Membership's deep analysis is gone?

Hello Chess.com support team,

I am holding diamond membership and I like the post game deep analysis a lot (actually the major reason why I purchase membership). From yesterday, I saw that the analysis UI page has been updated to a new version. But the analysis seems to finish very fast, so it is the shallow analysis, not the deep one I used to have? I am very sure about this when I compared the lines it provides to stockfish 8. It used to be almost the same when deep analysis was available, now it is very different and a lot worse. Also I really dislike the new UI page, because it is almost impossible to use on cell phone. 

Is there a way to get back to older version of game analysis and the deep analysis? If not I will definitely not renew my membership next year. Also, I think it is not so good to cancel the deep analysis which was the reason I bought the membership. It made me feel like I paid for deep analysis and now it is gone. 

Might be something in here for you...

https://www.chess.com/blog/News/meet-the-new-analysis-game-report-retry-mistakes-more

It would be better if there is a button to switch back to older version. That would give us the freedom to try both. 

@big_data_is_cool the above quote is from another post thread on the new engine that I wrote. I am including it as I noticed you picked up that the new analysis appears to be inaccurate versus an analysis you have run comparing the two. If I may ask...what program did you use to generate your comparison analysis? Also...just for your fyi...Stockfish is at version 10 now (not that there is anything wrong with Stockfish 8...just Stockfish 10 is improved and stronger!) 

I am awaiting feedback from customer support on the major inaccuracies I found with the new engine when I started using it. Right now based on the hard evidence I have I personally don't trust the new analysis's accuracy levels.

@flashlight002 

I tried stockfish 8. The new analysis deviates notably frequently from stockfish 8 (10% ~ 20% of time when the positions are complicate). In the older version, they are almost the same. 

I also tried alpha zero using the codes from github and run on GPU. The new analysis deviates even more frequently compared to older version. 

I feel like chess.com is trying to save computation resources by degrading the analysis power while it makes the appearance "fancier" to cover the truth beneath it. 

It looks sexy...but the engine is not accurate.

All I usually want to do is scroll back to a few select positions and look at the top few most playable lines. Efficiently, without endless silicon regurgitation, not trying to become the next chess.com patzer with some illusion of grandeur as a GM candidate interested in absorbing all the mess I'm looking at now. I tried the new analysis 2 or 3 times, and said uhhh...nnnno!.........mmmm no. No thank you.

So the analysis is faster and stronger due to the servers (big strong machines) analysing the game...

As per this page:  https://www.chess.com/blog/News/meet-the-new-analysis-game-report-retry-mistakes-more

Finally, a note on the strength of the analysis this feature offers, because this is a huge difference between the new Analysis and the old. The "full" analysis that runs on your game to create the report, analysis data, and Retry Mistake positions is extremely fast. It is better than the old analysis you used to wait for, because instead of leaning on client-side/browser-based tech, the new analysis is done by a large cluster of server-side engines.

Hi @WillieVisagie

I also read the full release article that chess.com released when they launched the new analysis system explaining how it works (the article you have put as a link). The thing is I have absolute proof that the engine is doing strange and inaccurate things. I have shared it all with support department and their dev dept are looking into it. I explained higher up in this thread what happened. In summary: I played a "retry" on the "retry page of the new engine". Directly underneath the green banner (it has the word "perfect" displayed in it) that comes up when one gets the retry move correct it populated a moves variation based on my new correct move that was 15.5 moves deep. I clicked on the last move in this "suggested" variation and it immediately took me to the analysis section...with this new variation of moves now in the move list as an "alternative variation". I then began playing through this variation to see what play the engine had suggested past my now corrected blunder. As I played its moves it was analysing at the top of the analysis page its own moves it had chosen, on the fly. Imagine my surprise when it suddenly classified ITS OWN MOVES AS A BLUNDER!! And it classified an additional move it had suggested as a blunder!! So it was not just a "one time event". Of the entire variation that the engine had suggested itself (and one would expect its suggested moves to be best moves wouldn't one?) that was 15.5 moves in length reclassified 2 as blunders and three as inaccuracies. Regardless of what technology is being used this scenario just should NOT be happening! How can the engine suggest a move and then when it is re-evaluated by itself class it as a blunder?? That is ridiculous to put it mildly!! I have the screenshots documenting all of the above lf you would like to see the exact outputs by the engine and the point where it reclassified its own moves as blunders and inaccuracies.

By the way, in the article chess.com put out, and where you quote "because instead of leaning on client-side/browser-based tech....." what exactly is meant by client side/browser based tech? What "tech" is being referred to exactly? Do you know? Can you give us an explanation of this perhaps? I do fully understand the concepts of server side computational resources being used to carry out the new analysis results.

Client side/browser-based tech is your computer and your processor.  Server-side engines are many computers with lots of computing power - as they already handle millions of users they seamlessly handle a lot of analysis power.

@WillieVisagie thanks for your speedy reply. I was also interested to know if you knew perhaps how Stockfish 10 was being implemented as a client side/web browser system. On all my apps one has to have the Stockfish code loaded to run a UCI engine locally.

So I guess I am asking how did the old system work using the client side processor on one's machine, considering that the Stockfish engine has to reside somewhere! 

I fully understand the mechanisms of server side computational usage having been involved in the design and setup of big retail websites...but no chess sites...they appear waaay complicated  lol. 

@flashlight002, if you view the stockfish engine like a black box, you put the PGN in and get the results out, then it really does not matter who where or what does the actual computation, it just processes the PGN and gives an analysed game as output.  If you run Gmail in your browser it's the same - all those lovely features are computed by the servers in order to reduce internet traffic.  So instead of sending the PGN to your computer for analysis, it sends it to the servers for analysis.  Behind what you see in the browser is for a long time not just web pages anymore - they are all applications with the browser as an interface.  So the stockfish engine resides on the servers.

@WillieVisagie 

If you really like it, are you willing to pay to switch to diamond? Yeah, the new feature is so good, but if the older version does not come back I will not pay  

I am on gold, but mainly because of time constraints - I rarely reach 25 tactics a day, and really don't have time to watch videos... But I really like the new analysis thus far. 

@WillieVisagie thanks for your insights. I guess that was the crux of some of my questions. So the old system had Stockfish on the chess.com server but used our computer resources to compute the results? While the new system is completely server driven? Have I got it right basically? That however really does not explain how the new system can generate results with such patent errors. All it means is that the results one gets is delivered faster. That's great...but the new engine architecture is delivering flawed line variations and results. 

Basically yes.  I guess the errors are due to some bugs that they are sure to sort out quite soon.  Watch this space - they surely value your results.

Just weighing in without really deeply reading everyone else's comment in depth.....But NO......deep analysis has gone......it was easy, simple and effective.  You could leave it go and come back later and look through the positions that you played badly (quite a few in my case) and try to fathom some understanding of why your thinking went astray.   I really liked CAPS too.  But this is a nightmare of a system.  Too much.  Sorry chess.com,  But its yukky.

I sincerely hope so Willie. I have given screenshots, the game pgn file and a screen video now too. Support have told me my data is sitting with their team. So I hope they figure this out soon, as no matter how sexy an engine looks in terms of its GUI, it's output quality and reliability that sits right up there in terms of expected requirements. I think the new system is clever, and has clever components and learning tools, but it all means nothing if one can't trust the data. So I am holding thumbs that the dev team squash this major bug/s