You can delete any comment you made, or any thread less than a week old, but you can't just blow away all your comments at once or older threads because that disrupts conversation. You can change your username or set your email address to garbage at any time, but remember that you're anonymous to begin with unless you chose to use your real name.
b) where are accounts "stored"?
In a database on the server.
c) Who has power to delete or act on accounts?
Patrick has the power to delete accounts, and there are a few trusted users who can moderate comments.
2/ Namespace architecture:
a/ Is it one global namespace or different "sub-spaces" (regions, subreddits, groups, etc)?
One global namespace.
b/ how do usernames work across the above? Do you get to take your username across each space?
a) what transparency do you have on content moderation actions (account suspensions, any algorithmic levers)?
You can still view comments that were marked personal, etc, by doing a "view source" of the html. That's inconvenient, but it's supposed to be, to keep personal comments out of the flow.
If split across multiple nodes, Node trust+incentives: what is the incentive schemes for various nodes/relays/intermediaries? What are the trust assumptions? How do you resist centralization?
The site is just one node.
7/ Privacy and data guarantees:
What privacy and data guarantees exist at various levels of the system?
Your email address is private, visible only to Patrick.
8/ Censorship resistance:
Which actors can censor content and how? (group owners, nodes,...)
Patrick could censor anything because he has control over the database. Moderators could try to flag and censor comments, but Patrick would see those actions and reverse them if not justified.
9/ Centralization:
Which parts of the service need an opaque centralized service in the middle?
Answering his questions: