Consider triangulating with other magnetic points
- There is always some margin of error, play with it.
- Mention the magnetic poles, a large deposit of iron, or some other object of nature that you have established.
- Triangulate with a. the initial crash site, b. the space station or moon in orbit, c. this other magnetic source (in 2).
- This would need some level of auto-computer driven calibration.
How to make it work:
What space ship doesn't have an astrophysics array these days? Starships had those before they had support for portable AI. Your space station probably has several young AIs in testing—it's run by the AI that runs other AIs!
Retrofit the onboard astrophysics array to detect magnetic metalics. It won't be perfect, but the cluster of iron deposits discovered to the southeast are large enough that the modified array should detect it anyway. You'll need to use a soldering gun to make the changes to the astrophysics array's integrated circuit (hardware mod) and—given the recent accident that caused all this (ahem)—your orbit isn't as stable as you'd like, and you'll only get one shot at this. You'll also need to replace the astrophysics processor with the orbital processor so that it can integrate the magnetic telemetry into your make-shift guidance system. That should be easy enough, if only the lock holding the orbital processor in place isn't welded from overheating...
You'll also need to copy a few lines of code (software mod) from your payload balancer mass detector recalibration subroutine* so that the AI's software knows how to calculate the telemetry from the magnetic mass. You'll have to take the AI offline for that, so it won't be able to help you through this whole process, but it will be thankful for the upgrade—because your plot is interesting enough to have an AI that you'll be able to take from the crash site with you (nods) because it is "portable" from one computer to another, and this AI of AIs has just been upgraded to calculate a position based on metalic telemetry.
*(This 'subroutine' is the small computer program/process which re-calibrates the mass detectors on spacecraft while docked at your space station thingy, those mass detectors help the propulsion systems balance the payload of their launches, of course. Every spacecraft has to have one of those, obviously, because, as everyone already knows, this is the age where what you're trying to do is even possible, wink-nod.)
The whole process should take four hours, but you only have 90 minutes. Even then, once you make reentry, the system will have a margin of error of about 20 meters (probably based on a degree trajectory margin of +/- 3°) because it must be crude enough that you could do the soldering by hand. And, if you get one of the wires crossed in your soldering (which you should do in all your haste, wink) you could end up flying right into the other crash site (or the lake 10 meters from it). This is because the improperly wired integrated circuit would make the AI confuse a guidance node (first crash site) with the landing destination.
(In a GPS, three satellites in view gives a position, a fourth allows to calculate elevation.) You'll also need to leave behind a drone to remain in orbit so you can get your fourth nav point so the AI can also calculate elevation to properly slow your descent; if the orbiting drone stops pinging (which it does, hopefully) you'd lose ability to make the automatic adjustments for your descent. In other words, if Murphy's Law holds true, you're going to end up flying in too fast, right for the other ship, needing to steer and break manually. But, not to worry, the grateful AI will be there encouraging you the whole time.
There's your ticking clock, which could last half a chapter or half the book, also some option exhaustion once the time runs out, and you have a few disasters to open up new opportunities, skew the reader's foresight, and make the situation more complex with a nice mix of victories and challenges along the way. 90 minutes is more than enough time to replace a CPU, copy-paste a few lines of code, do some bumpy soldering work, retrofit and launch a nav drone so it can break, all without the help of the offline AI.