User talk:A930913/Archives/2015/Jan

Great catch, thanks!!

Great catch, the stray bracket in fn. 211 on the Mark Satin page. I'll never edit this late in the night again! - Babel41 (talk) 08:18, 1 January 2015 (UTC)

ReferenceBot appears to be down

ReferenceBot appears to be down. It does not show any edits since December 29. – Jonesey95 (talk) 04:44, 7 January 2015 (UTC)

DefconBot source

Hi! Mind if you send me a copy of DefconBot's source? (via git, sftp, email, etc) Cheers, --L235 (talk) Ping when replying 02:22, 8 January 2015 (UTC)

@L235: It's on the labs, /data/project/defconbot/ and follow run.sh for all its messiness. 930913 {{ping}} 06:50, 8 January 2015 (UTC)

Thanks, I've made a copy in my homedir to look at. Cheers, --L235 (talk) Ping when replying 20:44, 8 January 2015 (UTC)

DefconBot percentages

Hi There.

Do you have any comment on the changes to the reported percentages at User:DefconBot.

Yaris678 (talk) 10:25, 4 January 2015 (UTC)

It seems clear that the percentages ought to add up to 100%.

However, the previous version of the table didn't bear much relation to current reality either. Over a 96-hour period in mid-December, when the bot appeared to be working continuously, the time periods at each level were:

Defcon Level	CVS range	Hours at this level	Percentage of time
1	> 5.7	22	22.9%
2	4.9-5.7	19	19.8%
3	4.3-4.9	13	13.5%
4	3.0-4.3	29	30.2%
5	< 3.0	13	13.5%
Total	-	96	99.9%

So although apparently intended to hit Level 1 only 1% of the time, it was in fact at that level 23% of the time. Does the bot manager need to adjust the thresholds?: Noyster (talk), 13:22, 4 January 2015 (UTC)

Hi Noyster. That is an interesting result. I imagine that this would change quite a bit from month to month. Perhaps it would make sense to remove the table from User:DefconBot and add in a statement along the lines of "Each defcon level will be in use an approximately equal amount of time."

That was just a suggestion. 930913 may prefer to modify the code to achieve the desifed result, rather than modifying the description to fit the measured result.

Yaris678 (talk) 14:13, 4 January 2015 (UTC)

@Yaris678 and Noyster: The thresholds currently used, I recall, are from the summer holidays when it was quiet. The thresholds ought to be regularly updated, as Wikipedia declines, but I've been busy and ill of recent. I could set the thresholds back to the old levels, but I should really do the maths to work out new levels when I feel better. 930913 {{ping}} 19:29, 4 January 2015 (UTC)

Thanks 930913 and I hope you're on the mend. I've suggested here that a different vandalism alert could possibly be derived from STiki queues, and would welcome any input there: Noyster (talk), 19:07, 8 January 2015 (UTC)

More disagreement on DefconBot percentages

Update: the following have happened since the above discussion:

• I changed the description of DefconBot, summarising (in my opinion) the discussion here.
• Creeper 919 reverted to an old version, with the edit summary "better version,school holiday part is not proved".
• Creeper 919 left a message on my talk page saying "I think the old version was correct, can you not change it again?"

I am pretty sure the old version is not correct, but I'd like to open up the discussion here, in case I have missed something.

Here are the two versions, side by side.

