...If reliability was getting better, the ESA would provide a profit for Tesla. Something smells bad....
That's right.
I spent most of my career running different parts of what became a really big software and hardware services business for a big name global company you'd recognize. The software and hardware business units who were incented to develop and sell product, wanted us to deliver post-sales support as cheap as we could since our warranty recovery costs went back against the cost of their product; OTOH our mission to the shareholder was also to generate profit from our "extended warranties" and service agreements by creating T&C that balanced our risk (it's similar to what insurance companies do setting rates) with what customers would pay (yes, we needed to maximize profit) while ensuring we delivered the best customer sat so customers would renew their agreement with us or in the bigger corporate picture, buy the next product from our company that hopefully started the cycle over again. It was a huge juggling act serving both masters so-to-speak, but IMHO probably the right way to do it if you care about product sales AND maximizing longer-term corporate profitability. The first couple years when I was starting up the software support business, we didn't have a lot of stats on the true costs to support each individual product, so we had to take some large swags at initial pricing and how we bundled products into logical sets for both the customers and ourselves -- fortunately I had 100's of products my team supported, so while some cost me an arm and a leg and I lost my shorts on them until our team could get products fixed back with those product divisions to reduce future failures that cost me more money than we collected to support the customers, others were good money making up for it. It took me several years to push prices up in certain problem areas, while maintaining or even reducing prices in others based on my costs and customer demographics in terms of what they were and were not buying -- again, trying to balance new product sales with loyalty and the customers that wanted to hang onto older products and just keep getting support on them seemingly forever (which we eventually priced long-term support in a way to help them justify buying the new product vs paying for really-expensive-for-us-to-provide support on something that had only a few customers demanding it).
How much of that can you draw parallels to in this thread? 
Tesla only has 3 products thus far, and what, 100,000 units to deal with ...and a much smaller number than that, that have really purchased, or will purchase, an extended warranty or service agreement. It's got to have been even harder for Tesla to have had early confidence when that first Roadster and MS hit the streets estimating actual support and development costs they have now seen and can project from, or try to balance good and bad profitability across 3 products -- especially with the very optimistic and public, often times very early stake-in-the-ground approach Elon has taken (and part of why so many of us appreciate what he's trying to accomplish, yet his team probably hates it when he Tweets something out-of-the-blue, having to then try and meet his perhaps unrealistic objective).
If you think about how Tesla is operating, again, they only really care about selling new cars. Nothing else matters to them at present, but as I've suggested in other threads, from my experience that strategy will eventually catch up with them if Customer Sat goes down with existing buyers, it will ripple causing less future sales... Not everyone buying a M3, or even future MS and MX as the volume expands, is an early enthusiast ready to risk $100K on a dream. We'll have to see if the rumblings of even some of the early adopters here on TMC expressing concern today over a number of Tesla practices or deficiencies is perhaps an early alarm supporting my suspicion. I only hope Tesla is listening somewhere and not operating only in a futuristic (and what IMHO would be an unrealistic long-term) vacuum.
FWIW, my general approach with terms was, if a customer paid per-incident after warranty, it cost an awful lot of money and was via T&C on a best-efforts basis in-part because I could not guarantee I'd have parts or developers that could open up old code somewhere in the world for everything we ever made, nor could I plan for it and guarantee service levels, except in very broad terms -- and as we told all customers, we wanted them to lock in a longer term agreement that allowed me to staff and set up infrastructure to support them the way we really should; if a customer paid month-to-month, terms could change month-to-month as could price but it was almost always better for the customer than going per-incident; if a customer wanted to lock in a price, I would do that for a limited number of years (based in-part on product age), lock in terms for that same duration, and generally give them some sort of discount -- sometimes as low as just the cost of money was to me.
