Parcel level bot detection

[Phil Deakins]

So am I? lol

[Paul Hexem]

1 hour ago, Phil Deakins said:

There is no requirement for bots to be registered as scripted agents. It's not against the rules for them not to be.

I'd double check on that.

If true, this entire thing, from the Belli estate ban to the actual scripted function, is a joke on residents that want to "block bots".

[xDancingStarx]

2 hours ago, Phil Deakins said:

Of course, traffic bots wouldn't be registered as scripted agents, or they wouldn't be traffic bots

"Traffic bot" is a bad term for traffic gaming users because it suggests that if there are no bots anymore then there also are no "traffic bots" anymore, which is wrong, because you can game traffic on an unattended text viewer that is not scripted or automated.

 

2 hours ago, Phil Deakins said:

There is no requirement for bots to be registered as scripted agents. It's not against the rules for them not to be.

I have no idea where you get your information from.

"If you own a Second Life account primarily operating as a Scripted Agent, you must specify this on the Scripted Agent Status page." source: https://wiki.secondlife.com/wiki/Linden_Lab_Official:Scripted_Agent_Policy

[Qie Niangao]

7 minutes ago, xDancingStarx said:

I have no idea where you get your information from.

I have a pretty good idea. He may have read the page before April 2023. Check this out to see how completely this policy was replaced at that point, with scripted agents fully redefined as part of the effort to rein-in the population explosion in roaming bots on the grid.

When the original scripted agent policy was written, landbots were very nearly the only roaming bots, with only a very few exceptions such as Tyche's venerable gridsurvey bot(s). At that time, the vast majority of bots were traffic bots and the sole practical significance of setting scripted agent on an account was to keep it from counting as traffic. If you ran a group-invite bot and kept it on a parcel that did not collect traffic (wasn't set to show in Search), nobody cared, and what the landbots were doing was a violation whether they were declared or not, so really only traffic bots needed their scripted agent bit set.

[xDancingStarx]

11 minutes ago, Qie Niangao said:

I have a pretty good idea. He may have read the page before April 2023.

Doublechecking...

Before that date it said this, or am I looking at it the wrong way? 

"If you own a Second Life account that is primarily operated by a Scripted Agent (a "bot"), identify it as a bot on the Scripted Agent Status page. By identifying bots, you can help us give you more accurate account statistics and inworld search results."

Furthermore, way before April 2023 Linden already encouraged to file an abuse report if someone suspects an account to be a bot but not set as scripted agent, and they have also been known to take action, before April 2023.

Edited August 20, 2023 by xDancingStarx

[Qie Niangao]

4 minutes ago, xDancingStarx said:

Doublechecking...

To be sure, it was already stated that scripted agents needed to be registered, but seriously: nobody cared as long as they didn't hang out on parcels listed in Search.

[Qie Niangao]

8 minutes ago, xDancingStarx said:

Before that date it said this, or am I looking at it the wrong way? 

Maybe the wrong way? Anyway, this was the policy up to March 30, 2023. It was pretty short. It did want the scripted agents registered, but… well, here's an excerpt starting from there:

Quote

Set scripted agent status

If you own a Second Life account that is primarily operated by a Scripted Agent (a "bot"), identify it as a bot on the Scripted Agent Status page. By identifying bots, you can help us give you more accurate account statistics and inworld search results.

Enforcement

Linden Lab routinely looks at search results, and will treat use of bots to gain an unfair search advantage as an abuse issue. You do not need to file an abuse report for violations; Linden Lab monitors the search results.

To begin, Linden Lab sends a warning message to the land owner. If the issue persists, then account suspension or removal from search may result. If you feel there has been an unfair decision in this regard, use the abuse appeals process.

Linden Lab routinely looks at message volume, and will treat use of bots to send excessive messages as an abuse issue. You do not need to file an abuse report for violations; Linden Lab monitors aggregate communication volume.

It's pretty search- and traffic-centric.

[Rowan Amore]

31 minutes ago, Qie Niangao said:

Maybe the wrong way? Anyway, this was the policy up to March 30, 2023. It was pretty short. It did want the scripted agents registered, but… well, here's an excerpt starting from there:

It's pretty search- and traffic-centric.

LL doesn't see the problem because the LL viewer doesn't have legacy search which shows completely different results.  

My most recent check of Escorts in Firestorm Places showed one club with a traffic score of 90K.  Using Escorts in the LL search brings up one of the oldest "oasis" escort places.  2 different results.  The former having 2 people in the club and a box full of unregistered bots. LL doesn't see this since even if they did monitor search results, they aren't the results a lot of people see.

[Phil Deakins]

If LL's policy on scripted agents has changed since they introduced the scripted agent status, then I've missed the change. It wasn't a requirement back then unless they were on land that was set to show in search.

Edited August 20, 2023 by Phil Deakins

[Phil Deakins]

On 8/20/2023 at 6:24 PM, Silent Mistwalker said:

Good luck! Let us know how it goes please.

Well, I got a reply on the ticket, but not what I was looking for. It was as I thought - a support ticket is a request for a fix the account's problem, and that's what it got. Wicked Leigh is now not returned as scripted agent lol.

However, my JIRA attracted a response from Maestro Linden. He asked me to take each of the false positive accounts over to Bellisseria to see if they could get in. That's how I discovered that Wicked Leigh is no longer a scripted agent lol. The other 4 can't get in. So we may make some progress on finding out what causes the false positives.

