Resolved != !stumped ?

July 7, 2010 9:10 AM

Should tagging an AskMe thread "resolved" automatically remove the "stumped" tag, if it's there?

Given that "resolved" and "stumped" are treated differently than other tags on AskMe, shouldn't marking a question as "resolved" automatically remove the "stumped" tag, if it exists on that question? I had assumed that it would, but it turns out that's not the case.

posted by DevilsAdvocate to Feature Requests at 9:10 AM (38 comments total)

I think your example is the exception rather than the norm, and it's probably just a mistake. Right now there is only one question tagged both stumped and resolved.
posted by pb (staff) at 9:13 AM on July 7, 2010

Well, it's not a mistake: it's my own question, and I had marked it "stumped" about a week after I asked it, and marked it "resolved" (expecting that would remove the "stumped" tag) just now after I posted my followup.

But yes, it is rare: is that because it very rarely happens that questions marked "stumped" are later resolved (in which case, "just ask the mods to remove it, it doesn't need to be specially coded" is fine with me, although it's disappointing that resolving a "stumped" question is so rare), or is it because it happens frequently, and everyone in the past already asked the mods to remove the "stumped" tag?
posted by DevilsAdvocate at 9:20 AM on July 7, 2010

Side pony: If someone does follow DTMFA advice, then I think it would be appropriate to add the dumped tag as well.
posted by special-k at 9:23 AM on July 7, 2010 [9 favorites]

It would be cool if the dumped tag was like the resolved tag and that once added you can never go back!
posted by cjorgensen at 9:29 AM on July 7, 2010 [1 favorite]

Going through my Contact form inbox, I see that we've had three or four requests this year to remove a stumped tag. There might be more if people are contacting one us directly. So yeah, I think it's a case where you can contact us if you spot it and we can remove the stumped tag.

I'm guessing not everyone knows about the resolved tag so not every stumped question that is eventually resolved is tagged that way. I don't think the lack of this problem is necessarily a sign that stumped questions remain stumped forever. Maybe some infodumpery could turn up some stats about the number of comments on stumped questions to see if that's the case.
posted by pb (staff) at 9:32 AM on July 7, 2010

Side pony: If someone does follow DTMFA advice, then I think it would be appropriate to add the dumped tag as well.

I smell a back-tagging project!
posted by reductiondesign at 9:33 AM on July 7, 2010

So yeah, I think it's a case where you can contact us if you spot it and we can remove the stumped tag.

Fair enough—consider this a request to remove the "stumped" tag from the question I linked.
posted by DevilsAdvocate at 9:49 AM on July 7, 2010

Done!
posted by pb (staff) at 9:52 AM on July 7, 2010

Thanks!
posted by DevilsAdvocate at 10:04 AM on July 7, 2010

looked up some quick stats: right now we have 153 questions tagged stumped. And here's how they break down by number of comments:

• 0 - 22
• 1 - 23
• 2 - 17
• 3 - 22
• 4 - 9
• 5 - 10
• 6 - 11
• 7 - 7
• 8 - 6
• 9 - 7
• ... [1 or 2 from 10 comments to 34]

This doesn't tell us anything about whether or not they're resolved, but it does show at least some activity for most stumped questions.
posted by pb (staff) at 10:17 AM on July 7, 2010

I'd be interested to know the comment counts with timestamps later than the stumped tag's timestamp. And maybe also a count of 'best answers' with such timestamps. If I was at home I'd be running the queries myself, but you guys always come up with these interesting Infodump puzzles while I'm at work!
posted by FishBike at 10:24 AM on July 7, 2010

There are also 10 questions tagged stumped that have at least one answer marked as best, so that might rule out the stumped tag for those at least.
posted by pb (staff) at 10:26 AM on July 7, 2010

I'd be interested to know the comment counts with timestamps later than the stumped tag's timestamp.

Unfortunately we don't directly track the time a tag was added. We have a log of tagging activity we can go to, but that data is mushy compared with tracking the time alongside the tag.
posted by pb (staff) at 10:30 AM on July 7, 2010

Darn, so, is the date stamp information on tags in the Infodump just an approximation, like the contact creation dates for early contact data is? I'd like to update the wiki to describe how this is generated, if anybody can describe the process by which this ends up in the Infodump.
posted by FishBike at 10:41 AM on July 7, 2010

Done!

Really? Still shows stumped and resolved for me.
posted by cjorgensen at 10:41 AM on July 7, 2010

Still shows stumped and resolved for me.

No stumped tag shows up for me, but I hadn't clicked on that link prior to the fix. Maybe it's cached?
posted by FishBike at 10:47 AM on July 7, 2010

The date stamp info stored with tags is the original link date. We're not normalizing (de-normalizing? unnormalizing?) that data in this case so we can have better performance on tag pages.

