Bill information verification

[EuroBillTracker]

For the past couple months, there has been a behind the scenes effort to crack the relationship between a bill's value, serial and shortcode. Thomas has helped us with much of the analysis.

Some time ago I shared with him some code that implemented the rules he (mostly) had infered and I finally put them onto the site last week.

Right now, it just logs unexpected entries, for example

Code: Select all

2003-09-13 21:28:46 : [User XXXX,V05311870681,L013F1,10] (i) Got printer L. Should be printers [M] or other value

. This is because the rules are not 100% shard and some values are even missing. I am hoping to infer the missing values from this.

We could also decide to make the system use them in their current state and just incurre the cost of making some people unhappy because they cannot enter a bill for a day. I'm not too happy about that solution but I'm open for discussion. Here are the number of bills flagged as invalid by the system by day vs the number of bills that day :
sept-8: 52 vs 4389 = 1.2%
sept-9 : 42 vs 3609 = 1.2%
sept-10: 41 vs 3389 = 1.2%
sept-11: 86 vs 3710 = 2.3%
sept-12: 96 vs 5204 = 1.8%
sept-13: 120 vs 4888 = 2.45%

On another note, it's a great way to detect serialists. I believe we may have identified one from this tool. An email has been sent to him and I am waiting for his explanation

[Thomas B]

On another note, it's a great way to detect serialists. I believe we may have identified one from this tool. An email has been sent to him and I am waiting for his explanation

Let me guess: __/-^-\__ This user?

But how do you want to check all those 'wrong' bills? There are a lot of them every day! If you have to check them one by one by yourself, it's a crazy job!

Offcourse you can make logfiles to track the serialists and see later what to do with the wrong entry's...

[EuroBillTracker]

Let me guess: __/-^-\__ This user?

How did you come to that conclusion ? Did you look for that bill ?!

But how do you want to check all those 'wrong' bills? There are a lot of them every day! If you have to check them one by one by yourself, it's a crazy job!
Offcourse you can make logfiles to track the serialists and see later what to do with the wrong entry's...

Well once we have the correct heuristics in place, the site will simply not accept the bill as genuin, just like when you make a mistake in the serial today.

[thomas]

Right now, it just logs unexpected entries ... This is because the rules are not 100% shard and some values are even missing ... We could also decide to make the system use them in their current state and just incurre the cost of making some people unhappy because they cannot enter a bill for a day. I'm not too happy about that solution but I'm open for discussion.

What I do on my own site is to reject a note if it does not comply to the heuristical rules, but an additional submit-button is presented which the user can press if she is sure that the data are correct (i.e. if my heuristics are wrong for this note), so that the note is inserted anyway. But I do not give suggestions about how e.g. a "correct" printer code should look like.

BTW, since I introduced this feature, about 15.300 notes have been inserted without error, 500 have been rejected by the heuristics (a typo error rate of some 3%), and 40 have been "forced" into the database with the above mentioned button (which means that my rules are a little bit too strict).

[EuroBillTracker]

Right now, it just logs unexpected entries ... This is because the rules are not 100% shard and some values are even missing ... We could also decide to make the system use them in their current state and just incurre the cost of making some people unhappy because they cannot enter a bill for a day. I'm not too happy about that solution but I'm open for discussion.

What I do on my own site is to reject a note if it does not comply to the heuristical rules, but an additional submit-button is presented which the user can press if she is sure that the data are correct (i.e. if my heuristics are wrong for this note), so that the note is inserted anyway. But I do not give suggestions about how e.g. a "correct" printer code should look like.

BTW, since I introduced this feature, about 15.300 notes have been inserted without error, 500 have been rejected by the heuristics (a typo error rate of some 3%), and 40 have been "forced" into the database with the above mentioned button (which means that my rules are a little bit too strict).

Using those same heuristics (actually yours are probably a little bit more up to date since we haven't synched in a while), we get an error rate (=potential reject) of about 1 to 2 %

[thomas]

Using those same heuristics (actually yours are probably a little bit more up to date since we haven't synched in a while), ...

not really; what I sent you the other day was the latest code currently used.

we get an error rate (=potential reject) of about 1 to 2 %

OK, in the above mentioned 500 errors I included trivial ones like wrong check sum, non-existing NCB-letter etc. The relevent part of the heuristical errors is, if you know what I mean (I don't have the list of error codes at hand :-)

Code: Select all

error  reject  forced
200        11       4
250         4       2
300        36       2
320        18       4
325       112      27
350         5       3

adding up to 186 rejected notes, of which 42 were forced in anyway. The other 300+ rejected notes cannot be forced in.

[EuroBillTracker]

Using those same heuristics (actually yours are probably a little bit more up to date since we haven't synched in a while), ...

not really; what I sent you the other day was the latest code currently used.

Well, actually I haven't used your code but mine from a couple months ago because it seemed somewhat more complete (it could help me determine the correct missing r values for finish notes for example).
It's really a shame that we did it two totally different ways, it's a lot harder to upgrade the code !

we get an error rate (=potential reject) of about 1 to 2 %

OK, in the above mentioned 500 errors I included trivial ones like wrong check sum, non-existing NCB-letter etc. The relevent part of the heuristical errors is, if you know what I mean (I don't have the list of error codes at hand

Code: Select all

error  reject  forced
200        11       4
250         4       2
300        36       2
320        18       4
325       112      27
350         5       3

adding up to 186 rejected notes, of which 42 were forced in anyway. The other 300+ rejected notes cannot be forced in.

The error rate I am talking about is the error rate with my code : this includes ANY error besides incorrect serial number vs its checksum (so it does include incorrect printer code)...