We've all seen them, endless machine-meandering about something that went horribly wrong because someone tried to modify their game. Because they're a self-replicating problem, searching for them quite often leads folks here.

That's .. great? - we love new people, but we really can't help them with something that only the author of the code they just threw into the mix can really understand and debug. We try, we offer blind advice, but in the end most of these types of questions just end up being a mentally taxing experience for us and new users. We don't want to turn people away, but we just can't help here; we're better at helping people learn the mechanics of a game once it actually loads.

Several discussions have been raised about how to deal with this. We'd like to offer a solution that means less work for you, and a better experience for the folks that try to ask these questions.

What if, through the magic of advancements that Stack Exchange has made in bringing up just-in-time help where it's really needed based on what users want to post, we could simply stop accepting most crash reports and lead the author to something that will probably help them immediately? That sounds like a win, and this is what we want to try:

  • We work on matching most crash reports through the blacklist, but instead of just blocking them, we show folks a link to a canonical question explaining how to debug buggy mods in the game they're asking about. This would start with Minecraft, which is probably the biggest source of this.
  • You tell us what canonical "I added a mod, and now my game broke" post the just-in-time help message should point to. It really needs to end with "If that didn't work, you need to contact the mod author".
  • We put the blacklist entry into place knowing that it won't match every single crash report, but enough that we match most, and some where the author neglected to post the report at all.
  • We then tweak this over time to be more effective, curate the canonical post as things change, alter the help shown if needed, and see what happens.

Custom close reasons can then be honed to better handle the ones where matching really isn't possible.

While no one has really suggested an outright moratorium on anything modded, folks are getting frustrated enough that they're casting close votes on perfectly answerable questions that aren't crash reports, while telling users that help is only available for the vanilla version of the game. I'm not going to point out any specific instance, I'm active in a few tags here, and I've seen it. I think we can help get ahead of this, and eliminate a lot of the noise.

What say ye? I've talked to some of our core devs, and they're happy to plug at this for a while (it's a really interesting exercise in regex).

What we'd need from folks to get this started is the ID of the 'truly canonical' mother-of-all-things-you-can-do post about Minecraft crashing due to forge / mods, ending with you need to contact the mod author, we can't really help until your game actually loads.

  • We're set to go with this at some point this week. Let us know when you've got a canonical post (or I can even take a crack at writing one).
    – user43038
    Jun 15, 2015 at 13:24
  • 2
    I've no idea how you're planning on doing this. It sounds like a pretty good idea in principle though. However, you'll get false positives if you are using an automated approach. I'd go for automatically booting suspected crash report questions in to a moderation queue so that they are not globally visible until they have been moderated. I'd also warn the user that the post is felt to be a crash report and point them at the useful support and help they should use instead. Maybe that's what you're already doing?
    – Benjohn
    Jun 22, 2015 at 10:04
  • Another thought – I'd guess some kind of Bayesian approach would work well for detecting these posts. Better than regex. You've probably got a pretty good training set available now!
    – Benjohn
    Jun 22, 2015 at 10:08
  • 1
    @Benjohn A Bayesian approach is the eventual goal. Tag prediction is Bayesian, we're currently looking at ways to (ab)use it for several other things. Think [crash-report] or [!crash-report] - totally theoretical at this point.
    – user43038
    Jun 23, 2015 at 17:29
  • Tasty. Like it.
    – Benjohn
    Jun 23, 2015 at 21:01
  • 4
    What is/where can I find the current status of this?
    – Ben
    Jun 25, 2015 at 4:58


You must log in to answer this question.