Yep, not seeing the stumped tag on that question anymore. Try a hard refresh on that page, you might have a locally cached copy.
posted by pb (staff) at 10:52 AM on July 7, 2010

Any question that I ask about cutting down a tree I will tag with "stumped".
posted by blue_beetle at 11:24 AM on July 7, 2010 [3 favorites]

Ooops, my bad. I looked at the stumped list and Devils Rancher has one on there that has a check next to it. I thought resolved got those? It's marked stumped and has a check.

I need to get some of that reading comprehension.
posted by cjorgensen at 11:42 AM on July 7, 2010

I really don't get why there is both best answer and resolved. I would never mark a best answer until the question is resolved, and vice versa.
posted by smackfu at 1:01 PM on July 7, 2010

Smackfu, sometimes it's nice to provide props for the best answer so far.

FWIW, I asked this same question last year and while it's never going to be worth pb's while to code, I still think it would be kinda cool if the tags were mutually exclusive.
posted by the latin mouse at 1:42 PM on July 7, 2010

Ah, I never do that, because I figure if there is a best answer marked, it discourages people from trying to answer.
posted by smackfu at 1:51 PM on July 7, 2010 [1 favorite]

I really don't get why there is both best answer and resolved. I would never mark a best answer until the question is resolved, and vice versa.

As the OP's own question which led to this MeTa shows, sometimes a question gets resolved by the asker. It would be a little weird to mark your own answer as a best answer (even if I have seen it done once or twice).

Also, as the name implies, it's a "best answer," not necessarily the right answer or the final answer. So you could very well have a question which doesn't really have an answer, yet you might want to mark as "best answer" the response which best explains why your question is unanswerable.
posted by Conrad Cornelius o'Donald o'Dell at 7:12 PM on July 7, 2010

The date stamp info stored with tags is the original link date.

Is it? I thought we attached a timestamp at time of tag addition to tags added after thread creation, but I haven't looked at this stuff since the last time I was adding tag stuff to the Infodump so I may be forgetting. I do recall finding some weirdness in how timestamps for some tags were handled where we were incorrectly backdating newer tags, but I thought I found that was only in the case of older stuff before some code changes.

I'll have to go take a look at some point.
posted by cortex (staff) at 8:19 PM on July 7, 2010

Is it?

Yep, just looked at the code to verify. That's not say we've always done it that way without fail for every tag-related thing we've done, but that's how it works right now.
posted by pb (staff) at 9:00 PM on July 7, 2010

Side pony: If someone does follow DTMFA advice, then I think it would be appropriate to add the dumped tag as well.

If someone follows a bungee jumping recommendation in a question about exciting holiday activities, I think it would be appropriate to add the jumped tag as well.
posted by EndsOfInvention at 2:52 AM on July 8, 2010

Yep, just looked at the code to verify. That's not say we've always done it that way without fail for every tag-related thing we've done, but that's how it works right now.

I poked around the tag data in the Infodump a bit this evening. Based on this conversation I was expecting to find that the link dates on every tag for a given post would all be the same. So I ran a count to see how many posts have tags with 2 or more distinct link dates. There are nearly 70,000 such posts in my copy of the Infodump.

I just took a quick look at one, post id 93188 on the blue, and there is one set of tags with the link date that exactly matches the post date, and another couple of tags with a link date about 45 minutes later. This post is from June 25th of this year, so pretty recent.

I'm wondering if there are 4 possible cases here?

1. Tags added by the original poster during the posting process
2. Tags added by the original poster after the post has gone live
3. Tags added by mutual contacts of the original poster
4. Tags added by the back-tagging crew

Any idea which of these use the post date and which use the current date, and have they always done that?
posted by FishBike at 5:10 PM on July 8, 2010

I checked the code for cases 1-3 the other day, and it's all the exact link date of the post. We haven't changed the code within the last couple of years that I remember.

I'll check into the back-tagging option, but that doesn't seem like a possibility for such a recent post. We have some admin tools that add tags and that could be off.
posted by pb (staff) at 8:02 AM on July 9, 2010

hmm, maybe we have some discrepancies between the live data and the infodump. I took a look at the tags for post id 93188 and here's what I see:

2010-06-25 11:48:38.113	messerchups
2010-06-25 11:48:38.113	russia
2010-06-25 11:48:38.113	scifi
2010-06-25 11:48:38.113	surfrock
2010-06-25 11:48:38.113	vincentprice
2010-06-25 11:48:38.113	CocktailDraculina
2010-06-25 11:48:38.113	munsters

Are you seeing different timestamps?
posted by pb (staff) at 8:56 AM on July 9, 2010

Ah ha! So! pb is right, and I'm not crazy, and here's why: when I wrote the infodump script that generates the tag data, I compensated for a lack of accurate creation-time records for after-the-fact tag additions by making a best guess about actual tag-addition times.

Which fact I apparently had not documented on the wiki. My bad. It's essentially the same deal as with the estimates for the Contacts timestamps, though, taking advantage of the fact that tag ids are assigned in order and so no tag can have been added earlier than the previous one. I talk a little about the ups and downs of this on the wiki page about the Infodump in the contacts data section; I'll go update that to take into account tags as well.

