This report by The Aerospace Corporation shows the frequently unquestioned assumptions currently crippling US human spaceflight.
The US currently has two operational unmanned launchers capable of carrying a substantial manned spacecraft: Delta IV and Atlas V. The Delta IV family has failed to substantially complete one mission out of ten launches, Atlas V one out of fourteen. The Saturn V that carried the manned capsules of the Apollo program had one clear failure out of thirteen launches with Apollo 6, and a close call with Apollo 13 which almost failed to reach orbit before the famous crisis en route to the moon.
Both launchers are relatively new, and their record should improve over time. More mature unmanned launchers like Delta II, Atlas II-III and Ariane IV had a successful launch rate of 98-97%.
A launch abort system (LAS) like the ones carried by Apollo and Soyuz could probably allow the crew to survive a launch failure about 80% of the time, so an otherwise unmodified Delta IV with a LAS might kill its crew once every 50-250 ascents. For comparison, the Shuttle averaged a loss of crew every 63.5 launches.
At the anticipated post shuttle NASA manned flight rate of two spacecraft a year, such a system might lose a crew once every 25-125 years.
To improve the safety of a manned version, NASA would want take additional steps to "human-rate" the launcher, primarily by improving structural margins, adding redundancy to some systems, increasing the qualification testing of the hardware, and increased inspections.
Option 6, the most similar human-rated version to the current launcher in the Aerospace report, uses a single RL-10 engine on the upper stage, and costs about $4 billion more for development and the first 14 launches than an unmodified Delta IV.. Other options with more powerful upper stages cost more but also improve payload.
Under those assumptions, human rating that goes much beyond adding a LAS and the associated hardware and software to allow a timely abort for a crewed capsule seems like an extraordinarily costly way to save human lives. If additional investment in human rating a Delta IV reduced fatal accidents to zero, it might save 1-12 lives over the course of the next 50 years.
If you are willing to spend $4 billion to save human lives, marginal improvements to a fairly reliable existing orbital launcher are probably not going to be at the top of your choices. Highway safety or infant nutrition will probably save many more lives for less money.
Much of the added cost seems to come from NASA overhead priced at 27-32% of contractor cost. Given that Lockheed Martin and Boeing are trusted to launch billion dollar spysats and planetary spacecraft with considerably less expensive oversight, it's worthwhile to question just how much value NASA adds for that additional cost.
And if you are proposing to spend over $300 million per life saved, then there really should be some discussion about how much the different improvements add to reliability, how many lives they save at what price, and how certain you are of your estimates. Probabilistic risk assessments for space launchers have a history of being very wrong and of grossly overestimating the reliability of their conclusions.
Perhaps the government has a good reason to put a very, very high value on preventing astronaut deaths on their way to space and back. Certainly the US public seems to care a lot more about such deaths than those of most other mortals, very famous celebrities excepted. And the US public is ultimately footing the bill for NASA's human space flight, so their preferences deserve some consideration.
It might be desirable to align the interests of NASA and a commercial launch provider by making a significant part of the contract conditional on a good safety record. For example, the provider offers 20 launches for up to $5 billion That amount includes a $500 million bonus paid for each ten flights in a row without mission failure, and an additional billion dollars bonus for 20 flights in a row without a mission failure.
Under this kind of contact, you can expect the launch provider to work very hard to keep mission failures to a minimum,