You seem to be missing that these things are not like for like exchanges. If you go buy a car you invest $25K for it, the return on the investment comes in the form of using the car not a return of the money you spent on it. A NFL contract is the same it is not a like for like exchange; the team gives the player money as an investment, the return on that comes in the form of production.
Aaron Hernandez is a sunk cost, we made an investment in him, and he is unable to recover because he is unable to provide the team with production.
You're now talking about ROI, which, again, has nothing to do with sunk costs.
Ask yourself this: can Danny Amendola's guaranteed money be recovered? If the answer is no (and it is, barring extraordinary circumstances), then it's a sunk cost. Period. There is no debate to be had here. If you think otherwise, then you are wrong and there is no further discussion to have on the matter.
Read this:
http://www.investopedia.com/terms/s/sunkcost.asp
In terms relevant to the Investopedia example, you're claiming that the cost is not sunk until the factory is closed. That's not just wrong; it completely fails to understand the question that's being asked in the first place, let alone answer it. You're answering an entirely separate question that isn't even tangentially relevant to the circumstances at hand.
Sunk costs = any cost that's already been paid out and cannot be recovered. This includes signing bonus, primarily, but also any other guaranteed payments.
Prospective costs = any cost that has not yet been paid out, and which you can avoid paying by electing to terminate the contract. This primarily includes base salary, but also any other nonguaranteed payments such as workout bonuses, future incentives, etc.
Since you're really stuck on this point, I'll reiterate it yet again: a sunk cost is not analogous with a poor ROI. An asset's potential for future returns have no bearing whatsoever on which costs are sunk and which are not. In fact, the entire sunk cost methodology is import precisely because it forces you to evaluate assets based on future returns, without allowing your judgment to be clouded by past events that you cannot change. It's simply money that has been spent and cannot be recovered. Period.
Wilfork is probably the best example of it. For the sake of simplicity, let's pretend that it's not possible to extend Wilfork and push his cap hit out into the future. The choices are to honor the last year of his contract or release him. This makes the decision a relatively simple one, since there are two possible outcomes:
Option A: You keep Wilfork, and pay $~11.6M against the cap for his services
Option B: You cut Wilfork, and pay $~3.6M against the cap for his services.
Now let's say that you look at what you can get on the open market, and decide that Wilfork's expected contribution to the team--including performance, leadership, intangibles, etc.--is worth about ~$6M, ie for $6M, you believe that you can get value on the open market equivalent to what Wilfork can give you.
An irrational actor will frequently say "well, I dunno, I already spent $3.6M on Wilfork, so it would be a waste to just cut him. I might as well keep him and honor his full contract". If that's your thought process, then you suck at being a GM and should be fired. Why? Because if your evaluation holds that you can replace Wilfork's production for $6M, then you're leaving money on the table. You're paying $11.6M, when your own analysis holds that you could cut Wilfork, take the $3.6M hit, pay $6M, and get the exact same production for $9.6M instead of $11.6M.
Similarly--and more relevant to Amendola--if you expect that Wilfork's value to you will be $9M, then cutting him on the basis of "he's not living up to his $11.6M cap hit" would be be the same fallacy taken in the other, equally idiotic direction. Why? Because to replace that production you'd have to cut him, take the $3.6M cap hit, and then spend $9M on top of that, leading to a total hit of $12.6M to get the same value that you could have had for 11.6M by just keeping Wilfork. Again, you suck at being a GM and should be fired.
Now, probably the single most important point to keep in mind here: whether or not you're
actually right in evaluating the expected value of a player's performance is beside the point. That's an entirely separate question, and while getting it right is at least as important as getting this one right, it still needs to be answered separately.
They're tangentially related in that one answer is an input to the next question, but they're otherwise unrelated. It's possible to make the wrong decision and get great returns, and it's possible to get poor returns on the right decision, in much the same way that hitting on 20 sometimes works and staying on 12 sometimes works.
In much the same way, sometimes you see that you can get $10M worth of value by finishing a widget factory for $5M. If that's the evaluation, then you made the right call by proceeding, even if some freak pipeline accident makes the factory explode days after completion (or, for a football analogy, even if the player suffer's a career-ending neck injury in training camp). In blackjack terms, you hit on hit on a 12. You made the right call, despite the fact that you ended up busting. Sometimes you just get boned by probability.
Alternately, you may elect to spend $5M finishing a widget factory that will only return $1M in profit per year, on the basis that you've already spent too much to stop now. If the alternative is to spend $5M building an entirely new gadget factory that will yield $3M per year in profit, then you've made the wrong call, period. This will remain the wrong call, even if some unforeseeable event quintuples the price of widgets, so that you're suddenly making $5M per year in profit. This is the business equivalent of hitting on 20 and getting a 1. You made a call that would screw you over 90+% of the time. The fact that it didn't this time makes you lucky, not smart.
In the Pats' case, my view is that they're
really good at acting rationally with perspective to sunk vs. prospective costs, and rationally allocating resources based on their evaluation of the future value of their assets. They're considerably less good at making accurate predictions for what those future values will be (but still pretty good, compared to the average team). But again, those are separate issues that must be handled separately.