Hi Liam, I’m one of the author. Thanks for taking the time reading our paper. Could you explain further your critic?
The COP-Kmeans was more of a robustness check and doesn’t change much the results, the F1 score is 0.997 when comparing the kmeans with cop-kmeans

Another issue with bias would be the changes in crediting practice over time. This is hard to adjust for, because you don't know if credits increase over time because games are more complex or because you're crediting people who might not have been credited before.

My biggest fear (for classification purposes) is not so much that the credits are biased, but that the bias would be different between categories. For example, if 10% of workers are not credited for all games, this would bias the raw measure but not change the resulting classification.

Still, large publishers generally invest in larger projects (with larger teams), hence it's not so much a problem. Whereas for playtesters and special thanks, this relationship is not monotonic. 3/3

For playtesters, the issue is we have some big outliers, with some games having dozens of (unpaid) playtesters that skews the credits.
There are other issues of course, like publishers including all their staff in the credits of a game even though the team is ridiculously small. 2/n

I totally agree with you. We took this decision because those two categories generally added more noise than signal. One example: crowdfunded games include the backers in their special thanks, which adds hundreds (even thousands) of unwanted names, unrelated to the amount of work in the game. 1/n

Erratum: Mailgun gives you only 100 emails/day actually. But for your 3 customers it’s plenty enough 🤗

8/ Productivity: @obsdmd for notes taking + the Relay plugin for collaboration. Apple's Reminder for collaborative todo lists

7/ Analytics: goatcounter.com is the GOAT. Simple, small, privacy-friendly, and doesn't slow down your site like Google Analytics. Sometimes the free options are actually better.

6/ Emails: Mailgun gives you 1000 emails per day for free. That's more than you need until you're making real money. Stop overthinking email.

5/ Frontend: Everything is free here. Use React if you're risk-averse. Go Sveltekit
if you want to develop twice as fast with half the code. Either way, it costs you nothing but time.

4/ Auth: Supabase auth. Dead simple, works out of the box, and you're already using their DB anyway.

3/ Backend: Serverless is expensive - unless you just use Cloudflare Workers. 100k requests per day for free? No shit they're losing money

2/ Database: Cloudflare D1 again (seeing a pattern?), or Supabase if you absolutely need postgres. Both free tiers will carry you well past your first paying customers.

1/ Forget about Vercel and Netlify, go Cloudflare for free CDN and smart cache. Their free tier is more than enough to get you started with unlimited sites, and you'll get better performance where it matters.