<<chessgames> In what sense are they not playable?>

Let's consider this page:

D Byrne vs Fischer, 1956 [analysis]

I assume this is the output from analysis - and as you can see the SF8 suggested improvements aren't playable.

Playable as in - I want to move the board pieces to follow the computer moves.

This would require a way to jump into and out of subvariations - basic (advanced) stuff.

* * * * *

Next - maybe I haven't been paying enough attention, but the subvariations are generated exactly how?

You mention a minimum of 60-sec/ply, which seems to imply (to me) that you did a "walk-thru", i.e. allowed a 60-sec min analysis of a position, took the leading candidate, then advanced/repeated for a set of moves.

Then you allowed another 60-sec analysis in the final position, and appended the eval for the subvariation's final position.

Is this correct?

BTW- if I was spending fuel to analysis a game there might be circumstances when I would want to do "walk-throughs", rather than picking a position and churning.

This raises the issue of how to run the engine. SCID has an annotate mode, when it plays through a game at a fixed dt for each move.

There is also a finish game mode, where it allows more flexibility - allowing the user to specific either a fixed dt or a fixed depth for each move.

Are you familiar with these kind of issues?

The idea is to delete the posts, not move them. There is no reason to preserve posts that nobody will ever read. They are taking up space and space is money in this game.

Do you have a reference to GM Rogoff having no objections? That he may be nice enough not to object is beside the point. In fact, it makes it even more egregious to trash his page.

<St. Louis Rapid & Blitz (Blitz) (2017) is not viewable either?> It's problematic, yes.

Anyhow I don't think anybody questions that you're doing good work, it's a question of where this material should be placed. Perhaps one day we'll have pages for "meta tournaments" (things like the Grand Prix and the Grand Chess Tour) that can branch off into the smaller tournaments that comprise it and when that day comes you can cut your tables out of the tournament pages and insert it into the meta-pages.

It's complicated.

It all starts with a 300 millisecond / ply scan of the entire game to get the "feel" for it. If there are drastic changes in evaluation around certain moves they are flagged for later review. Just takes a minute and is surprisingly accurate as a starting point.

Next it moves through at the minimum requested (n) seconds per ply, but when it gets to any of the flagged moves, it spends 8 times longer than normal (i.e. 8n).

Now it goes back and looks and sees if any NEW drastic changes in evaluation have appeared as the result of (presumably) understanding the position more accurately. If so, it gives them the 8x treatment as well.

This loop can continue several times, especially in a situation where the incredible move is too deep for Stockfish to understand. It first thinks that move 25 is the "?" culprit, but upon deeper analysis decides that 24 is really the move that deserves the "?", then upon still further analysis thinks that the combination started after 23._? — the process can take hours especially if you set it on maximum time.

Any position that is given a "?" or a "better is" indication automatically is given an 8x evaluation, as it's important to know the evaluation of the position after the inferior move was made. Oh, and the final position (assuming it's not checkmate) is always given extra time, since people always want to know about that.

Finally when there are no more "mysteries" in the eyes of SF, it releases the annotated game.

Because of this, we're already coming up with a scheme in which these behemoth forums can be archived up to some certain point in time (or a certain page number).

We have no plans on deleting posts but if they were placed in some archive available only for those who wanted to search them, I have a hunch they wouldn't get much use anyhow.

Moreover, it would make the server faster. It's embarrassing to say how much fo the server is consumed by the routine algorithm that figures out what posts to show somebody every time they refresh one of these monster pages. Not just the "off topic" pages, but things like Carlsen and So etc., they all have become an incredible drain over the years. Our hardware gets better but the data we serve gets bigger still.

http://english.vietnamnet.vn/fms/sp...

I suppose that one thing that is really degrading the performance is providing the correct page count. Simplify the algorithm: count the pages without applying the ignored list, and apply the filtering only when rendering the page. Some pages may end up sparsely populated or even blank, but this way you'll never have to fetch more posts in one try than 25. Additional benefit: page numbers will be identical for everyone, not like it is now that the numbering depends on the ignore list.

If having sparsely populated pages is not good, then consider getting rid of the "Jump to page#" feature. I imagine that this one does put a lot of load on the database, if used, but I'd venture a guess that it is used rather infrequently. Most people probably click the Earlier/Later links, which can be handled by fetching the candidate posts, filtering out the ignored ones, then fetching more until there is enough to fill the page. Once you know the last post fetched, the approximate page number could be displayed.

No, don't just delete them. Burn them. Yes, yes, burn them all!

Burn the posts alive at the stake!

https://antifilmschoolsite.files.wo...

https://vimeo.com/43315666

(Gheesh, the music is so great, yet the vimeo video so awful!)

Let's watch me in action:

https://www.youtube.com/watch?v=3U5... (short version)

https://vimeo.com/190315596 (long version, beware vimeo)

https://www.npr.org/2017/06/15/5329...

I agree with this solution offered by <offramp>: