tweetindex
PT

Gabriel Geiger

@gabriels_geiger

Investigative journalist @LHreports. Surveillance and algorithms. Bylines @guardian @bellingcat @derspiegel @lemondefr @wired @NRC Signal: gabriel_geiger.06

3 420Followers
1 735Following
1 146Posts total
2.8MViews on collected posts

Posts recentes

@gabriels_geiger @fk_media Yeah FOIA in Sweden is basically a scam. It works for non-sensitive stuff, but not for the interesting. Typically you need to win two times in court, after which the court will process the FOIA on behalf of the fraudulent authority. 32.1K views · 449 likes · 3 reposts · 2 replies Open on X →
As usual this work wouldn’t be possible without our expert reviewers who generously took the time to review our methodology. Special thanks to @AmnestyTech, @DavidN1194 for their support https://t.co/PBjcEJHqGt 71.8K views · 592 likes · 10 reposts · 4 replies Open on X →
This investigation would’ve died a year ago if it weren’t for Sascha Granberg, who it was an incredible pleasure to work with. I literally felt like a crazy person with this story until he came onboard. He’s one of the best freelance (data) journalists out there — do reach out to 53.9K views · 714 likes · 16 reposts · 1 replies Open on X →
I’m really excited about an article interrogating the agency’s projections of widescale welfare fraud. I’ll write a thread on this when it comes out, but you can already see our analysis in the methodology. 57.4K views · 681 likes · 13 reposts · 2 replies Open on X →
If you want to read how we tested the data, we’ve published another methodology (1,000 words shorter than last time!) here. It also includes a link to our Github, where we’ve published all the code and data underlying our analysis https://t.co/L0pcAcOu97 68.1K views · 760 likes · 20 reposts · 1 replies Open on X →
Over the next few days, @SvD will publish a series of articles with our findings. The first article unpacks the scale of profiling and highlights the results of our analysis https://t.co/vk3jsy2KPF 62.7K views · 807 likes · 29 reposts · 2 replies Open on X →
That’s because @fk_media ‘s fairness procedure allows, for example, 4 immigrants to be selected for every Swede without having to test if this is at all justified. Even if the model fails these 'generous' thresholds there’s an escape hatch: It's okay if legal says it's okay https
89K views · 964 likes · 25 reposts · 1 replies Open on X →
Can we see the results of this fairness procedure? Confidential — aggregate fairness data would allow fraudsters to game the system, apparently. 69.2K views · 817 likes · 7 reposts · 2 replies Open on X →
But through FOIing another agency, we obtained a dataset of applicants for temp child support who had been flagged for investigation by the algo. It was broken down by demographic characteristics (gender, immigration status, education, income) and outcomes (mistake vs no mistake) 122.5K views · 1.5K likes · 24 reposts · 1 replies Open on X →
But there was a catch. The agency said it had created a new fairness procedure to prevent discrimination. We obtained it via FOIA. When we tested it against our data, all the problems re: discrimination seemed to disappear 95K views · 1.1K likes · 12 reposts · 1 replies Open on X →
With support from academics, we tested six different statistical fairness definitions. The results are disturbing. Women, immigrants, low-income earners and people without a university education who were innocent more likely to be *wrongly* selected for investigation 109.7K views · 1.7K likes · 44 reposts · 5 replies Open on X →
At one point we even conducted an experiment: FOI for information we knew @fk_media had published in its annual report. “The information you are requesting is confidential” 133.1K views · 2.4K likes · 53 reposts · 3 replies Open on X →
How many people receive a risk score? Confidential. How many people has the system flagged? Confidential. How much did the system cost to build? Confidential. How do you define fraud? Confidential 135.1K views · 2.1K likes · 22 reposts · 1 replies Open on X →
We sent 100s of emails, 12s of FOIs, 2x in court. @fk_media refused to answer the most basic questions arguing baselessly that it would allow fraudsters to game the system https://t.co/N8epa8swSY
0:43
144.1K views · 2.2K likes · 34 reposts · 4 replies Open on X →
I do a lot of FOIA and I’m used to stonewalling. But Sweden surprised me. After 3 years of telling me no, a high level official at Sweden’s welfare agency @fk_media accidentally left me on cc, writing “Now we must hope that we are done with him.” https://t.co/qtNiXLPmCg
1.5M views · 20.3K likes · 1.2K reposts · 77 replies Open on X →
Big thanks to experts who graciously reviewed various parts of our design, methodology and findings, including @AmnestyTech, @DavidN1194, @informusiccs, @merbroussard, @eikepetersen1, @acidflask, Moritz Hardt, and Alexandra Chouldechova 72.9K views · 17 likes · 1 reposts · 0 replies Open on X →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

144.1K27 Nov
135.1K
133.1K
109.7K
95K
122.5K
69.2K
89K
62.7K
68.1K
57.4K
53.9K
71.8K
32.1K

Last 14 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

1.57%27 Nov
1.58%
1.88%
1.55%
1.21%
1.26%
1.19%
1.11%
1.34%
1.15%
1.21%
1.36%
0.84%
1.42%

Reactions — likes, reposts, replies and quotes — divided by views.

What the audience does

Likes88.7%37 153 in total
Reposts3.6%1 511 in total
Replies0.3%107 in total
Quotes0.2%94 in total
Bookmarks7.2%3 025 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Contas semelhantes