I am just confused with the flag score?

Why the flag score increase is non-linear after 500? I know there is a max cap score of 750.

Why don't just make it as linear and increase the max cap, say 2000?

  • 1
    500 was originally the cap, I can't recall why it was then changed but there's probably a meta post for it somewhere. It does, however, mean that the "high flaggers" have separate scores and so can be still be sorted. – DMA57361 Aug 9 '11 at 13:03
  • @DMA57361: I can't find the question, so i post it out ;p – Kit Ho Aug 9 '11 at 13:04
  • @DMA57361: I saw someone got 750, and that should be the max – Kit Ho Aug 9 '11 at 13:05
  • 750 takes a hell of a lot of flags to reach – DMA57361 Aug 9 '11 at 13:07
  • Related: meta.stackexchange.com/questions/78343/… – user162697 Aug 9 '11 at 13:07
  • @Siva: it is not really the same question what I am asking for. I just am curious on the non-linear increase rather that linear increase – Kit Ho Aug 9 '11 at 13:09
  • 1
    @Kit Ho: Yes, I am aware of that. I didn't use the word duplicate. I have only mentioned related. – user162697 Aug 9 '11 at 13:10
  • 2
    @DMA57361 meta.stackexchange.com/questions/97890/… covers how many flags it takes to reach 750 – Kate Gregory Aug 9 '11 at 13:21
  • possible duplicate of Unnecessary precision displayed for flag weight – Cody Gray Aug 9 '11 at 13:26

Lets start by looking at a graph of the flag weight for 6 imaginary individuals flagging 400 times at various success rates.

(because I don't know the actual algorithm used, I'm going to assume each successful post-flag above 500 will increase the flag weight by 10 * (750-weight) / 250. That should be close enough for our purposes.)

Flag weight over 400 flags.

As you can see, anyone with a success rate over 50% will eventually reach 500. Once our imaginary people pass 500, their flag weight starts to plateau at various levels depending on their success rates. If the cost-reward was symmetric then we everyone over 51% would eventually meet up again at the top.

Not only does this allow the system to distinguish between a "good" flagger and a "bad" flagger, but also between a "good" flagger and a "long term ok" flagger.

For simplicity, the graph assumes that people flag consistently with their good and bad flags evenly distributed. In reality, there will be some local variations and some people will change their ways (for better or worse). I found that the flag weight continues to (roughly) track the same path even with small local variations, but a sustained change will cause the flag weight to find a new plateau.

Flag weight variation Flag weight divergence.

Update As per Hendrik's recommendation, I revisited this using 10 * Pow(10, -(weight-500) * 0.008) which is reported to be the actual formula used.

More accurate algorithm

The overall result is mostly the same but with more differentiation in the upper regions.

Although its not shown in this graph, the 99% user will still hit the 750 cap in a little under 800 flags past the end of the graph, whereas the 98% user will not reach 716.3. However, I'm not sure such specific details are particularly meaningful in the real world ... except perhaps to determine the minimum number of consecutive successes required between each failure to guarantee some sort of "forward motion" (probably just short of 99).

  • It was around 780 flags to get to 750, with a pretty good accuracy (~1% declined), a lot of it depends on when you had the flag declined. – user7116 Jan 20 '12 at 18:25

The idea is that this way your flag weight will always approach a certain number depending on the current success rate of your flagging.

If you're 50% successful, your weight will approach 500. If you've been 100% successful lately it will approach 750. For any rate in between it will asymptotically approach a number between 500 and 750. If your recent success rate changes, it will start adjusting towards the number for your new success rate.

Brian's answer here has some great graphs that explain the concept really well. Have a look at it.

  • So this scoring system is to distinguish who is the "real" good flagger? – Kit Ho Aug 9 '11 at 13:27
  • @Kit Ho: The idea is that the flags get sorted by the flag weight of the person who made them. So the higher your flag weight, the more likely it is that your flags will be handled quickly as they will be near the top of the list. – hammar Aug 9 '11 at 13:30

Speculations about the formula are in What's the flag weight formula in the 500-750 range?. The rationale is basically "with greater power comes greater responsibility." Once you're established as a good flagger, another good flag doesn't provide much more information about how good a flagger you are.

Not the answer you're looking for? Browse other questions tagged .