To mention combat in brief, what I was formulating was the concept of expressing an idea of action, rather than mathematical one. [...]
This is something I really enjoy in some of the P&P systems I've played, and something I try to push toward in games I'm DMing. My one of my personal favourites was putting the players up against a mad wizard in zero gravity. Even though all actions had at least one arbitrary roll attached to them, it really encouraged a good deal of thought of choice and consequence beyond "I walk over to him and swing my sword."
There was also a great deal of ambiguity as far as how much the players could measure the success of their actions in advance. As it started, they we're all asking "can I do this?" and I'd respond "try it."
Aside from that, I put the same group up against steam tanks, and once again, the onus was on creative problem solving. However, the biggest factor in these situations being playable, was firstly, the fact that a human DM is in control to make appropriate judgments, assess difficulty of player actions, and assign appropriate skill checks.
Second of all, the campaign was pretty obscure, and actually relied on player death being a frequent thing. It was the ultimate munchkin campaign, where players and NPCs alike were all completely over the top, phat loot was in constant supply, and each death caused physical mutation and mental disorders.
But these two things serve to highlight a couple of issues with ambiguous player actions. In the case where the player doesn't make the dash to cover, and the enemy is ready for him, is he fucked? The acceptable margin for error has to be broaden in parallel to the degree of ambiguity. In a tight turn based tactical game like X-Com, a bad move would mean your squaddie is fucked, but in the scheme of things, most squaddies are little more than pawns, and so death doesn't have to be a game ending experience.
In an adventure-ish game like you propose, it's my character that's in the line of fire, and so you need to provide leeway somewhere. An interesting idea to overcome the cycle of save->fail->load->repeat, is the idea of a rewind button, so each turn is groundhog day! That way, you can be as unforgiving as you like, as long as a situation isn't utterly impossible. Repeatable actions, like shooting 4 times in a turn, and rewinding until all four shots hit should be covered by using the same random seed, to encourage the player to make different choices.
Or, alternately, you take the approach of being more forgiving, to encourage (hell, just allow) the player to continue despite a "wrong" move. If you wanted to be smart about it, you could calculate enemy results according to
desired player actions. So, if your player wants to try the mad dash, and doesn't make it, the computer is a little more lenient.
It's a precarious balance, because on one hand, you have frustration, and on the other, a reasonable challenge. When you introduce more ambiguity, you present a wider range of seemingly valid choices, so you have to account for the player making them.
To use your feedback example with FPSs, the player doesn't need to know the exact numbers when it comes to weapons, or monster health, because they rely on their own skill and dexterity to avoid damage, and inflict damage, so as long as they're able to surmount the challenges the game offers, it's largely irrelevant.
But imagine taking away the crosshair from a FPS. Now you've added an ambiguity that actively makes the player less effective, despite their skill. You can overcome this by either reducing the degree of ambiguity (ie Doom doesn't require a crosshair, since accuracy is more or less determined by the X axis) and/or, by fudging the shot within a margin of error (auto aim).
So basically, as long as your combat system provides challenge without frustration, then ambiguity is fine. To build on your example:
The player decides they want to make a mad dash from cover to cover. Let's say making it 75% of the way is deemed "close enough" in accordance with an appropriate margin for error, and fudge accordingly. A player capable of making it 100% of the way makes it with no penalty. A player who would almost make it, runs most of the way, and dives at the last second out of desperation, making it, but starts their next turn lying on the ground, so at the very least, they have to spend some of their turn getting up again, or maybe the landing knocks the wind out of them, and they need to spend time recovering. Maybe it jolts their gun out of their, or helmet off their head, and so they have to take the time to pick them up again.
Of course there's also potential for the character to make an evaluation, regardless of the player's input. Just don't be completely totalitarian about it. The player should be able to make really stupid decisions post warning.
Player: <click>
Character: Er, there's no fucking way I'm going to make that...
Player <click again>
Character: ...but fuck it, you only live once.
What I'm thinking about is doing away with stat-sheet altogether, be it verbal or numerical. That is, the player (not the character) will see only the effects and symptoms, not casues and diagnosis.
The second hardest is probably making the symptomatic display of your character, the analysis of his bodily and mental conditions, but without diggin in his brain and extracting all the knowledge he has.
I think it's reasonable to have a record of the character's self perceptions. The more you do something, the more accurate it becomes. It would certainly be interesting if it was calculated according to a limited perception too.
Imagine your character has just spent a whole day cutting up giant rats with a sword, and that posed no challenge to him. A smart character realises that rats are hardly a reasonable test, and so his assessment is positive, but not overly so. A dumb or self-absorbed character believes that he's a fucking hero, and a master swordsman to boot.
Just think of character stats as the same thing as journal entries/dialogue trails and what not.
The first hardest thing here is to make a stat-improving system properly, and that's a problem in many games, even with standart numerical stat-system.
I think that perhaps a goal based, skills-increase-with-use system would work well, especially if it's abstracted well. Think Betrayal at Krondor's system of tagging skills to focus on improving, but not done so specifically. Maybe part of it is affected by character interactions, like the typical:
NPC: You look like you can handle yourself, can you swing a sword?
1. That I can, in fact, someday I aspire to be the best swordsman in the land.
2. I prefer the slightly clumsier art of swinging a mace, but it gets the job done.
3. Actually, I prefer to fight my battles with words.
etc.
So you're introducing reactive roleplay. Rather than using numbers to define a character, you're using their own choices during the course of the game, which helps to reinforce a game response to RP choices.
In any case, this an interesting discussion, and beats the fuck out of over-analysing another identical Oblivion preview