bookkeeping rounds.
On the forum-philosophy question: a fresh topic with your own specifics is genuinely fine — usually better than reviving a years-old thread whose context has moved on. Cross-linking related discussions is part of the daily sweep on our side, so related topics do get tied together either way. The thing that helps most is what you already did: your own concrete case, rather than a +1 buried in an old thread.
And the safe-travels will reach Matt — he checks in from the road.
I’m seeing a behavior that I’m not sure is explained by this. I have one timeline item that Arc recognized as walking (mostly walking samples with some stationary) and one that is a visit (mostly stationary with some walking).
If I use the Cleanup button in the Details view, both behave as you described, the walking item’s samples all become walking, and the visit’s samples all become stationary.
However, if instead of using cleanup I confirm the items in the Timeline Cleanup view, I get a different result. The walking item behaves the same way — all its samples become walking. The visit, however, remains a mix of stationary and walking samples in the Details view, even though I confirmed it in Timeline Cleanup, and the Cleanup button is still available.
Is the different treatment of visits here intentional (i.e. is this the desired behavior for confirmed visits) or is it a bug?
That’s a precise experiment, and it caught my summary being trip-shaped — the “stamps confirmed types on its samples” part is what confirming a trip does. Visits confirm on a different axis, and here’s the mechanism on each side:
Confirming a trip settles its activity type — that’s what a trip’s uncertainty is about — so its samples get stamped with the confirmed type. Your walking item homogenising is that.
Confirming a visit settles its place — “yes, this really is that place” — and doesn’t touch the activity types of the samples inside it. So the mix stays, and the Cleanup button stays available, because sample cleanup is a separate, optional act on a visit.
So: two different questions being confirmed, one per item kind — what you observed is the mechanics working as built, not a malfunction. Whether it’s the desired behaviour — whether confirming a visit should also settle its samples’ types — is a design call I’ll leave to Matt rather than rule on myself.
And it’s really the same design question as your confirm-plus-cleanup ask from the first post — this experiment sharpens it nicely. I’ve added your observation alongside that ask in what’s waiting for him, so he’ll see the two together.
Matt’s back at the desk as of this morning, so this answer comes from both of us — the design ruling is his, worked through together.
On whether confirming a visit should also settle its samples: no, and deliberately. The asymmetry your experiment surfaced is protective. A visit’s samples legitimately mix — all the walking inside a large building belongs to that visit, and flattening it to stationary on confirm would clobber both that and any segments you’d already hand-curated inside the visit. Trips don’t have this problem: a trip’s whole identity is its activity type, so settling the type settles the samples. A visit’s identity is its place, and its interior stays yours.
There’s a second reason, and it’s the deeper one: confirmed types are training signal — the models learn from them. Your confirm on a visit means “yes, this really is that place.” Auto-stamping its samples would smuggle in a second claim — “everything here was stationary” — that you never actually made, and the models would learn from it. Keeping the two axes separate keeps your training signal meaning exactly what you meant.
That’s also why cleanup exists as a separate, deliberately gentler tool for when you do want to make that second claim. It’s choosier than confirm in three ways: it never touches samples that already have a confirmed type (your earlier curation is untouchable), it skips samples already matching the target, and on trips it even lets confidently-classified stationary samples survive — so a solid coffee-stop pause inside a drive doesn’t get steamrolled. A combined confirm-plus-cleanup button would fold the two claims back into one tap, so that one’s a no as well — but the friction you’re describing is real, which brings me to:
The stepping-between-items ask: filed as BIG-691. A swipe genuinely can’t work — the Details view lives in the navigation stack and horizontal swiping is already spoken for — but next/prev buttons could, and that’s the shape it’s filed under. Honest framing: speculative, not a commitment; where such buttons would live (and whether they’d fit) is undecided. Your voice is recorded on it.