Also, hydrogen's specific impulse advantage comes with major disadvantages. If you just took Starship and gave it hydrolox engines, it wouldn't have the performance needed to reach orbit, because you simply wouldn't be able to fit enough LH2 in the available tank volume. Despite the higher specific impulse, you need a far larger tank volume just to match the performance of denser propellants. And now you've got a lot more surface area to shield against reentry, and it also has to be insulated to keep the LH2 from boiling off too quickly. Notice how the Shuttle didn't even try to bring its propellant tanks back...

A notable difference is that replacing a single Shuttle tile could take three days. Most of Starship's tiles are attached with metal clips and can be replaced in seconds, they can largely re-tile a ship in a day. The vast majority of the tiles are also a few uniform shapes...each Shuttle tile was unique, made for a specific location. So even if not perfect, it's already a huge improvement.

"the experience in building the cargo craft really helped them."

It certainly did, but it's not like Boeing lacked experience. In fact, early on there was a serious push for sole-sourcing the crew contract to Boeing based on their history with the Shuttle, that SpaceX was too inexperienced and couldn't be trusted to get the job done.

This is a microcontroller, not an embedded Linux SoC. It has a minuscule fraction of the power consumption, runs a simple RTOS, and can achieve levels of determinism, latency, and timing precision that are impractical with a full Linux OS running on the chip. A microcontroller can bit-bang protocols at speeds the Zero 2W couldn't match, or which it simply couldn't manage the timing requirements for. The programmable PIO stuff enhances those advantages.

The Cortex M0+ of the RP2040 was rather underpowered, though. Its main use is in very tiny systems where transistor count or power consumption are the main considerations. The M33 is a much more capable microcontroller, with things like hardware integer division and a FPU.

SpaceX's approach to "move fast and break things" can also be described as "test early, test often" and "fail fast". It's heavily based on real-world testing to root out failure modes in systems too complex or poorly understood to handle otherwise (as demonstrated with the Shuttle, the alternative is "move slowly and break things anyway, with people on board"). What's industry standard practice may be overly conservative, obsolete, and even flawed, but you need extensive testing to demonstrate that.

This philosophy has made Falcon 9 not just the launch vehicle with the highest flight rate and lowest cost, but also the highest reliability, and has let them do 11 crewed Dragon missions to the ISS (the demo flight, the originally contracted 6 flights and 2 additional ones, and 2 privately contracted flights) while Boeing's Starliner is working through issues discovered after a failed first attempt at a demo flight and problematic second attempt.

OceanView's approach was apparently more "skip testing too because it's too expensive and takes too long".

It's not even just that the price is unreasonable. A lot of the apps are billed yearly, and they're looking at huge API bills coming in 30 days. They don't have the finances to eat the losses until enough of their subscriber base renews at a higher price or leaves that they can start making that money back. More than the prices, the timeline makes it clear that the intent is to kill off these apps.

Their suborbital flights take place almost entirely within the atmosphere, only reaching the Karman line after the ascent burn completes and only starting up for the landing burn (for the propulsion module) after an aerodynamic flight back to the landing site. The experience isn't totally irrelevant, but its relevance is rather limited.

It could in principle (though a more likely approach is to refuel with a tanker/cargo-crew transport in cislunar orbit), but NASA's Artemis plans would have it sitting empty for years between uses. NASA hasn't even bought a second landing, and won't even ask for one until they have another lander from someone else under the "Sustaining Lunar Development" program. Between those SLD missions and Tollbooth-specific missions, it could be 2 or 3 years between Starship HLS missions. That's a long time to sit in orbit unattended, and probably a lot of work to test things out and do required maintenance when the next crew finally arrives.

Maybe SpaceX will be able to offer the HLS Starship's services to others in between Artemis missions...someone who actually wants a lunar base.

SLS is not "insurance" against anything. If the HLS version of Starship doesn't get developed, its replacement will be a new lander to be launched using Falcon Heavy, New Glenn, Vulcan, etc, because the one and only SLS launch available for each Artemis mission is tied up getting the Orion capsule out to near-lunar orbit. SLS isn't a lander, and even Block 1B won't be able to deliver one alongside the Orion (co-manifested cargo capacity being only 10 t). SLS is purely a jobs program.

Starship can be an insurance policy for SLS/Orion, as a Starship variant could easily do what SLS/Orion does. However, SLS/Orion is no insurance against problems with Starship, since all it can do is deliver a few astronauts to NRHO. It can't even get them to low lunar orbit, let alone the lunar surface. The only backup for Starship is another lunar lander...one that doesn't use SLS, since the SLS is needed to launch Orion, and it can't fly often enough to do both jobs.

Spacecraft propulsion doesn't require power production. A fusion reaction with a Q of 1 will not be self sustaining, but will still double the power output of a fusion propulsion system.

While yes, that's true...Starliner's only contracted for 6 operational launches, currently scheduled for 2023 to 2026. That uses up the available Atlas V's, and Boeing's not shown much interest in adapting it for Vulcan, which would require certifying Vulcan to carry crew...they want someone else to pay for that. So, for about half of the ISS's planned remaining lifespan, Starliner won't be around to be a backup.

The full Saturn V was actually not capable of launching useful payload to LEO, the "payload to LEO" figures include the third stage and propellant load (the SLS figures are similarly misleading). Skylab actually used a cut-down version of the rocket, the third stage being modified into the ~76 t space station.

The Starship figure is the actual useful payload to LEO. It's designed to be most effective at delivering mass to LEO, to make the most of orbital propellant transfer. And while Saturn V could get 43.5 t to TLI, and SLS Block 1 27 t, Starship's payload to TLI is 100 t, because that's what fits in its payload bay, and it's well within its delta-v budget to make TLI from LEO with a maximum payload and partial propellant load. The total number of launches depends on boiloff, launch cadence, tanker performance, etc, but could be just a few additional launches. It's feasible they could do it with entirely expendable Starships...more expensive, but they'd only have to do it twice, with NASA planning to spend a couple years on the Tollbooth before going back to the moon after Artemis III.

Antarctica and the ocean are essentially illegal to settle due to international treaties and agreements.

It really isn't even good for that. Their scaled-up, full-power version is going to have a max payload of about 200 kg. You're going to have to cram orbital propulsion, guidance, power, and rendezvous systems in that along with the ice, and robot arm/airlock operations of some sort to retrieve it, all for a small fraction of the amount of water that could be sent alongside astronauts on a Dragon, or about 0.1% of a Starship load.