Live CEOing Ep 750: Language Design Review of GeometricSolveValues and GeometricScene

All right. Hello everyone. Welcome to another version 14.0 of language design Review, part of our intense season of these things. This one is about synthetic geometry. So geometric solve values. Wait a minute, we haven't. We don't have a geometric solve, do we?

No, there's a geometric solve resource function. The issue with geometric solve is if you. At least. This may not be quite as much of an issue now as it was when I made the resource function was if you're. We're solving for certain, say like area of some abstract polygon, you at least used to get a message if that was if you know, the whatever region should fix it.

I mean, I understand the left hand side of the thing.

Right. It does actually seem that that no longer really issues quite as many messages as I recall. So for the. For the resource function, we just inactivated all those heads, which we thought would be sort of awkward for a kernel function, but I could go through and check.

Let's just look at this. So this is pretty cool actually, what this is doing. So this says solve for a particular angle, for example, given this. And there are multiple solutions. Okay, so this is basically you're giving a set of constraints which are like equations and you're saying solve for things. Why didn't we have this before? This is a really cool idea.

We had it as a resource function, but it wasn't quite powerful enough to be a system function yet.

Okay, so if I were to say. I see, so this is asserting that those make a triangle.

Right.

So. Wait a minute, why are you saying instance one? Oh, I see. That's the specific.

Right, so it. The output is dependent on. Yeah, right. So if you. Yeah. So instance one just takes the numerical values for that scene. Instance for scene one. It's the abstract scene.

But I don't understand because these are determined in this case. So this is.

Right. Yeah, so they are.

So let me have an underdetermined thing. So.

You can scroll down. I have. So that's. The next example will be what you want.

Or I want to do my own. So I just want to say triangle abc.

You're the list around Triangle.

I don't think I should need it. You shouldn't insist on that. I should just be able to assert a triangle. You should make sure it doesn't need that list.

Yeah.

Well, what do you hope to solve for. For that? Nothing. Nothing. Watch what I'm going to do. Okay, there's my triangle. Now what I want to say is geometric solver values. Geometric. And now I want to say.

Right.

So this is.

So if you wanted to find the abstract version you do geometric cell values right. Of that SC.

And it will hopefully tell me. Okay, now I'm totally confused by what that did.

So. So the. It uses index for point coordinates.

We need to com. We need the coordinates of the abstract points.

I mean. Okay, was thinking it would just come back and say Euclidean distance of ab because it doesn't have any way to say what it is.

No, but it cannot say you gladian distance AB because the Euglidean distance wants something with the right form.

If What? Sorry, if. So if I now say.

I see what you mean.

Right. I mean it could just come back and say just like solve would say that in now here. If I say here something like how do I say. How do I say equilateral?

You would do for a triangle would be or so geometric assertion polygon string equilateral.

Yeah. Come on. So close. You got zero at the end. But why all the crud?

We. There'll be some bug bug fix time after the freeze.

All right. Okay, well bugs in a few seconds.

It is overall more optimized to vaults concrete exact numeric Q and approximate solution. Not for symbolic ones because this soon gets quite out of hand.

I understand.

Right. But I think there should be a quiet there somewhere in case something like that happens.

Okay, it's cool. Now why is this in one level of braces, not two levels of braces?

That's for.

Because it's a singleton.

Right. Well, so solve values is different from solve in that regard. So we're mimicking solve values. I had the same question.

I see, but so if there were two possible cases here. How can I make two possible cases here? Is easy way to generate geometric figure. There are two possible solutions. I mean we had examples of that.

Whether did you have unit documentation or not?

I for. Yeah. What is documentation? There's one in the documentation.

Okay. For what geometric solve values? Yes.

Basically again mimics what Solvay use does. So they are just also single list as now list with two elements.

It's I think bottom.

By the way, you shouldn't commit things to this documentation that are full screened like that. It will cause trouble.

Oh well, that's my bad. I always forget to. Okay, so scroll down near the end. I think. Further. Basically all the way to the bottom.

I think.

I know, I'm just reading these ones. I really like this. This is really useful. I mean this is something where you know this is very practical thing of being able to actually solve for these things. I Mean, it's one thing to generate these pictures which, you know, kind of Euclid re rerun, but this is actually something where, you know, given a practical, you know, engineering type problem, being able to conveniently represent geometric constraints and then solve for them seems really useful. Okay, right, there we go.

