How much of an advantage should lead to a win?

Two caveats:

1. How much of a net advantage would someone playing flawlessly (e.g. 3300elo+ chess engine, or a grandmaster performing on-point) need secured to be certain that the match will eventually end in the opponent getting checkmated?  Does it depend upon how much the match has progressed (i.e. how many moves have transpired and\or pieces have been removed and\or space gotten controlled and\or opening_vs_middle_vs_endgame)?

2. In a match between two similarly-skilled adversaries, at what point would a third party analyzer be able to make a reasonable bet (e.g. >70% favorable odds) as to which side will win (instead of concluding in a draw), assuming that the player with an edge does not make any foresaking blunders later in the match (which may have already occured, especially between two novice players)?

 

On a related query:

Human analysis often includes the annotations 

+-   White is winning.

+/- White has a significant edge.

+/= White has a small edge.

=     Equality.

=/+  Black has a small edge.

-/+  Black has a significant edge.

-+    Black is winning.

{^^ copied from the "please read" sticky, conveniently)

 

Chess engines utilize algorithms that assign specific values to each piece (such as 1 for pawns, 3 for knights, 3.2 for bishop in a pair, 3.1 for solo bishop, etc.) and weights to various positions, based on calculations down the gametree (as many plies deep that it is set-to), and with each move deliver a score in favor of one side or the other (or 0.000, which is completely neutral).  Stockfish, moving for Black, assigns Black's first move as high as +0.38 in response to dubious opening firstmove for White (e.g., 1.b4 followed by 1...e5), but mostly scores slightly negative (e.g., 1.e4 e6 computes a -0.24 for Black, at 22ply).

 

So the question is, what would a "small edge" equate to, in terms of engine analysis? What about "significant edge"?, and vice-versa.  Obviously, human beings don't typically analyze thirty-plus halfmoves deep (outside of theory..), so what would an approximate formula be?

 

You seem to forgot the most important which is that depends the position you take the advantage.

And what engines evaluate is just a number of pawn loss centimeters that actually most of the times does not give accurate evaluation. There are some examples were engines evaluate 0.0 totally equality for the next 25 moves but since you continue play the position, engine realize that this evaluation was wrong and starts renew the numbers which again don't give the exactly assessment of the position.

Most of the times...the advantage is in your anticipation about the position rather what engine suggest.

Well true, if you don't see what it is figuring thenits analysis won't be all that helpful.

 

As far as anticipating the position -- Both sides are anticipating. So.. where is the crunch?

My chess instructor from highschool taught to "play the board, not your opponent".  Obviously, if you are playing a noob, or if you have the power of invasive mental telepathy then you could use that easily to your advantage.  But outside of playing contrary to opening lines that the opponent may be adept at, high-level play doesn't assume anything onto the competition --afterall, underestimating the opponent leads to defeat more often than not.

Engines do not understand chess.

They don't "understand" it since computers do not "understand" anything.  They do, however, make the best possible move given their algorithms, which compute( which must include some level of underlying 'analysis'), at a level higher than any human player.

It doesn't matter. A GM knows the logic of his inferior to engine choice. The engine don't. If you take from the engine the opening book theory it's a useless calculator machine.

There is a video in YouTube were Carlsen beat his App " Play Magnus " finding a positional trick that engine can't understand, not in the opening, in the middle game. That shows something.

That app obviously was flawed then.  Modern high-end chess engines do pretty devent at "analyzing positions".

decent*  (at playing positionally, not just tactically)

As far as chess engines using opening book theory:  chess engines have repeatedly owned human super-grandmasters, without referencing any opening book, and continue to do so.

Let me tell you an example to understand what I mean. Komodo engine has an ELO around 3450 the highest right now between programs. If you ask Komodo to give you a evaluation about the position when starts the Berlin Wall ending it will tell you that white has 0.9 cp advantage over black which means a total pawn up. If you ask Komodo to prove this advantage over Stockfish...the answer is : I know shit!

A GM with 2800 ELO has more chances to prove this advantage than engine has.

""  A GM knows the logic of his inferior to engine choice. The engine don't. If you take from the engine the opening book theory it's a useless calculator machine.  ""

I'll grant you this.  The engine does not necessarily 'know' the logic behind its move choice, although that doesn't mean that the logic isn't there. Stockfish provides what it quantifies a 'mainline' for every move, for example.

My original questions remain.

Right.  The engines, as currently available on the market, don't do a great job at *proving* an advantage/disadvantage; this is because they do not articulate their 'thought process', let alone in full paragraphs.

I'm not hailing chess engines as perfect or anything. They don't offer the ingenuity that comes from human reasoning. But they are still useful tools which out-perform any human.

outperform when left to their own devices, which harks to the point about understanding the engine's complete logic. What tactics might it see possible, why is one future position favorable to another, etc... sometimes more obvious when it plays through, other times (especially relating to positional analysis) more obscure.

Generally, a +1.25 pawn advantage should be enough for an engine to convert the win. 

But that's the caveat, "generally". There are times when engines will consider a position completely won (+3 or more), and not recognise that the position is drawn. 

Here's an example of a position that engines consider winning (+7) but humans don't:

1.5 points is typically considered a "winning advantage."  As long as you don't get checkmated (from a suprise attack), you should win the game.

Exchange down into an endgame, and win through pawn promotion.  

Fairly Simple.  In most cases.

Conversely, a tempo is worth about 1/3 of a point.  Three aimless moves (by you), and your opponent can probably start playing for a win.

But Exceptions Abound.

I was playing a game against an FIDE IM which offcourse I lost, but was intrigued at the analysis which gave me a 1.5 advantage about 25 moves into the game, and after one seriously bad move, the advatage shifted -17. Theres an eye opener for ya! So I played the game out afterwards to see how the computer would have played out the position with the 1.5 advantage and I learned quite a few things from the game. So the adavatege theory is almost like a scale to say that either you are still in the game or that the game is lost completely. As far as being out of the woods of blowing an advantage like I did, I go with positional pawn structure for endgame possibilities, and available pieces that can still deliver mate. As long as your opponent has enough powers to deliver a mate threat, dont count your eggs before their hatched.

I was having a problem with th en caviat thingy, whatever it's called. Is it always fair to capture a pawn with a pawn when it lands next to you on it's first move?is there a twist or something?

 

1. It is called En passant

2. It is not always allowed to capture the pawn. Only when a pawn from their starting column (2nd for white, 7th for black) and the pawn lands right beside your pawn, then you can capture. 

For additional help: https://en.wikipedia.org/wiki/En_passant