A new project is coming to the end of its crowd-sourcing phase which could be interesting. It's a bit of a man-in-the-middle processor with memory to be between your computer and printer. If the print job fails, it can resume at the point of your choosing.
Sales promises offered to the crowd:
"Currently Resumer 3D could not work together with Octprint, however, our team is testing and updating the firmware, once it's finished, they can work together."
"I am sorry that currently resumer3d can not work with octoprint now."
"Resumer 3d will work with Octoprint, my team is upgrading the firmware. Once it's finished. it will work."
They indicate that the molds are in production and they'll be fulfilling some of their kickstarters' outstanding orders.
The thing that bugs me with them is that they appear to be in China (their contact info uses WhatsApp with a Chinese mobile number), but claim in their kickstarter campaign to be in CA. Do you know them at all, perchance? They claim to be "Mark Cole" from Chino Hills, but use broken English and all their hackaday posts are from "robert". I'm always a little skeptical of KickStarters that don't give any actual background for their team and have incongruities with their posts. They also waited until backers were freaking out about it being July 31st to tell them their products would ship the first week of August and arrive in late August (which also implies China, not CA).
Good stalking, there. It gives the appearance of a legitimate project.
And yet, it could be all con job.
My own interest would be to divert one branch of my man-in-the-middle project to provide this functionality. On the surface, it sounds like something that could be accomplished with a Raspberry Pi Zero (1.3) plus the add-on UART hat, plus a button or three.
LOL It's not really stalking when it's public information provided under the guise of transparency, though, is it? It's just scrolling down on their own links and actually reading what they say (and looking at who is giving "official" responses). KickStarter added the creator verification in response to the issue of vaporware, IIRC. It was really their odd phrasing with a CA location, their odd statements about when backers would get products, and the whole "I promise OctoPrint will work by the time you find me and chase me down" routine that made me curious enough to scroll to the bottom of the KickStarter page. Too many decades on the internet of chasing down shady, pseudo-anonymous people who think they're fully anonymous and no one will ever be able to identify them (for the purposes of OpSec and the occasional suicide intervention), I suppose.
I'm all for you adding it to the man-in-the-middle project...basically all these guys did was take Prusa's power failure module concept and stick it in a dongle, and then add some special sauce to try to figure out what the printer was doing by being a MITM. They basically decided they could either a) build a control unit that does what Prusa's does, or b) fake it at a much lower quality level and sell it to people who don't want to modify their printers. Prusa controllers can handle you moving the head to load filament, and actually know their position at the time of failure...any MITM concept is limited by not knowing how many lines are in the cache and machine coordinates, so they're always going to be guessing, though some will allow for better tuning than others; yours could theoretically allow for moving the head off the print to load filament without drooling all over the print, and then let the user reposition the head in the right spot via your project, which would then let you extrapolate which line of the last few lines it was most likely on and even modify it to resume from the position the user jogged the head back to. At that point you could even move to a safe position to preheat and prompt the user to wipe the drool and hit "GO". Of course, that would require 5 buttons and a small display (1" OLED, 20x4, or even just a 16x2).
I used to run a CNC machine. There were times when you'd have to hit the big red button, jog the Z up, remove some wood stock that's gone rogue, jog the Z back to where it was, rewind the GCODE a bit and resume.
It's obvious from their design that they're saving layer changes and the +/- is to allow you to jog through the layers; this would be the granularity offered. I really can't imagine a working product that doesn't have some sort of mechanism for talking back to the user.
Layer number is easy enough, but it'd be far more interesting to use the user specified position to guesstimate where in the layer it was (I can all but guarantee that 99% of the time, it will not fail during layer change, as they are a very small fraction of the print time). Given how small most printer buffers are, you could probably extrapolate the correct line 99% of the time and minimize/eliminate the overextrusion artifacts from redoing part of a layer. Just sayin', since you're already intercepting and inspecting the gcode and reconciling a position within say 8 lines of gcode is pretty simple math, given 3D printers and slicers generally use straight segments, not curves (pet peeve that curve operations are excluded).
Personally, I like the idea of finding the correct layer (where it crashed) and then slowly going through a hovered "dry run" with the user, letting them click a button to say "yeah, that's where it stopped", then rewind, heat up, add an extra G0 command and resume at the next G1, as indicated. In theory, this might be a perfect restart.