So the. Or no, no, the. The last one. The under possible issues. So here is one where.

Yeah, I get it. I could. Okay, this is the case where. Why is that a flat list?

Because it's. If you have multiple solutions but you're only solving for a single variable, the.

Second argument doesn't have a list.

Yes.

Okay, to be clear, I'm not, I don't necessarily agree with that, but that is the precedent.

Okay.

There are cases where we can't get a real nice closed form, right?

So especially here it's not an invariant of the scene. So, you know.

But wait, wait, I'm complaining about the inactive there.

Right?

So why isn't it just plainer angle? I mean, if I just type in planar angle, nothing bad is going to happen. It's just going to sit there.

No, something bad used to happen.

Well then fix it now because it isn't happening now and it better not.

I will, I will take away the inactivates. The only issue would be, I have to think if there's a possibility where you could somehow be solving for something that would evaluate, not issue a message, but actually evaluate. Well, like if you have Euclidean distance and you actually list abstract coordinates, then that will actually evaluate and we would have to.

So what. I mean, then it should evaluate.

I was going to say kind of depend.

I mean, it doesn't have to say symbolically Euclidean distance. It can give the X square plus Y squared type thing.

Oh yeah, yeah, that's fine. Never mind. I was thinking of what geometric solve would do. That would be an issue for geometric solve. Exactly. Yeah, you're right, you're right, you're right.

Which is a good reason why geometric solve might not be useful.

I mean this was one of our reasons why we have this.

I mean that was basically the reason.

Yeah, well, I think you're probably right. I mean, I think if you know what you're solving for. I mean the fact that solve, you know, the fact that solve can say X error, whatever, Y arrow, whatever is useful in the way that solve is used because you would be saying evaluate X squared plus Y cubed. With these values, I think it's much less likely that you're going to be doing. Well, I don't know, not Completely unlikely you're going to be doing that here. But that pipeline is a little bit less likely here.

I agree that the other one.

Less likely. What's that? I agree that it's much less likely.

I mean. Right.

Substituting into a geometric scene is. Yeah, it's not a common thing. I mean, you know, being able to say, because once we got this geometric scene, if we substitute planar angle is 30 degrees. It's kind of a goofball thing here. Yep, exactly. This is actually really quite spectacular functionality. I'm really quite pleased. I mean, what. So. I mean what happens if. Okay, where are its boundaries? What, What? Where does it start to break down?

Well, So basically we complicated enough scenes, you're going to almost you, you, you. There's a good chance you're going to end up with one of those conditional expressions as your output. If you know, because there are limitations for explicitly solving for stuff.

There are various boundaries. One obvious one is as long as you go straightforward forwards, meaning you have a construction typically you never leave basically nested square root territory. But if you have say a tessellation of a triangle into smaller triangle, you have a circular loop. So you can get much higher degree polynomials and we will not be able to do them right now.

Right, yeah. So basically for the, you know, for the, for the finding the exact values like it's decent with quadratic irrationals and if you're looking for an angle, a radian angle, you know, rational fractions of PI, but some exact values are unlikely to be found.

The other boundaries is if you have sufficiently many points on a line that must come in order because then you need inequalities to represent it and then you are in cat territory with double exponential complexity.

Right. So basically the fail safe for everything is one of those conditional expressions where it's basically the thing you're solving for and then.

A bunch of equations.

Yeah, but basically the equations that were used by random that would have been used by random instance to draw the scene which you know, because this is.

The case for symbolic things for the numerical case can you blast through and when you have specific numerical values for things.

Right. So if you are talking about running geometric solve values on a fully specified geometric scene, so the one that actually has the picture drawn, then it's just a straightforward rule replacement.

Well, you might have to compute something depending on which quantity is asked for.

But well, yeah, well, but yes, you're.

Basically substituting the coordinates into a formula and out pops an answer.

Yeah, right. So that in general, that is what.

The Heck is this?

So that's.

If you do random instance, you'll see the scene. It's just an example where you don't have instructions. You have mutual dependency.

Right. I think there's a geometric scene.

There's a P4 sitting alone in the middle.

Yeah. There's a P3. There's some commas and stuff.

There's a multiplication P3 time style.

Yeah, I must have copied wrong.

And there's a P2. Is. There were. There are a few other.

It's in the neat examples on the geometric scene ref page.

Yeah. This one might take a while. I think this one takes a while to evaluate.