My comment in my code:

        # we provide approximate tag-addition dates by "guessing" that a new
        # tag was added at the same moment as (i.e. no earlier than) the 
        # previous tag was added.  It's an imperfect method, but the actual
        # add times are lost to time.

Question of why I tend to use the first-person plural "we" when writing comments in code only I will probably every look at left as an exercise to the reader.
posted by cortex (staff) at 9:19 AM on July 9, 2010

aha, just took a look at the infodump code. For the infodump we're estimating the time a tag was added based on the tag_id. So if a tag_id for a particular post appears out of sequence, we use the last available link_date for the previously added tag in the ID sequence as the link_date for that particular tag. It's a way of estimating approximately when the tag was added.
posted by pb (staff) at 9:20 AM on July 9, 2010

2010-07-09 09:19:32.113 jinx
posted by pb (staff) at 9:21 AM on July 9, 2010

Heh.
posted by cortex (staff) at 9:22 AM on July 9, 2010

Excellent, mystery solved.

So... if I'm thinking this through correctly, this algorithm of estimating link dates for tags added after the post was made¹ would tend to result in timestamps that are slightly earlier than when the tags were really added.

Which means using the Infodump link date to count answers and best answers that were posted after the "stumped" tag was applied would tend to over-estimate these results. But maybe only slightly so because of the frequency of new posts (and hence, new tags with accurate link dates) in AskMe.

All of which is a long, rambling way of saying I'll give that analysis a try now that I understand the limitations of its accuracy.

pb is right, and I'm not crazy, and here's why:

Well, pb is right, but the explanation provided is, I think, insufficient to prove sanity. In fact, some might say it leads one to the opposite conclusion.

1: I almost used the term "post-post" for this.
posted by FishBike at 9:51 AM on July 9, 2010

So... if I'm thinking this through correctly, this algorithm of estimating link dates for tags added after the post was made would tend to result in timestamps that are slightly earlier than when the tags were really added.

Yep. Where the value of "slightly" is the tricky bit: the approximation is always on the earliest terminus of a range of possible actual dates, and that range is the distance from the estimated link_date to the next-later link_date in the file.

So the size of the potential error will itself vary from link_date to link_date as well. I haven't looked at the actual data recently, but it might be interesting to give it a once-over to measure the delta between link_date values over time, see just how much give there is—whether it's typically minutes or hours or what. In this case, a fairly busy Ask Metafilter is helping us out a lot since every new post anywhere on the site is a guaranteed reset of the error clock and we get questions posted all day long at a decent pace.
posted by cortex (staff) at 10:09 AM on July 9, 2010

The short of it being that I'd guess the delta is relatively small most of the time—in minutes or maybe hours if it's the middle of the night server time—and so measure for post-"stumped" effects should be pretty doable especially given that people aren't going to be (I'd bet, but maybe look at this too!) stumping a thread that's still fresh enough to be getting any large number of answers incidentally still.

And, for that matter, you can also look explicitly at the error ranges of the stumped tags in question to have specific delta value on a per-stumping basis if you want to really get into it.
posted by cortex (staff) at 10:12 AM on July 9, 2010

[...] if you want to really get into it.

You know I do!

I did some analysis of answering activity after the stumped tag is applied to questions. First I computed the range of times when the stumped tag could have been applied. The range starts with the timestamp shown for the tag in the Infodump, and ends with the next later timestamp for any tag in the AskMe tag data.

I then counted answers, and answers flagged as "best", with timestamps later than when the stumped tag was applied to that question. I did this twice for each question, for the earliest and then the latest possible timestamp when the stumped tag was applied. This gives a maximum and minimum count of answers and best answers for each question.

Here are the numbers:

• 155 questions with the "stumped" tag.
• Only 41 to 48 of those received any answers after the "stumped" tag was applied.
• Only 3 to 4 of those answers were marked "best answer".
• Of the 41-48 stumped questions that got any subsequent answers at all, most got 1 or 2. A handful got 3 to 6. One got 34.
• The uncertainty of timestamps on "stumped" tags averages 34.7 minutes, and ranges from 37 seconds up to 152.8 minutes.

Is the "stumped" tag typically removed manually once the question is resolved? If it is, then it's impossible to tell from the Infodump which questions were once tagged with it but no longer are.

However, even if that is what happens, for the purposes of this discussion it doesn't really matter. We can still see all the questions where automated removal of the "stumped" tag might have had an effect, and there seem to be very, very few of those.
posted by FishBike at 1:11 PM on July 10, 2010 [1 favorite]

I just re-read pb's list of total answer counts and noted he saw one stumped question with 34 answers, the same as I got counting only post-stumping answers.

It's an anonymous question, and from the context I think it's really likely that one was posted with the stumped tag already present. Sort of pre-stumped as it were.
posted by FishBike at 2:41 PM on July 10, 2010