1. For everything we learn, feel or think when we are reading a narrative we can and should find a reason in the language of the narrative itself. That’s why learning about narrative technique is learning about the way language works. The more you know about the way language works, the better you are able to explain what you feel and how you ended feeling it. Is that enough? No. You also need to understand how poetry, or narrative or drama, for example, are constructed observing some rhetorical formulas which differentiate them from each other as literary genres. That’s why you need to read works like Aristotle’s Poetics and Wayne C. Booth’s The Rhetoric of Fiction.
2. Reliability is always related to the author’s linguistic expertise and literary aptitude first of all. If the work we are reading passes that preliminary test there are still many other tests ahead. As a rule we do not tend to question the reliability of the narrator in a third person narrative. If we question the reliability of the narrator in a third person narrative, reading the story he is telling becomes very problematic or impossible. The narrator of Machado de Assis’ "A woman’s arms" is not himself a character in the history he tells us. He is nobody. He is a voice, an “employee” of the author. He has a particular task to fulfill and beyond that he doesn’t exist. This particular existence and behavior of the narrator as the “voice” who tells the story do not surprise us because they are the consequence of a well-known convention that we accept without much questioning it (it’s a rhetorical device). Nothing of what the narrator says in "A woman’s arms" reveals him much as a real person, as a real character. For that reason we do not have particular reasons to question the truthfulness of what he says. He may have an opinion about the events and the characters in the story, he surely has his own way of interpreting what he sees, and he may let us know in a more or less clear way what he thinks about the character’s personality and behavior. But as a rule he does not go far enough on that direction to make us doubt the trustworthiness and consistency of what he says. Otherwise we would stop reading the story in disbelief.
3. It is true that we could submit everything the narrator says when he is telling a story without being a character to a thoroughly analysis and we could question the way he presents the events and the characters. That’s something we can do about any narrator, about any speaker. But as long as a perspective of common sense does not question the reliability of what is told we don’t stop reading, the illusion of reality is preserved.
4. In Machado de Assis’ "Midnight Mass" the narrator is himself one of the two main characters of the story. He is older now than when the events he is telling us took place (how much older exactly is not clear). The story is about something that happened to him when he was a young boy. He says that he didn’t understand the conversation he had that night with a lady when he was waiting for a friend to go to the mass of midnight. We are allowed to think however that he is not (anymore at least) so naive as he pretends to be and in consequence his reliability has to be questioned. But his diminished reliability is not the result of any defective ability of the author nor is it a weakness of the narrative. It is, on the contrary, an important element of the technique used by the author to build the story. We could go farther and say: everything he says seems to point to the attraction that the married woman and himself felt that night for each other. He never clearly utters it though. What he does is to give us a description of his behavior and of the behavior of the lady that allows us to come to such conclusion. Now we could say that he was the one who was disturbed and that’s why he is telling us the story. When he mentions at the end of the short story that after her husband died the lady married a man working for him the narrator may be in some disguised way insinuating that the lady was already ready to start an affair even before her husband died. But the only reason we may have to think that the lady herself behaved in a suspicious way - the only reason we may have to think that she was disturbed during the conversation – is given to us by the narrator himself and by the technique he uses to tell the story. Does this factor diminish the interest of the narrative and the reliability of the narrator to the point of making the story less interesting and believable? I think we all would answer: NO.
5. You should be able to understand from now on how much the reliability in a narrative is closely related to the illusion of reality (verisimilitude). In "The Horla" by Guy de Maupassant we may question the mental health of the narrator. But in doing so we are just reading the story as the author and the narrator want us to read it: it is the narrator himself that is questioning his own mental health first of all and in understanding it we are not behaving in contradiction with the way the author expects us to behave. The narrator’s insanity - and our disbelief in what regards the narrator’s report of the events disturbing him – is an important factor, the most important one, of the meaning of the story. Reading the story is to learn little by little that the narrator is falling into madness. His unreliability in what concerns the amazing events he describes becomes progressively indisputable for us. It may seem for a moment that he could understand himself that he is falling into madness. But it doesn’t happen. We cannot fail to see however that he is the one who, being a candid first person narrator, gives us the reasons to understand that what he is saying doesn't make sense. I guess that that is the reason why we go on reading the narrative until the end, interested in the development of the mental state of the narrator, in the incoherent plot that he is methodically developing, and in what may happen in consequence of that. His madness and his behavior as a man whose relation to reality is seriously disturbed became the main subject of the narrative, the one we are interested in witnessing. The "illusion of reality" and our interest are preserved because the story is about the madness of the narrator as told by himself. “Reliability” is not a concept related to truth exactly or in the first place; it is a concept closely related to the “illusion of reality”, that something that makes a narrative believable, readable and interesting.