A Text in version edited by Yaris678 at 09:32, 8 January 2015 Update, based on User talk:A930913#DefconBot percentages. This doesn't necessarily describe the desired situation, but it describes the situation as it is at the moment

B Text in version edited by Creeper919 at 02:54, 11 January 2015 better version,school holiday part is not proved

It uses ClueBot NG to estimate how much vandalism is being performed, and sets the defcon level based on thresholds calculated to give specified percentages of time at each defcon level in a sample period. The percentages are as shown in the following table. The sample period was during a school holiday and so in other periods the percentages will be different. In particular, Defcons 1 and 2 may be more frequently used in more busy periods. Defcon Level Percentage of time 1 1% 2 10% 3 20% 4 60% 5 9%

It uses ClueBot NG to estimate how much vandalism is being performed, and sets the defcon level based on the threshold calculated by the calculated percentage of time that each defcon level should be at. The current percentages are shown in the following table. Defcon Level Percentage of time 1 90% 2 60% 3 20% 4 10% 5 1%

As we can see:

• Version A gives the percentages stated in the request for approval and then states why these are not correct in many cases.
• The percentages in Version A add up to 100%.
• The percentages in Version B add up to 181%.

I have not seen any source for the numbers in version B or any explanation as to why the numbers should not add up to 100%.

I have a feeling that Creeper919 has misunderstood something but I don't know what. Maybe I am missing something. Maybe something needs to be clarified about what the percentages mean. I am open to suggestion.

Yaris678 (talk) 11:39, 12 January 2015 (UTC)

@Yaris678: I have recalculated the thresholds using the period from the beginning of the year and have reverted the DefconBot userpage back to the original. (If anyone can utilise 181% of time, please let me know how.) 930913 {{ping}} 12:39, 12 January 2015 (UTC)

Thanks. Yaris678 (talk) 13:13, 12 January 2015 (UTC)

In the real defcon, defcon 5 is the least important, and defcon 1 is the most. I see no reason why these should differ whenever defcon is used; if you disagree with me, then you can visit the defcon page. You may be able to adjust the digits so it equals 100%, but I still think that defcon 1 should have the most weight. They should also not have to add up to 100, because each defcon unit is independent, and does not rely on the previous.

Wikipedia vandalism information
(abuse log)

Level 4

Low to moderate level of vandalism

[view • purge • update]

---

3.90 RPM according to EnterpriseyBot 01:10, 12 December 2024 (UTC)

As you can see, defcon 4 says LOW to medium level. 60% is much more than medium, so this should be changed. Creeper919 (talk) 03:41, 18 January 2015 (UTC)

@Creeper919: The percentages are for the percentage of time. Using historical data, we can calculate the thresholds such that, for example, 60% of the time will be spent on defcon 4. Time should most definitely add up to 100%. 930913 {{ping}} 04:02, 18 January 2015 (UTC)

@Creeper919: It looks like you were interpreting the table to mean the levels relate to particular percentages of edits being vandalism. Am I right in thinking that? This is not what table is about. The levels are based on an estimate of the number of vandal edits per minute. This rate is converted to a level by referring to thresholds that were calculated to mean that 9% of the time the level will be level 5, etc. Yaris678 (talk) 17:32, 18 January 2015 (UTC)

DefconBot down

DefconBot hasn't edited in two days. I've also noticed that your tools on WMF Labs are down as well. Mind checking them out? --I am k6ka ^{Talk to me!} _{See what I have done} 11:42, 29 January 2015 (UTC)

@K6ka: Yeah, problems on the labs recently. Should be fixed now. Thanks for notifying. 930913 {{ping}} 23:46, 29 January 2015 (UTC)

Just a word

I have noticed the concerns you voiced about David Mellor's article in my personal messages section. I am a Wikipedian, but didn't bother logging on as I usually would. The only reason I deleted one of the brackets around "MP" was because I found one opening one and two closing ones. This seemed to me fairly superfluous, so I edited it. I hope that clarifies everything, but will follow the other half of what you (or the automatic notification that came with your message) voiced about logging on with my username if and when I edit so that I can be more accountable for what I do.

Thanks. 78.145.230.27 (talk) 03:07, 26 January 2015 (UTC)

One of the closing brackets belonged to "(MP)", the other to "(then a Member of Parliament...)". I think it's Wikipedia's stance to properly match parentheses in such a case, even if it results in two subsequent closing parentheses. Fixed. Huon (talk) 13:42, 31 January 2015 (UTC)