The Suggestions panel in the Identify interface is broken

I’ve been participating in the May 2021 Id-athon, and find that, when I open an observation in the “Identify” interface (not sure how to refer to this), the IDs suggested in the Suggestions tab are truly horrible, in general, compared to what a user would see in the usual observation interface. Because of this, I’ve had to open all of the observations I’ve reviewed in the regular interface.

Here’s an example of the difference…

In the Identify interface,

In the observaton interface,

1 Like

You have to change the “Source” to “Visually Similar” on the Suggestions tab of Identify.

Once you do that, they will still usually display different results though, at least in part because there isn’t currently a way to “Only view nearby suggestions” in the visually similar filter of the Suggestions tab on Identify.

1 Like

Adding to @bouteloua’s reply, the Suggestions tab can be helpful if you’ve already got a good idea of what something is. The Suggestions then list the species of that taxon seen in that area from most often to least often observed. That’s different from the visually similar matching of the iNat computer vision but still often quite helpful.


Ok. Why is that not the default? I didn’t have to change any setting to get good ID suggestions in the regular observations interface.

In short, I think this is broken, even if your suggestion fixes the problem.

I only started using the Identify interface because of the ID-athon. Having an interface like this is a great idea, but it needs to work at least as well as the usual interface.

1 Like

Perplexed. The Suggestions tab doesn’t have any idea how good my idea of what something is might be. It needs to give good suggestions, period.

If the observation is already identified to genus, it will by default show the most observed species in that genus in the region (super helpful, e.g. leading to my own recent feature request which illustrates an example).

1 Like

In the ID-athon, all observations I’m trying to ID are either “Unknown”, or “State of Matter Life” (which will probably be the subject of another bug report)

I’m also participating. :) You’re right that the Suggestions tab on Identify is less useful by default for those situations. Nothing has changed recently or is broken from what I can tell.

There’s an existing feature request to make these filters “sticky” where you could vote and/or weigh in with your thoughts:

I appreciate your responses. I’m a software developer myself, and understand the tension between “nothing is broken - it’s working as intended” and “why isn’t this doing what I expect it to do?”. But, even if it’s currently working as intended, it’s not working in a way that’s helpful to people trying to use the Identify interface to resolve “Unknown” observations. I’d like the developers to consider that.

I have my own workaround, which is to open every observation in the regular interface. This works very well, even if it’s a little bit less efficient. I just expect the Identify interface to work the same way.

1 Like

Another feature request could be something like “If observation is unidentified, default to “Visually similar” instead of “Most observed” on the Suggestions tab of Identify”

Are you suggesting that I register that as a feature request, or are you communicating to the developers through this chat? I think your suggestion is good, for what it’s worth - “most observed” could be birds, when the observation is of a plant, which is enough to make anyone think the system is broken.

1 Like

Either way works: we could turn this topic into a feature request instead or you could submit a new topic.

1 Like

“Either way” What are the two ways? How do I bubble this up to someone who can evaluate the problem and decide on a course of action?

Whether a fix for a problem is a “bug fix” or a “feature request” is an arbitrary distinction.

I hear your frustration, but note that I like the way it works at the moment. That’s because it forces identifiers on the Identify webpage to not just rush through accepting plausible visually similar suggestions. iNat’s Computer Vision can often be wrong, and we already have a big enough problem on iNat with many identifications being made by people just accepting what the Computer Vision recommends.

If you look at an Unknown or State of Matter Life observation and only know it’s a flowering plant, for example, just add your ID as flowering plant and move on. If you’re a botanist with the skills to assign it a genus, then you can add the genus as your ID and the Suggestions tab becomes a quick way of seeing what species in that genus have been seen in the area. That can be enough to jog your memory to assign the correct species ID.


I appreciate your positivity. Putting reins on people’s ability to jump on plausible, but not well-considered, IDs would make sense in the usual interface, but on the Identify interface, you should expect to be dealing with people who know what they’re doing. Most people never see this interface.

What I really don’t get is why the regular interface can produce good suggestions, without my having to tell it it’s a plant in the Ukraine, when the Identify interface just fumbles and offers the most common living things in the Ukraine? Does iNat really expect that identifiers need more guards than regular observers?

1 Like

Forgive me, but I can’t see where to set the “Source” on the “Identify” interface.

Identifiers are not all biologists working in their own niche. They may also be enthusiastic newbies with hopeful good intentions. Not yet understanding that even if iNat says, pretty sure it is X - that they need to look at that suggestion very carefully. Especially if it is NOT seen nearby.

I have dug down to the bottom of the barrel for Cape Town CNC unknowns. Now chewing thru the Needs ID backlog that has accumulated in the meantime.

1 Like

It’s the symbol just to the right of the word “Observations” in the “Suggestions.” If you click that, you can set to “visually similar” and it will stick for the following observations. There was a previous discussion on this topic, if you’re interested: