Fudging is a necessary evil, but tell your players you don't!
In my humble opinion, fudging is a (very, very occasional) necessary evil. I try and avoid it. As a DM, I have told my players that I don't do it.
RPGs are a combination of collaborative storytelling with added chance, provided by the dice, in order to send the developing story off in unexpected directions. Further, the dice are dispassionate, and help absolve the DM of personal responsibility for unfavourable outcomes.
On this basis, the DM's role is to provide a series of challenges for the characters, some of which they will succeed at most of the time because they are proficient in handling the challenge, and some of which they will mostly fail at because they are unequal to the challenge.
An RPG is a series of carefully considered gambling opportunities for the players.
In a situation which is dangerous, or even a matter of life and death, they must therefore consider if they wish to take the gamble and risk the lows of failure or the highs of success. That is fundamentally where the excitement comes from.
If the DM fudged ALL his rolls, everyone might as well discard the rulebooks and just agree to collaborate on a story (and there's nothing wrong with that, in fact I recommend it for a novel experience).
If the DM fudges SOME of his rolls, he must choose what story events to 'fix', and this gives a muddy experience: why choose to reject a result which kills a character, for example, but not one which allows a character to succeed spectacularly? The DM is in danger of being held personally responsible for unfavourable outcomes, and players may make riskier choices expecting to be let off.
If the DM fudges NONE of his rolls, he must provide carefully considered challenges. More importantly, the players must carefully consider whether they wish to engage with the risks presented to them (such as by gathering as much information as they can about the risk before they charge in).
Of course, designing encounters remains an inexact science, which:
A) places even more onus on the players to gather intel (even an encounter designed to be 'fair' can throw up unpleasant surprises)
B) might occasionally mean that the players should cut their losses and run away from a challenge
And, of course, the DM is entitled to design challenges from which the players SHOULD run away, or that are easy to beat. If all encounters were moderate gambles, the players would eventually learn that there's limited need for intel, since they already have a premonition of their odds of success.
So when do I fudge? My DM style is quite cinematic, and I am particularly sensitive to maintaining the right pace and mood for each scene in a session. I fudge when the dice cause an event which is utterly disastrous to the momentum of the ongoing session.
I once DM'd a new player, whose play style was to charge in to combat before I'd even finished describing the scene. I soon realised that this wasn't just roleplaying his character. He was playing a kind of meta game: daring me to kill his character and seem cruel, or relent and ameliorate the encounter so that he won out. I kept my encounters exactly at the risk level I had designed them, from the easy intro fight, to the 'a smart guy knows when to run away to fight another day' encounter. Our collaborative story that night turned out to be about the rash fool who ran into every fight and died.
