2) You cant choose the hatcher, it chooses you. You can request some parameters, such as trust level, min fee, but your broadcast goes system wide.
EDIT: Your next transaction is sent to a different hatcher, you can not send a transaction to the same hatcher twice in a row, and the network can see this as the hatcher signs the transaction.
I'm basing this on the above. If clients have a configuration for "trust level", then hatchers with higher trust are going to be favored by clients over hatchers with lower trust, and hatchers with no trust at all will probably have to wait until there are no other hatchers available to service transactions.
This is the bit that gets me confused.
I run a client on my local network - it can only connect to my own hatchers. What stops them from hatching its transactions - noone else is going to see them to get the chance? Or do nodes only ever approve transactions that they saw before they were processed (meaning immediate fork as soon you connect if any transactions were already being processed)?
Is there something that stops hatchers below a certain trust-level from processing transactions? Do they have to back-date the time-stamp so it looks like they waited to give more trusted ones a chance to process them first?
And what's with the claim that honest nodes can detect ping-pongs (not quoted above)? Is there some penalty for sending funds back and forth? Do hatchers need to analyse all transactions to make sure they won't be penalised if they process them? If I send money to someone then they send it back will it be taken away or something? A general statement that something can be detected doesn't explain anything without detail of what happens when it IS detected.
If generating key pairs is fairly trivial then there'd be no need for anyone doing transaction padding to use same address twice anyway. You just have X blocks of cash that you keep moving to new addresses. Or one big block if its weighted by transaction size.