It's a complicated one. It's not the limit. What we can do.

Yeah, this is. This. This could take a minute or two to evaluate.

Okay, well, let's go on. I mean, this is quite lovely. Yeah, I'm.

And also. Yeah, send me this notebook afterwards so I can find a. That reduce message leak. I thought I had all of those holes plugged.

I'm sorry, what did you say?

I said, go ahead, send me this notebook.

Not somebody else will deal with that. This is not for me. Who's project managing here? Dorothy.

Dorothy will take care.

I will. Okay. So I should give up on this, probably.

Yeah, it'll.

Yeah.

There we go. Well, that one was not a great. This is one where. Yeah, sometimes.

That was uninspiring.

Yeah. Yeah. I would.

Would dragging a point help?

No, this time probably not.

No.

I think that's too hot. Yeah, it was.

Now it's spinning right.

It's trying to. It'll eventually time out from trying to drag the point.

Can't you detect this degeneracy?

We. We do have some code for that does try to detect it. It doesn't. It doesn't detect everything. And sometimes it's hard to tell whether it's actually a bad scene or whatever we think is degeneracy is actually a facet of the scene. So basically we detect the degeneracy. Then we see if we can do better. And if we try but can't do better, then we just return what we originally had. Which. Yeah, some of the more complicated scenes, it's, you know, sometimes, you know, I.

Don'T understand why there isn't some way of keeping derivatives around so that you can tell what, whether you're in a stationary point or something. And when this isn't, then, you know, it's kind of a. A random degeneracy.

Well, right. Well, we, I mean, we, we. For this instance, we presume I think the code would detect that it's in a degeneracy. The issue is trying to, trying to find, find something that's not in the. That's not degenerate. And we have a time, we have a time constraint on that. And so for something like this, where, you know, it's a long complication, no matter what the. That, you know, we detect it's in a degeneracy, then we try to find something that's not degenerate, but it's, it's almost assuredly going to time out because this is a very difficult scene to find.

Yeah, okay, fine.

We made good progress over the last two years with degeneracy, but for hard ones, we still suffer from it.

Fair enough. Okay, let's look at this. What is this? Wait a minute, you just sent me another thing. I. Okay, remember what's happening is. Yeah, you're putting equal in there because the Unicode values were not pasting correctly. Okay, let's talk about this because we're going to run out of time. What is this?

So this is Jeremy Stratton. Smith made this, so I'll let him present it. Yep.

This is essentially an interface to allow you to build a geometric scene without having to type the code. So it's divvied up into the figures and relation sections. Figures.

This is the beginning of our effort to make an interactive way to enter geometric.

Yes, but at the same time a little bit orthogonal because this is. You have a pure text problem and you want to convert the text problem as convenient as possible into something, but not yet drawing it, but kind of more formalizing it.

And also remember to hit enter.

Yeah, I was about to say that.

Yeah.

So if you want to spe, you can specify whatever points you'd like in the input field. These will take you to links for the various documentation.

Okay, I have a couple of meta questions here. First question is, has this been seen by either UX or Design?

What you've done in a preliminary fashion, I think. Yes, it's been a while. It was published originally as a resource.

Function, so yeah, it was a WF off of a. Yeah.

Okay. Did they look at it then?

I think somebody gave some initials.

That's not good enough. I mean, okay, because I can see tons of things here which are not quite right. I mean, which are, you know, design UX type things independent of functionality. All right, so. Why is that plus not doing anything?

Because you have not specified fig. The primitives that you've specified are not properly.

Obviously you've got to. Okay, let. Let's talk about how this interface, basically. I mean, you know, in a situation like that, the. You. What am I trying to say? You know, and something like that. It's obviously got to have a mouse over that says, you know, you can't press this yet because you haven't specified the tech.

The grid underneath does say.

Does.

I'm aware of what the grid underneath says. I read it. But I'm also channeling what a typical user will do, which is not read it.

Yep.

Okay. Okay, so let's, let's start at the top here. So, okay, first of all, I wouldn't do it like that. I would do a question mark thing there. I don't think you want to do this, but that's an admission of defeat to say you won't understand this. Go read the documentation. A nice little question mark. Everybody knows what it is. But it's not like you're, you know, it's not like one of those textbooks where it says you're not expected to understand this type thing. That makes sense. Yep.

I mean, keep in mind, it is literally the WFR which was made for the vain audit.

