[BPIP] Bitcointalk Public Information Project [Back in Action]

[marlboroza]

<0 == Red

I spotted "light green" message on one profile:

...but account has 2 negatives, 1 positive feedback and big red warning:


Score is -1 and this should be red, right?

[Timelord2067]

Score is -1 and this should be red, right?

Wrong.

2 X 2 = 4

10 - 4 = 6 (assuming the one positive has matured into a value of ten) (now watch suchmoon and ibminer pretend I didn't answer the question...)

[suchmoon]

Score is -1 and this should be red, right?

Yes. Score should be actually -4 I think. The scoring algorithm tries to simulate the old trust score (greens count 10 each, when fully "aged" over 10 months; reds count exponentially as degrees of 2) but unfortunately it doesn't implement the reset that the old system had - when you get the first red rating, it should reset your score to 0. This should really be fixed one way or another... possibly made more consistent with the new trust system and account for flags too, which it currently doesn't.

[LoyceV]

10 - 4 = 6

Score should be actually -4 I think.

You're both wrong This is the profile: TheCarm (Trust: +1 / =0 / -2). It would have been -3 before theymos' changed the trust ratings. Like this rating: -3: -2 / +1.

See the algorithm:

- The first number is the trust score.
- The second number is the number of unique users who have given that person negative feedback.
- The third number is the number of unique users who have given that person positive feedback.
~

I also completely changed the trust score algorithm to this:

Code:

if there are no negative ratings
	score = 0
	for each rating, oldest to newest
		if this rater has already been counted
			continue
		score += min(10, round_up(months since rating))
else
	score = unique_positive - 2^(unique_negative)
	if score >= 0
		start_time = time of first negative
		score = unique_positive since start_time - unique_negative since start_time
		if(score < 0)
			return ??? (orange)
			
move score to range [-9999,9999]
return score

but unfortunately it doesn't implement the reset that the old system had - when you get the first red rating, it should reset your score to 0.

That wasn't part of the previous algorithm. Checking a few weeks of OgNasty's old Trust scores shows this.

This should really be fixed one way or another... possibly made more consistent with the new trust system and account for flags too, which it currently doesn't.

Why hold on to the old algorithm anyway?

[suchmoon]

but unfortunately it doesn't implement the reset that the old system had - when you get the first red rating, it should reset your score to 0.

That wasn't part of the previous algorithm. Checking a few weeks of OgNasty's old Trust scores shows this.

With +55 he would have a score of 550 minus 2^negatives if there was no reset. The reset has two steps. First, positives no longer count as 10, but rather as 1 each, once you have at least one red rating: "score = unique_positive - 2^(unique_negative)". Second step happens if this "score" is positive - then the start time of the calculation is changed to the time of the first negative (i.e. positive trust from before doesn't count anymore - that's what I meant by a reset to 0) and red trust no longer counts exponentially, so both count equally.

So basically this means at that point in time Og had 19 more green than red ratings since the reset.

But you're right about the TheCarm's rating, it would be -1 because it doesn't meet the conditions for the reset-prior-positive-ratings-to-zero part.

Why hold on to the old algorithm anyway?

Because there is no new algorithm, there are no scores anymore at all. And we need some number to represent a trust score if we want to have a ranked list. We haven't been able to come up with a new way to calculate a score that would account for flags for example. So the old one - unfortunately not implemented exactly by the formula - is sticking around.

Edit: grammar.

[Timelord2067]

I was wondering, when are the ten points counted as being added to the score?  Is it month to month e.g. positive feedback is given on the 8th of the month, so the first point is added on that day then each of the next nine months on the 8th of the next month?

Or, is it the first point is added when the positive feedback is given, then a fraction of the year (one tenth of a year??) for nine times until the score is ten?

[Vod]

Or, is it the first point is added when the positive feedback is given, then a fraction of the year (one tenth of a year??) for nine times until the score is ten?

Each individual feedback gains one point per month until it reaches 10.

[GazetaBitcoin]

I just found a weird behavior of the site, which seems to not parse "50 merits bombs".

At first, since Thursday the 12th I noticed a delay of updating my merits on the website. MoparMiningLLC sent me a merit and the site updated it to my profile on Friday evening (if I remember correctly; it could have been though Saturday morning). This may not look unusual, but so far I could see my new earned merits within a few minutes on the site.

After that, also during Thursday, I earned some more merits, including a 50 merits transaction from theymos (this being the last merits transaction on my account on Thursday). The other merits were received on Friday and Saturday: 1 from philipma, 2 from nullius, 4 from Ddmr and 1 from Pacimaker.

Again with delay, these last merits were updated on my BPIP profile, but the ones from theymos still don't appear, although they were earned before. I'm thinking if there is an issue with counting 50 merits transaction by the website

[LoyceV]

I'm thinking if there is an issue with counting 50 merits transaction by the website

My guess is it's because Investigations is not scraped by BPIP. I've sent you 4 more Merit as a test (your post deserves it anyway).

[GazetaBitcoin]

My guess is it's because Investigations is not scraped by BPIP. I've sent you 4 more Merit as a test

Uhm... I can say the test succeeded, as those 4 merits don't appear as well on BPIP. So your assumption is correct. But in this case, is this the designed behavior of BPIP or is it a bug?

