Consider the following variant on Parfit's Mineshafts case. Suppose ten lives are in danger, and an agent is presented with five options (A - E), with the following information. Exactly one of options A and B will save all ten, while the other will cause all ten to die, and the agent has no way to discover which is which. Option C is guaranteed to save nine and kill one. Option D is guaranteed to save one and kill nine. Option E is guaranteed to save five and kill five.
I take it to be clear that the agent ought to opt for the safe option C, so as to save 9/10 lives. The standard concept of 'ought' (the one that's relevant to deliberation, etc.) is thus evidence-sensitive. We can stipulate a technical usage according to which the agent 'objectively ought' to choose option A (supposing that that is the one that will actually save all ten lives), but that doesn't seem as central as the evidence-sensitive 'ought'.
Let me add some further details to the case. Suppose that our agent was raised to believe that happy lives are bad whereas pain and death are good. (He finds this just as subjectively 'obvious'-seeming as we find the opposite claims.) Should this affect how we are to evaluate the agent's options? I'm inclined to think not. The agent should (in the standard, evidence-sensitive sense) still choose option C, but because he's so morally misguided he likely won't realize this. His messed up upbringing might excuse his error when he instead chooses D, but it sure doesn't justify choosing to bring about eight extra deaths.
Some philosophers - let's call them 'subjectivists' - respond differently. They say that, although the agent objectively (or with respect to the facts) ought to do A, he subjectively (or with respect to his beliefs) ought to choose D (not C), killing 9/10 people. This seems a serious cost to their view: surely we should not be attaching any positive normative status to such poor decisions. Objectivists can (and should) insist that it isn't rational to take the means to irrational ends like this.
The prospects for subjectivism look even worse when we add a further detail to the case. Let's suppose that the agent is logically, as well as morally, confused. So although he aims to cause as much death as possible, he confusedly believes that this end is better achieved by causing five deaths than by causing nine. Does this then mean that the agent rationally/subjectively ought to choose E?
Here the subjectivist faces a dilemma. If they endorse ever-increasingly subjective norms, then their view becomes empty -- it fails to accommodate the obvious datum that people can be irrational without realizing it. So suppose they instead draw a line in the sand, and insist that our logically confused fellow is irrational to choose E, and what he really ought to have chosen is D. But why stop there? Once we appreciate that people can be irrational without realizing it (maybe even blamelessly so), isn't it overwhelmingly more plausible to favour option C (saving 9/10 lives) as what any reasonable agent really ought to choose in this case?
(Note that there's a principled stopping point here at option C. The agent lacks access to the empirical evidence that could ground a reasonable belief that option A is better than C. But no such evidence is needed to appreciate fundamental moral truths, which are presumably a priori knowable if they're knowable at all. See also: Expecting Better of the Ignorantly Unreasonable.)