What are the differences between RDT1.0 and RDT2.0? What improvements has RDT3.0 made over

Don't just argue that deep access is preferable because it's faster; provide some concrete reasons why a developer would want deep access.

Suppose a piece of software that lets a surgeon operate on a patient who is physically situated in another area over the Internet. Why would someone continue after the curtain has closed? How much harm do they want to cause? Which of your shortcomings do you think they would emphasize in an effort to discredit you? Is it feasible for harm to happen without a malicious aggressor trying to take advantage of these weaknesses?

Is there a certain set of guidelines that the OS's microkernel follows? Where does this approach diverge most sharply from the modular approach?