Meta & Children's Online SafetyChat: God knows what happened to those kids
Thread View
quoted in an excerpt in the archive.
Wednesday 4 November 2020
06:50LisaPouya wanted to highlight S214841
06:51Pouyayes so basically we need to make sure any Messenger based DMAT subproblem first sets up a correct escalation path and can handle Mat2 in reporting mode
06:51PouyaIf not I would argue they are not even basic maturity in dmat
06:51PouyaFor example IIC is currently not even basic
06:52PouyaCan we do an audit of which problems rely on messenger report escalations and if they have already implemented an escalation path correctly?
06:52PouyaI am working on solving this SEV for IIC on D24690159
06:53PouyaThe million dollar question here is whether we need to deep copy EntReported Conversation for escalation jobs
06:53PouyaI'm talking to both MWB and Jessie on this as he seems to be the leading Mat2 expert these days :-(
06:54PouyaHappy to collaborate on this @Vincenzo to make it part of the framework
06:55PouyaBTW this SEV is particularly painful. Thousands of minors have been reporting IIC T1 abuse and even though we knew that there is IIC T1 going on (more than 50% of which is Sextortion which can lead to suicide) we haven't done anything. We had a broken escalation path and no measurement. God knows what happened to those kids
06:56Jennifer MaliaThe perpetrators have been disabled, yes?
06:56PouyaYes but what does that do?
06:56Jennifer Malia(Just checking, but I know we were meant to do much more.)
06:57PouyaAnyway it's just an FYI
06:57Jennifer MaliaAt least they weren't continuing to abuse directly. But I agree it's a small measure.
06:57PouyaIn Sextortion normally the abusers have a network of fake accounts so it's probably not very helpful
07:01LisaNote I only have 2 FTCs staffing English Escalations only, about 200 hours more a week requested but not on the way yet. 2 more English FTCS on the way by EOY but that's it.
07:02Lisabacklogs are 3000++ across IG, FB etc. -> IG escalations were never worked, someone created a Queue before I started and COINV/Chelsea/Zach didn't know about it (IG English backlog 1897)
07:03LisaPLTF EX this exists already, it's all the 'ESCALATIONS' in the table in this doc
07:03Lisahttps://fb.quip.com/daAAASNCMp5d#UMGACAC1mf2
07:03LisaShared an image
07:04LisaQueues are any Queue with name like '[Messages]'
07:11PouyaWe need to solve these backlog issues and measure our turn-around time. As Dom said in the larger meeting this is one of the requirements for getting to measured on DMAT framework. I will prioritize fixing some of these issues when I get a moment. Realistically in Q1 though.
I'll do anything in my power to empower the triagers and investigators so they can be much faster
07:14LisaInvestigators don't matter it's all Triage now. It used to go:
Scaled Review > Triage > Anything not dropped to COINV
now
Scaled Review > Triage send standard report > less than 10 a week that meet specific conditions go to COINV
07:16LisaThey're working more and more each week, they worked 177 jobs last week, but the better any text model gets at increasing Scaled Review escalation rate, the less they'll be able to meet demand
https://fburl.com/scuba/single_review_tool_main/2nzqz6wo
07:16PouyaDidn't we say that there are some tooling issues slowing them down? For example the very manual nature of the reports to NCMEC?
07:16LisaCOINV have backlog of 1300 also (was 2000) and we know that's reportable
07:17Jennifer MaliaI think we can all agree that this whole situation is Not Good and there are Several issues blocking us from correctly reporting all the T1 we know about to NCMEC/LE.
07:18PouyaSorry @Vincenzo if it was just too much information dumped too quickly onto this chat. Happy to provide more context if needed
07:18Jennifer MaliaLet's make sure we separate them and address each one.
07:18PouyaWe need to have proper measurements for each piece of the puzzle and we should keep in mind that as @Lisa said things might change unexpectedly
07:19Jennifer MaliaYes agree.
07:20PouyaLet's make sure we are never again the reason why a kid did not get help.
07:20LisaI have a fancy GRT now that provides enough labels to auto populate a report that's the tooling gap we have They have to leave SRT and add info to MRSNCMEC UI to send a report, this adds up to 5 min per job.
Working the jobs is the time consuming part.
Example labels below. Minor is 13, Responsible is adult. Met for Sex + Proximity.
NCMEC", "decision": "ESCALATE_NCMEC", "admin_action_id": null, "prepopulatecr": "", "aid":"escalate_ncmec", "acti on_id":"escalate_ncmec"}],"labels": ["default_question:: NCMEC_BLUE", "AgeGap:: 13_ADULT", "CSSE:: NONE_CSSE","
07:23Lisa: NONE_SEXTORTION", "CEI:: NONE_CEI", "MFS:: CONFIRMED_CONTACT_OFFENSE", "SUPERMINOR::MINOR_ON_MINOR _ABUSE", "Agg Facts:: PROXIMITY", "NCMEC:
We are losing 80-90 hours of German/French Scaled Review so I am requesting that gets diverted to sending reports instead. I have no been sacrificing any Scaled Review Tier 1 Market time as getting the text model done enough seemed top priority, then sending reports, then more Tier 1 Market expansion.
If this is top prio I can tell Ops Readiness that, future proofing enforcement in general is what we agreed when we all met and shouted about Ops Readiness priorities into a spreadsheet.
07:43Pouya@Lisa please make sure to list anything Eng can do to improve AHT here. I'll try to bump the priority of such tasks in H1 roadmap
10:48Lisa@silent yes I have it on our roadmap asks and will put together a Task for it with better details, night night