Given those two statements, how much testing ($$$$) are agencies like NASA going to require before attaching self-loading cargo to the ship?

A large part of SpaceX’s success in winning commercial crew approval is because the stack is actually very traditional. The primary innovation is first stage recovery which arguably doesn’t add any risk to the crew.

Fully new materials and propellants completely change the game when it comes to risk. I hope the succeed and I think they can succeed. It’s just that this round is going to take a little more money than the Falcon/Dragon program and I hope they’re able to find enough of it.

There is a lot of innovation in Falcon 9 compared to other crewed rockets - pneumatic separation systems, subcooled propelants, propelant loading with crew on board, liquid abort engines on Dragon and of course the stage 1 recovery systems.

As for human rating Starship - I think the best bet is to launch them as many times as possible unmanned to demonstrate it is safe, kinda like the thousands of flight hours needed to certify a new airliner. This should be perfectly doable on the fully reusable Starship + Super Heavy combo.

Thanks for pointing out the other aspects. I was aware of the propellant loading with crew onboard and liquid abort engines but just forgot about them. You’re right. There are a substantial number of other innovative aspects.

Yep - also the computers. AFAIK they are both much more powerful (which is actually needed for the first stage to compute the best landing trajectory) as well as using relatively commodity hardware and and normal (IIRC C++) programming languages.

I'm not saying rad hardened aerospace computers and specialized programming languages (ADA and co) are bad - there was a time where they were absolutely the best (any only) choice, but thankfully today there are often better options.

IIRC for the Dragon computers I remember an article that mentioned that if they went with aerospace computers, even just a devkit would have cost >150 thousand dollars, severely limiting available resources for developers and the flight hardware would have lead time in years.

So they went with redundant (IIRC 5 times) x86 with Linux on top & each engineer working on the code can have effectively flight hardware on their desk for use during development.

As the other commenter pointed out, the Falcons have some innovation in there. But Starship's has even mroe. It uses SpaceX's new Raptor engines, which are full-flow CH4/LOX engines. No full-flow staged combustion cycle (FFSCC) rocket engine has ever been used to deliver a payload to LEO (or even used in an actual rocket at all, AFAIK). This is somewhat surprising (and indicative of the difficulty of manifesting the design), given that the design has been around since at least the 1960s and has a lot of benefits over other rocket engine designs.

In that regard, FFSCC rocket engines actually remind me of LFTR-type MSRs:

1. been around for decades

2. clear benefits to the design

3. hasn't managed to be used viably

So I would say Starship + Super Heavy is definitely a riskier investment than NASA has made with Falcon.

Quite a lot, I imagine, but given they're (supposed to be) fully reusable, the idea is that testing will be $$ instead of $$$$.