No, I, I understand, I understand. So what's your point? I mean, your point is that. That I shouldn't be so critical or your point is that.

No.

What?

That some of the classical form language design principles might not quite apply to this audience.

Okay, but. But insofar as we're putting it in the system. Yes, we're expecting.

Yes.

So as I understand this, if I start off here. Something like this and then I say constructor.

This does not yet work.

Okay, but that will work, is that right?

It is the goal to make this.

That's the goal.

Okay.

Basically, it. It should in some sense behave similar like molecule draw.

Okay.

Where you also can give a molecule to start with.

Except it doesn't do the modal dialogue thing.

Correct.

Do people know that this exists?

I think so. I think at least Jason Donneberg promoted pretty heavily at various talks and others.

Pretty nice. How do I. What does return mean?

I think you get it in the.

Notebook and so you do. I keep on forgetting that things like this exist. Wow, look at that guy. Okay, now the real question is whether I can solve the geometric constraint problem based on this in here. How would I go from talking about your geometry stuff here? I mean, how would I go? Well, you wouldn't.

Right?

Because our geometry is strictly planar geometry. Well, okay, fair enough. This isn't a plan of molecule. Sorry about that. Here's a question. Could I go from this, this thing here to a graph? Is there an easy way to do that or take one of these other things?

No, there's no easy way to do that.

And I don't know that that's even makes sense. I mean, because what if you have circle through. Yeah. Or. Fair enough. Okay. And what about the relationship between these geometric proofs and find equational proof type proofs. I'm just trying to understand the interoperability here.

Peter Barenzi has been working on the proverb that is more a discussion with respect to the prover, not draw.

Okay, so this is a separate thing. Okay. What do you expect? I mean, the first question is, do we have any hope of, of NLU stuff even through an LLM being able to synthesize like if I, you know, have we tried to get LLMs to synthesize geometric scenes from descriptive text?

We've made some training sets for them. I'm not sure if.

Yeah, what about just trying them raw out of the box?

I had no success with it.

You had no success?

No, I think it has, I think it has seen too few pairings.

Okay, so we have to fine tune it at least with our form here.

Exactly.

But, but I mean, even if you ask it certain questions, could you, like, like here as I'm as a human trying to fill out this form, you know, you could imagine interrogating it. What primitives, what, you know, what figures are there in the scene, blah, blah, blah, what relations are there between these things? Does that make any sense? Probably might be worth trying because, I mean, the, the challenge here, by the way, it drives me crazy that those are not in a grid. But that's just me. Okay. So if I were to pick a different thing here, let's say a circle through what happens? I just hit return just to make it. Okay. And then it's, then it, it's, it generates.

You have to put in the points.

Right. These are just suggestions.

Like you need to put in the eight points for triangle.

Yep. I took it both with. With braces and without.

It's.

Cool. So I don't think those are quite the right words there. I mean, I think you want to make it clear that this is a random instance visualization.

Yeah, yeah. The real problem is, and we knew is we cannot use literally the WFR for basically this purpose and this audience. We have to use different words.

Yeah.

And I. It is now set up so we can, we can modify it for this however we want.

Really? Okay, so what do you what do you want to ask me here? Because, I mean, this needs a bunch of UX and design work and I, you know, I mean, we can set up, for example. Okay, let me. Let me start just giving you a few random things. Given that you have icons for these, you should use them. Do you understand what I'm saying here?

Yeah.

In the pull down. Okay. And it just helps people understand what. Primitives is a very complicated. It's a very weird word here. I mean, these are the.

We discussed for a long time. Yes.

What's that?

We discussed for a long time what stride world.

Objects. Probably.

Geometric objects. Geometric figures. I mean, I do point out that in.

I mean, for what it's worth, primitives.

Is exactly what we call it everywhere else in the documentation.

Yeah, well, fair enough, but.

Yeah, but it's. Does circle through actually count as a primitive? Because that. Because if you circle through, return a circle. I guess not. Because that evaluates.

Yeah.

So what about primitive objects? Yeah, it's not bad. I understand. I mean, here. I mean, this vestigial piece of, you know, heading is very, very confusing. Do you see what I'm saying?

I do. If you click the plus button.

No, I know. Then I'll get something. Yep. But you shouldn't leave it vestigially when there's nothing there. Well, this says select figures. What's figures? Oh, that's figures.

It probably has popped up on a different window somewhere.

