What are the main arguments brought forth in favor and against the unmediated nature of consciousness?
Chalmers argues knowledge of consciousness is unmediated in his book The Conscious Mind, chapeter 5, section 5 The Argument from Self-Knowledge:
On a reliabilist theory, beliefs about a subject matter are justified if they are formed by a reliable process; that is, if they are formed by a process that tends to produce true beliefs. Perceptual beliefs, for example, are justified if they come about via optical stimulation from objects in the environment
The trouble is that if our beliefs about consciousness were justified only by a reliable connection, then we could not be certain that we are conscious. The mere existence of a reliable connection cannot deliver certainty, for we have no way to rule out the possibility that the reliable connection is absent and that there is no consciousness at the other end. The only way to be sure here would be to have some further access to the other end of the connection; but that would be to say that we have some further basis to our knowledge of consciousness. This situation is often deemed acceptable for our knowledge of the external world: we do not need to be certain that chairs exist in order to know (in an everyday sense) that chairs exist, so it is not a problem that we are not certain that there is a reliable connection between chairs and our judgments about chairs. But we are certain that we are conscious.
Beliefs justified only by a reliable connection are always compatible with the existence of skeptical hypotheses. These concern scenarios where things seem exactly the same to a subject but in which the beliefs are false, because the reliable connection does not hold. In the case of perceptual knowledge, for example, one can construct a case in which the reliable connection is absent—a case where the subject is a brain in the vat, say—and everything will still seem the same to the subject. Nothing about a subject's core epistemic situation rules this scenario out. But in the case of consciousness, one cannot construct these skeptical hypotheses. Our core epistemic situation already includes our conscious experience. There is no situation in which everything seems just the same to us but in which we are not conscious, as our conscious experience is (at least partly) constitutive of the way things seem.
First, one from communication theory:
If the encoding delay seen experimentally is real, then experience is always retrieved from memory. That is a medium, and the process is mediated by it. The nature of what we can and cannot experience is restricted by the medium on which it is recorded. Not to go entirely in the Marshall McLuhan direction, but the medium of presentation hugely shapes the kinds of messages than can be faithfully borne.
I think the 'arrow of time' is a restriction of that medium, as I have argued elsewhere: https://philosophy.stackexchange.com/a/18190/9166. So are concepts like color and the continuity of shape.
To argue that experience is unmediated leads us to consider it more reliable than it really is. Experience is the frame from which we must start, but it is not just a representation of what we are immersed it, it is heavily pre-processed and strongly biased by the nature of our embedding.
Second, one familiar mostly to educators and workers in computing:
We are constantly confronted with things that we have learned to experience differently. For example, we more seldom experience ourselves moving backward when something moves forward beside us, unless we are otherwise lost in thought or processing things poorly for other reasons. The original experience we used to have as children just disappears most of the time. Our learning intervenes before that experience happens, now that we are more sure of the world. If nothing intervened before information became experience except some mechanism of perception, that would not happen: We would know the original experience, even if we also knew the corrected one. If experience is unmediated, how can we learn to do it better?
One may object this is still about perception, about the outside world. But our internal processes are altered by common repetition just as strongly. Consider the experience of 'flow'. If we do algebra all day, we do it automatically, and it evaporates from our experience. When we really know how to program, it feels like speaking, and not like the intricate planning and balancing of factors that it really is. So the experience is 'cleaned up' before we have it.
When we want access to the gritty details of our own stream of consciousness, they are actually quite hard to get at. Any good systems analyst has to notice that most high-expertise professions contain absolutely false characterizations of how the work itself is done. Novices know this because it can present a quite high barrier to entry, not on purpose, but because experts come to experience their own internal worlds differently and become alienated from the more natural state. Automating a field where such over-processed notions of how one actually thinks are prevalent is painful, and creates the industry-wide impression that 'clients are morons' who do not even know objectively how they spend their days.
The following may not be strictly relevant to what you want to ask, but ...
There is a time lag (~100ms) between an event and you becoming aware of it, this allows the brain to associate data from different processing channels each with its own lag into a single experience. Most of the time you do not notice it but it is there, and on occasion will even give rise to some apparently pretty odd experiences. See David Eagleman's essay Brain Time.
As I understand it you may even get a reversal of experienced temporal order (but that is my reading of the sources, yours may vary) ... didn't I just say that...