Moderation Service

This is incorrect. Please show one scrap of evidence of this 'personal grudge'. "One of my mutuals got labeled and added to the block list" is not 'evidence'. Labeling policy is unchanged, and available for review via pinned post.

Block-listed the 'flydubai false flag' accounts - 300+ new out of 500+ total. Manual review, prompt= give oxygen to 'flydubai false flag' trope => unreliable narrator, irrespective of motivation Account themes rarely change on social media - why waste precious online time arguing with these?

PS one useful enhancement here was to Tag the auto-reports with the rule-id that triggered automation. This means that once a string match rule is added - "false flag", "Zionism is racism", - the Ozone UI allows filtering on just those auto-reports.

Block-listed the 'flydubai false flag' accounts - 300+ new out of 500+ total. Manual review, prompt= give oxygen to 'flydubai false flag' trope => unreliable narrator, irrespective of motivation Account themes rarely change on social media - why waste precious online time arguing with these?

1. Moderation tools (which are much improved now) for bulk labeling 2. SQL bulk 'insert if not found' to the table that drives block-list automation This was still time-consuming, but highly leveraged - I don't see any more auto-reports to review once they are blocklisted

State of the world update. First, a new 'antizionist' trope bsky.app/profile/stev... The account is a 'network hub'. Soon: monitor by building out subgraph real-time activity tracking in Neo4j. Most tooling is also in place to reconstruct all of an account's activity and network from T=0 ❤️ #atproto

This is analogous to the old Twitter "state-affiliated media" tags. Visual clues in general on social media are over-trusted. [Hence, no AI moderation and limited auto-labeling of accounts.] This leads to and confirms bigotry. This toxicity easily bleeds into your network via hasty interaction.

State of the world update. First, a new 'antizionist' trope bsky.app/profile/stev... The account is a 'network hub'. Soon: monitor by building out subgraph real-time activity tracking in Neo4j. Most tooling is also in place to reconstruct all of an account's activity and network from T=0 ❤️ #atproto

Only the most motivated will do the work to build these. Passionate users will migrate to their communities of choice and presumably have a say in moderation. Formalizing discrete 'town square corners'. communities.bsky.app

State of the world update. First, a new 'antizionist' trope bsky.app/profile/stev... The account is a 'network hub'. Soon: monitor by building out subgraph real-time activity tracking in Neo4j. Most tooling is also in place to reconstruct all of an account's activity and network from T=0 ❤️ #atproto

I have lost count of how many dumbass lists I am on. Block lists speak to a Pavlovian response learned on the proprietary pro-authoritarian platforms - X, Meta: Most people just view moderation as a black box for wish fulfilment, for good or ill.

State of the world update. First, a new 'antizionist' trope bsky.app/profile/stev... The account is a 'network hub'. Soon: monitor by building out subgraph real-time activity tracking in Neo4j. Most tooling is also in place to reconstruct all of an account's activity and network from T=0 ❤️ #atproto

People are often surprised to find that their strongly-help opinions get their account labeled as 'abusive,disinfo'. Why this is? An open secret ⬇️

State of the world update. First, a new 'antizionist' trope bsky.app/profile/stev... The account is a 'network hub'. Soon: monitor by building out subgraph real-time activity tracking in Neo4j. Most tooling is also in place to reconstruct all of an account's activity and network from T=0 ❤️ #atproto

Original plan was to build a team. Problem: it's too demanding. In a way, this made the job easier: No AI Automation is needed, must be almost perfect Auto-report uses simple, qualified string matching Limited auto-labeling, mostly confirmed links/slurs Manual review before most account labels

State of the world update. First, a new 'antizionist' trope bsky.app/profile/stev... The account is a 'network hub'. Soon: monitor by building out subgraph real-time activity tracking in Neo4j. Most tooling is also in place to reconstruct all of an account's activity and network from T=0 ❤️ #atproto

I continue to believe that block lists are best avoided, but it's there if you want it. Better to configure labels to get a cognitive hint that a post might not be representative. Example: - follow a 'woke bloke' - unwittingly buy into toxic post-Soviet 'antizionism'.

State of the world update. First, a new 'antizionist' trope bsky.app/profile/stev... The account is a 'network hub'. Soon: monitor by building out subgraph real-time activity tracking in Neo4j. Most tooling is also in place to reconstruct all of an account's activity and network from T=0 ❤️ #atproto

@tomaszeman.bsky.social fyi

receipts are always on hand in case of gaslighting by labeled accounts ⬇

Some 'veil' 😂 From the pinned thread ⬇️ To recap, the queries for which you were labeled and added to block-list are: bsky.app/profile/.bsk... bsky.app/profile/.bsk... bsky.app/profile/.bsk... You are free to post antisemitic hate. Users who subscribe are not obliged to have to see it.

Notes from the coalface: Gaza fundraiser spam account in action. Hundreds of posts over a short time period, minimally spaced out to avoid bsky rate limits. These were auto-reported, resulting in an obvious outlying account in the Ozone moderation DB.

All 10 observed in this wave are added to the service's Soft_Deleted block-list - if account owners regain control after a possible hack, they can appeal. Now I have rate limiting logic working (long story), it will be possible to implement detection for this type of thing as a new report type.

The list of accounts that would be "taken down" were this moderation service on a private PDS is building out here. bsky.app/profile/pef-... 'Soft-delete' decisions are subjective, made according to service policy, and can be appealed via the usual channels.

Update. Ramping up active review of reports ahead of the US mid-terms, when I expect disinfo from all sides to increase - MAGA/russia both desperate. Prioritize manual reports first - rerunning tool to Tag all of those. Then, steady state review of worst offenders all-time/recent.

Noise reduction: Enabled 'soft delete' for worst offenders. Could make the 'soft delete' list into a block list if people want it. Ignore accounts deemed harmless or reliable. Cleared out dead accounts (custom tool).