No, no, no, no, no. Select figures means this. All these. I don't understand. This here is all figures. Right. So view documentation refers globally to the whole thing, which. But it's on. It's aligned with figures, which isn't. Right. Yeah, I'm, I'm brutal here. I'm sorry, I just. I've done this stuff for so many years. It's kind of like you've got to have that tucked in a corner. So. And then the figures make it clear that that's not documentation for the figures and clear all versus clear relations. No, you know, I wouldn't put the clear there. You know, in other words, clear is not the primary action here. Clear is an action after you, you know, have all these things set. You know, clear goes somewhere different. And you don't want to clear all when you have nothing there. See what I'm saying? Yes. There's nothing to clear if there's nothing there.

We'll have more things popping in and out of existence.

That's okay, so long as you keep the thing. I mean, it was also wiggling around as things came in and out, which isn't right either. And that's just a detail of implementation. Now, when it says name, what is the point there? That's the name for this.

Yes. In Geometric Scene, there's that. You have the capability of naming.

I got it, I got it. But you can't say name paren this because that is the name. See what I'm saying? There's no reason to say. If, if you want to say this thing is called Triangle pqr, then you want to put that somewhere else here. And its name, because that is the story of what you have here is. It is triangle pqr. Do you see what I'm saying? And that's the summary of this as well, has nothing to do with parenthesis in the name, because it suggests it is the name. Do you get what I'm saying?

I do.

I told you, I'm. I'm totally brutal. I've just. This is what I get for having done this stuff for 40 years. It's, it's like.

Trust me, Stephen, I've seen you in brutal mode. You're not in brutal mode.

Okay, the. Okay, but, but look here, I think the name should be above or to the side or something. Name underneath is a very weird construct.

Agreed. It, it will then shift the elements around once you have.

No, you can't do that. You can't do that. You gotta, you gotta device where the thing. Also, these large fonts don't make any sense. Just don't make any sense in this case. Look, see if you can get.

We should have a meeting with Noah.

And, and Jeremy Davis and, and this just doesn't, I mean, it's, you know, in general terms, this is great, but let's get the details right. Yeah, right.

We knew you would say all of this.

Right.

You can just have the LLM version of me and it's like, like, forget the meeting, go ahead. Can I make one suggestion? Yeah. I think there's just too much going on in one interface here. Maybe that just like, just the triangle thing can be one sort of slide and multiple of, you know. Well, there's definitely too much going on. I'm not sure that's the solution. Like, think about if there's like 10 or 20 of these things, right? That that's going to be too tall for your screen. Like, is there going to be a scroll bar? Okay. The other possibility, arnode, is that all that's here is the summary Triangle pqr. And then whatever is the case, not all the choosers that the choosers go on a modal pop up, like hit the plus button like 10 times. I understand what you're saying. So I'm saying that the thing you want to do is to list the, the actual geometric objects here. Boom, boom, boom, down the page, right, with a plus button. And when you say add a new one, when you click one and you want to edit it, it brings up a modal. Yeah, it opens up. Yeah. You know, thing. Because depending on what it is, you know, it's a circle, it's a this, it's a triangle, it's a whatever. Then you give yourself a lot of freedom. If it's, you know, if you're putting in a, A, you know, a pentagon with all kinds of different tentacles on it, then you can use the whole screen to, you know, you can use a whole region to deal with. I mean, I'm not sure you want it. You know, you can also open it up in pain here. Rather than having to pop it up. It would be better both for the cloud and for this to not have to do a pop up. But you see what I'm saying? So that, so that what is listed here, all the this, all the crud is gone from here. And it's literally a list of the figures, the relations. Boom, boom, boom. Vertically. It should just be a one line.

Per thing, like the name.

You could have a little scrolling thing that, a left, you know, a fixed size, you know, box with, with elision at the end or scrollability that lists what properties it has. Yeah. Does this make sense to people?

Yep. I think the, this, we, we can sort out the specifics.

Yeah.

But it's, you know, this to me. Okay, so let's just talk about workflow for a second for what a user of this is going to do. So the user is reading a description. I don't know whether anybody reads. I mean, you know, does every version of this actually come with a diagram or does a user read a description?

There's a fair number of word problems in textbook that have no diagrams.

Okay. And probably there are also patents that have no diagrams that just describe in their incredibly obscure way the, you know, some geometrical thing. But. Okay, I'm just, I'm still just kind of a little bit, you know, as a practical matter. How do people, I mean. Well, it's going to be much better when we can draw stuff because.

