Maybe I've missed how to make this point all along.
In any systems analysis, whether electronic, mechanical or human-in-the-loop, we study risk and we have definitions within the discipline.
Risk = (complexity / maturity) ... in other words, it's a ratio
If something is complicated, but we've done it a super-long time, the risk is lower.
Because the numbers for complexity and maturity don't serve themselves to us on a silver platter, we use comparative norms we develop by gauging against past activities and lessons learned.
Next - comes risk weighting and risk consequence. If the matter at hand is complicated, we break it into steps and assign a risk to each step - each individual risk is then weighted, so something with a low risk - but a big deal - doesn't get ignored because of a step that has a high risk number, but is really peanuts in the big picture.
A final note before focusing: maturity is related to experience - I've met people claiming 10 years experience doing something, and some have that - others have 1 year of experience repeated 10 times.
Why do I mention that? Rooting the captivate is not mature - we can't add up the many successful rooter experiences and say it is. Each new rooter spends - what? a day or two, max? - so if you repeat that for 100k rooters - you still get a relative maturity of 2 days, not 200k days.
Rooting risk = complexity / lower_maturity
The only way to lower that risk value is to judge the relative complexity as lower - and that varies by each human presented into the loop.
The maturity value for rooting is such that it can not be adjudged that risk is not there.
The risk can only be mitigated by pre-study so the rooter knows what to expect (any maturity increase significantly lowers risk) and because that pre-study also allows the candidate to self evaluate the complexity component of the risk for their own personal case.
Finally - risk consequence. We all like to talk about bricks - me, too. In actual practice, that's a single risk consequence and there are others. There are humans in this loop.
We've danced around saying it, but because this has applied to me personally at times in my life, I'll just come out with it:
Make something idiot-proof and a better idiot will come along.
I personally believe that rooting risk can be mitigated for many and made safe enough. The Evo's had a longer rooting history than the Captivate and has a larger dev community contributing to rooting techniques - it's more mature, more people are doing it with fewer consequences than at first. With time, the Captivate will join that ranking.
To say this is a no-risk proposition simply defies the math. If systems analysis has taught us anything, it's that risk math is not wrong.
Soft vs hard bricks should be better defiined.
I see battery and USB disconnects as common causes but it's certainly good to know that it can only result in soft bricking.
Please - words mean things. Bricks only come in one way.
Which is what I do. 