For example, coinlocket$ also earned merits in this board, but his merits from Investigation board are shown on BPIP. That topic made by him is from 2019 though, so it seems like after later website update those merits were not counted anymore.

your post deserves it anyway

Hehe, thanks! I'm not sure, but I think you merited the wrong post. I believe you wanted to merit my update from November 12th, as you already merited the OP in the past (Unless you wanted to extra-merit a previous merited post - no matter which is true, I thank you!)

[Csmiami]

Uhm... I can say the test succeeded, as those 4 merits don't appear as well on BPIP. So your assumption is correct.

Not really.... the merits from suchmoon, nullius, Ddmr, and the old 8 ones from Loyce do show up. The board is only shown as (Deleted/Off-limits/Ignored); but when you click them, they take you to the actual post.

The odd behaviour is that it's not counting the new ones; not that it doesn't count the ones on the board

[GazetaBitcoin]

Not really.... the merits from suchmoon, nullius, Ddmr, and the old 8 ones from Loyce do show up. The board is only shown as (Deleted/Off-limits/Ignored); but when you click them, they take you to the actual post.

The odd behaviour is that it's not counting the new ones; not that it doesn't count the ones on the board

I didn't realize the merits from Ddmr and nullius were counted (these being sent also to my post from Investigation). So it's correct. Maybe the 4 merits from LoyceV are too new and the site didn't index them yet - I have to check that later. Then the problem is indeed with the 50 merits transaction though.

So it's not like the site doesn't count the new merits from Investigation board - it does. The merits from Ddmr and nullius came after those from theymos and they are counted. Also the old merits I received for that topic were counted in the past (similar to those of coinlocket$).

[hosseinimr93]

------

Since Investigations board isn't visible to guests, I guess that some merits given in this board aren't shown because BPIP has scrapped some merits without logging in and missed them.

[suchmoon]

Since Investigations board isn't visible to guests, I guess that some merits given in this board aren't shown because BPIP has scrapped some merits without logging in and missed them.

We're using this page: https://bitcointalk.org/index.php?action=merit;stats=recent

It requires a login so the scraper is always logged in. It doesn't have access to the Investigations board so the thread title appears as "deleted". LoyceV's four new merits are showing up. The only reason why the 50 wouldn't show up is if it didn't appear on the recent merit list. Unfortunately the list goes back only 24 hours and it's been 3 days since then so I can't verify it. I'll see if I can figure it out.

[GazetaBitcoin]

LoyceV's four new merits are showing up.

I confirm, the merits from LoyceV are shown, together with the other received recently. Only the 50 merits bomb is missing.

The only reason why the 50 wouldn't show up is if it didn't appear on the recent merit list. Unfortunately the list goes back only 24 hours and it's been 3 days since then so I can't verify it. I'll see if I can figure it out.

This is very curious... So the website shouldn't act like that indeed. Thank you for investigating, suchmoon.


Edit:  after some hours the information was updated on the site. Did you find the problem, suchmoon? Or did the site make the update with big delay, for an unknown reason...?

[Vod]

https://bpip.org/TrustLog

Seems to be a lot of changes recently.

Do you guys by chance have a graph showing number of changes per period?

[LoyceV]

I know BPIP uses https://loyce.club/trust/latestversion.txt to get https://loyce.club/trust/2020-09-26_Sat_05.08h/user_ids.txt to update it's weekly links, but that doesn't work for users who had a custom Trust list before, but don't have one now.
It also leads to incorrect links on users such as Furfuraceous: the Trust list link gives a 404, because it should link to last week's Trust list.

By my math, BPIP should add or correct the Trust list link on 542 profiles.
To fix this: I made a list of the last known version of each user's Trust list: https://loyce.club/trust/latesttrustlist.txt (currently 476KB).
This file gets updated right after latestversion.txt, and should be done after a minute delay.

Bumping my slightly edited quote for attention
Betwong noticed it too.

[Timelord2067]

Or, is it the first point is added when the positive feedback is given, then a fraction of the year (one tenth of a year??) for nine times until the score is ten?

Each individual feedback gains one point per month until it reaches 10.

Hi and thanks for your reply - so my score should be 74 not 73 at this point in time?



There are seven other positive trust feedbacks (more than 12 months old) - adding @Yoshi's score (one in August and three months thereafter) makes four - yes?

YOSHIE

2020-08-29

@Timelord2067, have good ideas/suggestions, on handling Alt cheating.

[Vod]

Hi and thanks for your reply - so my score should be 74 not 73 at this point in time?

You also have one negative.   Not sure how that affects the score anymore, but maybe the BPIP team could add a breakdown.

Example:

Seven positive over ten months = 70
Two positive for two months = 4
Five negatives = 2^5 = -32
Total:  42

[Timelord2067]

Hi and thanks for your reply - so my score should be 74 not 73 at this point in time?

You also have one negative.   Not sure how that affects the score anymore, but maybe the BPIP team could add a breakdown.

Example:

Seven positive over ten months = 70
Two positive for two months = 4
Five negatives = 2^5 = -32
Total:  42

Hi Vod,

My question was posted pre the most recent DT lottery.

What I've been trying to nut out is is it a calendar month e.g. the 17th day of the month, a 28 day month, or something akin to (365.25/12) days as the value only seems to change when the DT lottery occurs. (after the initial one point when the trust merit is given out for a positive)

At one point my score should have been advancing roughly every two weeks, but wasn't.

You also have one negative.

Again, that one negative comes and goes - pre this month's DT lottery it wasn't there (or, rather, the user wasn't DT (1or2) positive.  There's a popular misconception that if a DT (1or2) has a negative after their number, then they somehow aren't anything.  This is not the case.