I'm still leaning towards my current theory that I posted on the previous page, even though Wicked Leigh was never used as a bot, at least not after traffic bots became illegal.

[Love Zhaoying]

Apparently, once the bit it set (for whatever reason) it doesn't get unset / refreshed.. very unfortunate.  It would help if you didn't have so many bot accounts.

The real disappointment for me, is that the bit is apparently not set real-time.

Edited August 21, 2023 by Love Zhaoying

[Qie Niangao]

13 minutes ago, Love Zhaoying said:

Apparently, once the bit it set (for whatever reason) it doesn't get unset / refreshed.. very unfortunate.  It would help if you didn't have so many bot accounts.

The real disappointment for me, is that the bit is apparently not set real-time.

Well, we know that in general the AGENT_AUTOMATED bit is set and cleared real-time (more or less) to reflect the setting we give on the scripted agent status form, but we know from Wicked Leigh's experience that the setting on that form is overridden by something else that Support is able to correct. The entire mechanism of this overriding is the mystery here. If I recall the story so far, Wicked's scripted agent status form would accept a change to not scripted agent as far as that form is concerned, but that change did not actually clear the AGENT_AUTOMATED bit. So we know AGENT_AUTOMATED is calculated from both what's on that form and this other mysterious overriding factor that's not reflected on the form.

I don't think we know when that complicated AGENT_AUTOMATED calculation takes place; it could be "real time" when the agent enters the region or when llGetAgentInfo() is called, or it could be static, with the overriding factor preventing a change made on the form from affecting the bit (but nonetheless recording it somewhere else so the form populates with the futilely-changed setting).

(Sorry, that's probably repeating the obvious.)

Presumably whomever in Support worked that ticket knows what they changed to "fix" Wicked. The interest isn't entirely academic because we'd kinda like to know how prevalent these false positives are, and if there's a sampling bias in which accounts are false positive.

[Silent Mistwalker]

Here's a thought.

Is it possible that removing the account while it is logged inworld makes any difference?  I know for some things you were better off not being logged inworld when certain changes to an account were made years ago. Meaning you had to relog for the changes to take effect (similar to changing some viewer preferences). So, if it was logged in and then out without logging back in for a long period of time it might not have "triggered/completed" whatever code it should have.

I think in this case the better option would be to not be logged in when removing an account as a scripted agent.

If any of that makes sense.

[Phil Deakins]

1 hour ago, Silent Mistwalker said:

Is it possible that removing the account while it is logged inworld makes any difference? 

The 4 remaining false positives have never had anything changed about their account. They are just 4 of many that I created way back and used only for the purposes they were created. Wicked's account was changed a long time ago, when it was briefly a Premium.

[Love Zhaoying]

10 minutes ago, Phil Deakins said:

used only for the purposes they were created.

Bots!!!

[Phil Deakins]

I'm still leaning towards my previous thought. The 4-byte (assuming 4 bytes) integer that is used for an account's flags has plenty of spare bits for flags we don't know about, and one of them has been, or is still being, used for other things, such as being thought to be a bot, perhaps from ARs.

When agent_automated is requested for an account, instead of ANDing the flags with 0x4000, it is ANDed with another number that examines 2 of the bits. Anything > 0 indicates either scripted agent, or the other, or both, and is returned as scripted agent. It would fit with being able to change the status in the Scripted Agent Status page, and it sticking as 1 or 0. For that page, only the 0x4000 bit is ANDed. And it could possibly fit with my false positives, especially if individually they were ARed rather than seen together with other ones.

It may be an old flag that is no longer used, but any that were set in the old days remain set.
 

14 minutes ago, Love Zhaoying said:

Bots!!!

Bots and Picks.

Edited August 21, 2023 by Phil Deakins

[Phil Deakins]

THE ANSWER

Maestro Linden posted in the JIRA that the false positives are actually flagged as scripted agents internally. It means that requests for AGENT_AUTOMATED check both bit 15 (the agent_automated flag) and an internal flag. If either flag is set, then 'scripted agent' is returned.

It is necessary to do it that way for LL to try and keep bots out of Bellisseria and other places, which is what they try to do. If only bit 15 is used, then bots couldn't be kept out because they'd be set as human, not scripted agents.

So AGENT_AUTOMATED works as intended, and there is no bug. I am satisfied

In the process, I've discovered 2 unpublished flags. I'm sure I know what one of them is for, but the other is a mystery.

Edited August 24, 2023 by Phil Deakins

[Qie Niangao]

25 minutes ago, Phil Deakins said:

So AGENT_AUTOMATED works as intended, and there is no bug.

Now that I think about it, it works better than we might have hoped. I'd been thinking it only reflected self-identified scripted agent status, so bans based on it would only discourage botrunners from identifying their bots. Now we know it's deeper than that, and llGetAgentInfo can be used to ban known bots that are not merely self-identified. Now it seems worth doing for some folks. Not a panacea—parcel level bans are useless against data collection—but not quite as counterproductive as it first seemed.

[Phil Deakins]

Yep. Knowing does have its merits

[Garnet Psaltery]

3 hours ago, Phil Deakins said:

In the process, I've discovered 2 unpublished flags. I'm sure I know what one of them is for, but the other is a mystery.

As a person who quite likes videos with people blowing things up for fun and showing the results in slow-mo, may I suggest you go to Beta and try changing it?  Please take pictures.