Yeah.

And then what's the interaction between this and drawing things? Are you going to be able to flip between this and a drawing interface?

I think that I don't Think we did deeply think about this, but might.

Be a possibility because it seems like there's always a backup. I mean, whatever you have in the drawing interface, there's always a version of it, which is this essentially textual form. I mean, you've got really three forms here. You've got the actual linguist, you know, language form, computational language form. You've got this essentially editable form of the computational language form, even though this is not exactly horrifying to edit, but this is perhaps more convenient to edit and then you've got the actual visual form. See what I'm saying?

Yeah.

To me, I'd add there's also a textual form. Put a WFR in chat.

Oh, wow. We use this thing.

We use this in the geometric scene formatting for conclusions and things and step labels.

I think there's also been user requests for this to make it into the kernel. Right?

Yes. There is at least one external report.

For this, but this is kind of read only. We don't expect to be able to parse this.

Right. This is kind of the inverse of, you know, this takes symbolic and produces natural mathematical language rather than.

I know. The question is, could we parse this one day? No, I don't think it's impossible.

Unclear.

I mean, this is not unpausable. We could pause this into that fragment. You know, in other words, our lesson from control equals. Is fragment pausing is still useful. So I'm saying. Yeah, if I say, for example, okay. Before I wanted, you know, I could say a regular pentagon. With center O and side length two, for example. Okay. I'd love to get something which is. I just want code out of that. So I'm saying.

Right.

So it would be very convenient.

And will you.

What happens if you hit enter?

Well, that wasn't so bad. I mean, that's not terrible. But I mean, it's also not quite what you said.

I mean, it, I assume, gave you the correct polygon that you wanted, but it's not in the same form as how you specified it. You didn't say anything about square roots and things.

No, but I said it's side length two. Yeah.

You'd like a symbolic representation that actually just says side length two somehow.

Well, only if I can make graphics out of it.

Well, if this is going to work with a.

Right, but this is going to be with the geometric scene language, we.

I know, but it's also got to work with graphics.

Synthetic descriptions of that.

But it's also got to work with graphics, both of them. So we. Whatever. Regular polygon Description we have here, you know, sure. Better be renderable in graphics. So I mean, what I'm thinking here is what we want then is four tabs, basically. And this is similar to what we have for the Question framework where you have a visual way of entering it, you know, a form like way of entering it, a computational language way of entering it, et cetera. So you've got these different tabs and you can go between them. And if you edit the computational language and go back to another tab, this, this kind of form type thing will change and the visual thing will change.

But it's a little bit of longer term goal. That is nothing that can be quickly achieved.

I agree. I'll be surprised if this is doable for 14 hours.

No, no way.

Right. So. Okay, hope this. Yeah, so I mean, I, I think what I'm saying is not that unachievable. I think what we want to do is this and this form are both enterable. Okay. And right now the visual form and this textual form are viewable but not enterable. Am I making sense? But the organization is. These are just tabs. You're looking at the same thing in these different forms and you can edit it in 2 out of 4 of the forms right now and over time you might be able to edit it in more forms. Does that make sense? Okay, I need to go.

So I think conceptually, yes, we have to discuss and think through some of the details.

All right. This is going to be very nice and you've done most of the heavy lifting here. So now let's polish it.

I think for 14, let's basically make improved visual version of this one and then the other forms.

Post 14. I don't think this is going to make sense as a kernel thing in 14, unless you think there's a strong reason for that. I think it should be left.

The other issue for this in 14 is there's the dynamic module issue, which is out of our hands.

Fine. I don't think. Yeah, I just don't think it's going to make it. I don't think it's going to make it at a quality level that we're interested in.

Okay.

I think this way of accessing it is kind of janky. You know, I think if it works properly, we need a geometric drawer or some such other thing or geometric. You know, I don't know what you want, but this is not the right way to access something like this. I mean, it's reminds me of canvases and things as well. There's more work here. This is a nice thing, but there's more work.

Okay.

Geometric solve values is great and looks like that's a finished thing, so. Okay. What's that?

Just said okay.

All right. Yeah. Our node is pointing out. It'd be nice to have name fields automatically filled out with short, sensible names. That's a good idea as well. All right. To be continued. It's really nice to see this geometry stuff going along. All right, see you guys soon. Bye. Bye.