01-16-2026, 06:16 AM (This post was last modified: 01-16-2026, 06:19 AM by Tjerk.)
Again, it looks really good. I can not really help you with clean or fast laps, but this looks like a great addition already.
I am starting to either fantasize or hallucinate (pick one that fits best) but I can now even envision a screen with the LP file seen from "the side" so to speak, where you could draw a speed line (a bit as if you were drawing a curved line like in SGE, to create the elevation of the track). So draw a line on the map for the actual racing line, and then 'draw' the speed on that other window. I think that would allow someone to create AI without ever needing to drive a lap.
Did you make more progress on the track map front? I'm not sure if the STP file format is already out there somwhere, but I could definitely look into it. The files seem to hold not a lot more than dimensions and color indices, but I have not yet done extensive testing as - like I said - I'm not sure if it is needed.
(01-16-2026, 06:16 AM)Tjerk Wrote: Again, it looks really good. I can not really help you with clean or fast laps, but this looks like a great addition already.
I am starting to either fantasize or hallucinate (pick one that fits best) but I can now even envision a screen with the LP file seen from "the side" so to speak, where you could draw a speed line (a bit as if you were drawing a curved line like in SGE, to create the elevation of the track). So draw a line on the map for the actual racing line, and then 'draw' the speed on that other window. I think that would allow someone to create AI without ever needing to drive a lap.
Did you make more progress on the track map front? I'm not sure if the STP file format is already out there somwhere, but I could definitely look into it. The files seem to hold not a lot more than dimensions and color indices, but I have not yet done extensive testing as - like I said - I'm not sure if it is needed.
I was just thinking the same - to make a speed graph. I was thinking even to incorporate some feature where the user can tell the app to start at a particular DLONG at some mph, and automatically generate a speed curve all the way to an ending DLONG. I believe the speed vs distance graph is concave, with it flattening towards the upper end.
I believe the STP file is in a run-length encoding format - similar to what I describe in https://www.icr2.net/forum/showthread.ph...6#pid13356 but I have not tested this. If you would like to try and confirm how the file works, that would be great.
Preview of speed curve editor. You can edit the speed precisely in the table, or use the WASD keys to select LP records and increase/decrease speeds in increments of 1 mph.
(01-16-2026, 05:41 PM)Tjerk Wrote: Heck yes, that is so awesome! AI help or not, you're really very good at this.
Thanks, I have no idea how real software companies operate, but with this AI (I am using OpenAI's Codex), I feel like I am leading a team of programmers here. It does seem to take a lot of attention to detail and clear communication to get good results. And so much testing and feedback to iron out the bugs and make the product enjoyable to use - definitely not just telling it "give me a track editor" and hoping for the best. But it's been a lot of fun to see what the technology can do these days, and I feel that when the human still puts in the care and effort, that's how you can get some genuinely great stuff from the use of AI (instead of "AI slop").
My workflow also includes periodically showing ChatGPT a dump of my entire codebase and asking it to review it, and often it will suggest me to refactor here and there. So then I tell Codex to do that, and I imagine keeping the code clean and organized has been helpful to the AI.
01-16-2026, 07:10 PM (This post was last modified: 01-16-2026, 07:13 PM by Tjerk.)
(01-16-2026, 06:38 PM)checkpoint10 Wrote:
(01-16-2026, 05:41 PM)Tjerk Wrote: Heck yes, that is so awesome! AI help or not, you're really very good at this.
Thanks, I have no idea how real software companies operate, but with this AI (I am using OpenAI's Codex), I feel like I am leading a team of programmers here. It does seem to take a lot of attention to detail and clear communication to get good results. And so much testing and feedback to iron out the bugs and make the product enjoyable to use - definitely not just telling it "give me a track editor" and hoping for the best. But it's been a lot of fun to see what the technology can do these days, and I feel that when the human still puts in the care and effort, that's how you can get some genuinely great stuff from the use of AI (instead of "AI slop").
My workflow also includes periodically showing ChatGPT a dump of my entire codebase and asking it to review it, and often it will suggest me to refactor here and there. So then I tell Codex to do that, and I imagine keeping the code clean and organized has been helpful to the AI.
I have started using it in much the same way. I haven been tasked with a non-standard Oauth2 implementation using JSON web tokens and validation of certificate chains with all types of mumbo jumbo in a specific environment. I had no experience with that, which is usually not much of a problem for me, but this time I did not even have any idea where to start. What does not help in that respect is that everything related to crypto appaears to have been made opaque on purpose by the people designing the standards.
I have been using Google's AI-mode to guide me through the process and that has been tremendously helpful. It can very clearly explain what is relevant to my needs, what certain terms mean and which parts will be relevant to my project. I never share my own code because that would definitely pose a security risk, but I do not really need to either. I think it spit out a total of maybe 20 lines of code for me as examples, over the course of this whole workweek. But that gave me a much better grasp of what I was dealing with. It made me more confident about this project and saved a lot of time for sure.
If you understand the limitations of an LLM then it can be very useful already. I do not care much for the slop being put out, but I also find it rather disengenious when people try to attack LLM's by treating them as actually intelligent beings of some sort. Of course that will fly off the rails sooner rather than later.
A hammer is not going to build a house, but it sure as hell will be a tremendously helpful tool if you (know how to) treat it as such.
Includes various updates and fixes, most notably adding support for importing replays and editing LP speed.
Unless anyone finds a significant bug with this version, I am going to shift my attention to "updating" an existing track - maybe something simple like TheStig88's Firebird - to see how my tools hold up through a more end-to-end process. And possibly make some other helper tools along the way.
Per my last post, I thought I was going to do some actual track modding, but I could not resist the urge to continue on with the SG Viewer and now I have one app that does both SGE and SFE editing.
What you can see in the screenshot below:
- You can add, delete and split the centerline
- Fsects preview on the track map
- Fsects diagram where you do the dragging of starting and ending DLATs. You can pan and zoom, and it will also "snap" to the fsect at the next section when you bring the endpoint close to it on the screen (so for very precise snapping, you can zoom in).
- Xsect editing similar to SGE. Grade is a slider instead of the awkward click and side drag in SGE.
- A function to copy all of the elevations and grade from the currently selected Xsect over to another Xsect. This is so that if you did all this work on the elevations and grade on one Xsect you don't have to do it again on the next Xsect. In Papy's SGE, when you change the grade in one Xsect it doesn't automatically change it on the next Xsect, leading to uneven elevations.
