Posts are hidden after a single flag

No.

In most cases, this is a single user messaging to say "I’m not sure if this message deserves a flag or not, but based on my interpretation of [the post | the mod policy], you might want to look into it.

Less novice users might do this as a PM to a specific moderator (or less likely to @moderators), but the most common form of such comments is a “something else” flag.

3 Likes

I see.

I don’t know that I agree with this, though. If it bothers a person enough to send a “something else”, that is indeed a flag. Granted the weight of it could be shifted down (or even up), but clearly something is amiss.

I’d support reducing weight on that particular flag though if you think it is warranted, because its meaning is kinda nebulous.

2 Likes

Thankfully this already exists!

1 Like

Oh, so you can adjust that particular “something else” flag weight? Do we offer that as a setting @eviltrout, the base flag weight for each flag type? What I’m getting at is that a site owner could decide “all something else flags confer no flag weight” and change the base weight of that flag to zero?

Right now it looks like “something else” and “inappropriate” have the same base flag weight?

I tested on try.discourse.org and I see 1.0 (flag) + 5.0 (trust level bonus) = 6.0 for both of those flags in the review queue, and the required score to hide a post is 4.0. So right now “something else” carries the same flag weight as “inappropriate” which doesn’t seem correct to me?

I don’t follow this statement, because ignore is already retained for all flags in my testing, above?

So if you want to give a particular flag zero weight… just press the ignore button? Am I missing something, here?

2 Likes

Yes - the setting is under ”settings” in the review queue. It only allows for low/medium/high, however.

1 Like

Aha, everything defaults to “low” so presumably 1.0 is the floor. Should we offer an “off” or “none” in this list @eviltrout?

3 Likes

Using Ignore directly reduced a TL3 member’s weighting, I have observed this and tested it by repeatedly Ignoring, not Disagreeing, flags raised by a specific TL3 member over a period of time until I got them down to 51%.

Would a re-test with a created account be helpful? I am willing to run that test and post exact metrics with screenshots, assuming that will replicate it.

If that means it will not reduce the flag-caster’s weighting nor add 1 to the Flagged Posts > max of 5 under Requirements for Trust Level 3 then yes, giving a flag zero weighting would be the hoped-for outcome.

2 Likes

Oh, I don’t think that should be happening at all. Ignore, to me, means “pretend this never happened”. I’m sure @eviltrout can factor that in next week.

5 Likes

The issue here is ”ignoring” a flag still adds weight to a post. If the plan is to have ignore really mean ignore, then it’s important that ignored flags not count towards flag weight on a post as well.

I suppose this would also change the user flag ”agreed” indicator as well, since that also takes ignored flags into account?

4 Likes

Okay here’s my action list for this week:

  1. Make sure ignore does not affect the user’s accuracy.

  2. Remove special case where TL4 people immediately hide non-TL4 user content with one flag.

  3. Prevent non TL4 from instantly hiding content with one flag, regardless of score:

    I will likely be adjusting the hidden thresholds. Right now they’re based on scores being above a certain percentile, which I think is far too low in the case of hiding. I will be basing my calculations on a default that’s closer to 3 flags per post and go from there.

    It’s quite likely I will add back a setting for min flaggers to hide in the meantime. I believe this will work for 90% of the complaints here. Scores are still useful for sorting the most egregious stuff to the top so it should work

6 Likes

Here are the first two fixes:

4 Likes

Actually following some discussion internally I’ve reverted the TL3 -> TL0 spam thing. We consider that exception important enough to keep. The TL4 exception is still removed:

3 Likes

Okay I’ve made a few more adjustments here. The first two are fixes for when forums are quite new and don’t haven enough data for good calculations. I realized this was a problem while browsing the source:

This one is more relevant to the feedback in this topic:

The calculations for priorities has now changed to be centered around reviewables that have at least 2 scores (flags). After upgrading to this version, it should push the thresholds for hiding content higher. A “medium” sensitivity should be roughly two flags.

Please try it out and provide feedback as to whether things have improved!

8 Likes

@ubik once you have pulled latest, try it out for a while, and then write a long reply about your new and improved experience :wink:

We will mostly be away for a week so this will be the last change in this area for a fair bit, but we believe these are all significant improvements to the existing flagging system.

1 Like

I am not seeing any change related to min flaggers to hide? Am I missing it or did you leave it out?
For me that was the most important one, if you do not plan to make other changes for a while, I will most likely be forced to revert back to an older version of the source.

Regarding
FIX: Ignored flags should not count in your accuracy score
Remove special cases for flagging
FIX: Put back the TL3 -> TL0 spam thing

If I understand correctly these changes will only effect TL3 users and above, we have a lot of issues with TL0 and TL1 users so they will not help us.

Regarding
FIX: Sensitivity did not work by default
FIX: Require a min amount of reviewables before calculating thresholds
Tweak calculation for reviewable sensitivities/priorities

These changes should help a bit initially, but seems like they can deteriorate after time?
The base sensitivity bump will be affected by users over time somehow, unclear exacly how.
The target_count seems to delay the buildup of accuracy since fewer posts are part of the calculations, if I understand what it does. But it will also exclude single flag posts from it, so for some users it may increase the buildup?

Can you explain if I am missing something, or just reading the changes wrong?

I see that self.target_count is defined in app/jobs/scheduled/reviewable_priorities.rb and set to 2. Would locally changing that and rebuilding give me 3 flags minimum before something is hidden? - Edit - I think this was wishful thinking when reading a bit more, but I’ll let the question stand anyhow. I currently interpret it as minumum number of flags needed for a reviewed item to affect user accuracy?

You do not understand correctly.

FIX: Ignored flags should not count in your accuracy score
True, this will affect TL below 3, my bad. We dont expose the ignore flag

Remove special cases for flagging and FIX: Put back the TL3 -> TL0 spam thing is only TL3 and up though?

Am I still not reading the first 3 fixes correctly?

1 Like

Do the new calculations allow a single spam flag to silence a new user and hide all of their posts? It seems a TL2 user would be able to do this with the silence_new_user_sensitivity setting at high, which is the default. Previously it took 3 different users casting flags to silence a new user. Does this setting still apply and how does it interact with the sensitivity calculation? Which one takes priority?