6. The question of reliability is one of the most complex ones in narrative. The author constantly manipulates reliability because what we think or believe about what the narrator or the characters do or say or see plays an important role in the construction of the story - and it all depending on how much we trust the narrator or the characters.
Reliability is not fixed forever either, we may find in any narrative reasons to rely or not to rely, to rely more or less in what is being shown or said. As the narrative progresses the characters acquire a more precise and complex dimension, they may also change, and since the meaning of the plot is intimately dependent on what we think of the characters we are in need of constantly proceeding to small or important adjustments and readjustments.
Reliability in narrative is not an ethical question - at least in the first place. It has only to do with what we are supposed to really think of what we see or hear. Manipulation in this case is not first of all an ethical question either: it is just a tool in the construction of the meaning of the story. A narrative may become more interesting when the narrator is good at playing with reliability in an intelligent and complex manner.
In "Le H´orla" by Guy de Maupassant the narrator shows interest in what other persons may think of what is happening. But despite his interest in confronting his conclusions with other people's perspective he is unable to escape his hallucination, what for him is reality and not fiction. Maupassant however takes advantage of this amazing behavior of the narrator to include reflections, comments and interrogations which show his own interest in more or less scientific debates going on at that time about the role of the unconscious.
You should read Wayne C. Booth's The Rhetoric of Fiction. Booth deals with the problem at different occasions and from different perspectives. Booth says that " a fact, when it has been given to us by the author or his unequivocal spokesman is a very different thing from the same fact when given to us by a fallible character in the story. When a character speaks realistically within the drama, the convention of absolute reliability has been destroyed, and while the gain for some fictional purposes are undeniable, the costs are undeniable too. " In part III, Chapter XII, Booth writes extensively about Henry James and the unreliable narrator"
7. Going back to our discussion about the reliability of the narrator in "Le Horla" - apparently some of you and me disagreed on the issue of increasing reliability due to the fact that he shows some serious concern at some point in what other people may think, or know, about the events that are disturbing his mind, and mentions it in his journal - I think we need to look at that detail from two different but complementary perspectives that would respect and combine both my comments and yours:
7a. Quoting another person's perspective points to his doubts about the reliability of his own senses in the evaluation of his perception of reality. If he didn't doubt he would not feel any need of taking in consideration another person's perception of reality or what science is saying about these amazing phenomena. It seems to me that he becomes more reliable in his effort to come to an honest understanding of what's going on.
N. B. We may feel here the author's interest in irrational phenomena and trying to make the case for his story - and I don't think we are wrong in doing that. But it's another aspect of the story that does not diminish the interest of what we are discussing.
7b. You are not wrong when you feel that he does not become more reliable just because he seems interested in knowing what another person may think of what is happening to him. His reliability may have increased in some way because he doubts and listens to a second opinion. But reliability does not increase to the point of making us believe that what he sees is what is happening. In other words: his progression into madness ends by revealing itself strongly even in the way he deals with other person's (and science's) version of what happens. For reasons not explained but which have to do with his mental health he cannot escape his hallucination.
7c. I guess that this is what characterizes what we identify and describe as madness: any new element, instead of opening the door to normal or objective reality, ends up by being used by the person who is going nuts as another element corroborating his obsession. This story is also about how we give meaning to reality: you (I) put
together all the elements (invested of a particular meaning) that in your (my) opinion are relevant in every situation and then you (I) see/establish between those elements full of meaning relationships that in the case of madness contradict causality as understood by a person considered "normal".
The topic is complex. We may see "The Horla" as showing a particular and modern awareness of the complexity of inner or mental life and of human experience in general.