Perfect play does NOT result in a draw

Now we have solid proof that black can play flawlessly and still lose. Here's the game which shows just that:

As you can see, black played flawlessly. Every move was excellent or best. In fact, white was the one who played one move that was "good", resulting in white's CAPS being 99.54, and black's CAPS being 99.55 - one hundredth of a point better. 

After move 34 white has an advantage of 2.55. This means that he will SURELY the game. I stopped at this point because that is enough of an advantage to prove that white will win if the game continued at the highest level. 

Perhaps the thesis you've proven is: "Computers are dumb".

You have a very loose definition of "flawlessly!" 

How the fuck can you say that only allowing every single move to be EXCELLENT and resulting in a CAPS of 99.55 is a "very loose" definition? 

Ig your computer is claiming that every move was "EXCELLENT," then there is something is wrong with your computer analysis, Look at that game with a decent engine and you will see several inaccurate moves by both sides. 6...b6 for instance is a move you will never (< 1%) find in master play . You are building a theory on a weak computer doing superficial analysis. Why bother?

I used Stockfish 9, which is currently the strongest chess engine in the world. I guess I forgot to mention that. 

This is actually a game I set up where both sides are Stockfish 9. The rating for that engine is about 3438, well higher than any human player today. 

It just seems very odd that the game resulted in a loss (well, it was heading in that direction that is)

How many ply search-depth? Stockfish 9 at a dozen or so ply is incredibly stupid.

Engines lose to other engines all the time, this is not news.

They're typically bad at evaluating these types of middlegames too.

Also as @blueemu suggests, I have some doubt that you set it up well, because anyone who is surprised an engine can lose against itself is liable to have done anything

How many ply search-depth? What do you mean?? 

Well I guess the first question is, did you download stockfish, or are you running it off a website?

If you downloaded it, what interface are you using to run it and what's your hardware?

But anyway, even in something like TCEC engines lose to other engines all the time.

And alpha zero was marginally better than stockfish, so stockfish doesn't play perfectly.

OK, if you don't know what search-depth is, then the chance that your experiment has any significance is pretty close to zero.

Search-depth (or ply) is the number of half-moves the computer looks ahead. To pick a trivial example, after 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 a computer that only looked one ply ahead would think that 4. Bxc6 wins a piece, because the recapture is two half-moves ahead and therefore beyond its horizon.

You quoted Stockfish's rating as over 3400. This is only true when it is running on hardware that allows 128+ threads, and when it is set to a search depth of over 30 ply (closer to 36 as I recall).

Neither of those conditions will be true if you are just running it on your home computer.

I'm running it off a website 

Then the website should tell you how many threads and what the search depth is. If they don't, then you'd better assume that it's just a cheap (and low-power) set-up.

"Now we have solid proof..."

I quit reading after this...

Websites tend to donate very little hardware to it and/or limit its search depth. If you want "full strength" (so to speak) stockfish then it's better to download it to your PC. You can download it (for free) and an interface (also for free) here:

https://stockfishchess.org/download/
http://www.playwitharena.com/

When I have a question about a position's evaluation and I want it to suggest a solid move, I ignore anything below ~25 ply and ~100M nodes. This might take my computer 20 seconds or so. If the position is something hard for the engine to understand, like a KID then I give it closer to 1 minute of thinking time... but even then it can be blind to better moves.

Your research is just awesome, a turning point in chess history

I thought Lyudmil's book did that?

Come on man...Keep up!  These are some great chess times we are living in.  Just think of it...

We have a online 1300 player that has busted the Sicilian Defense.

We have a OTB 2100 player that is stronger that any chess engine, and GM (except for the WC)

And now we have an online 1800 player that has solved chess.

Just think...you're going to be able to tell your grandchildren you were around during all of this.  This is like being there when man walked on the moon.

You're living in historic times!