4NCL Online

[MartinCarpenter]

Although you could check to see how much cheating of that sort there might have been - if much more theory hitting the board on average vs an over the board 4NCL weekend then....

[John McKenna]

On soon tonite - Division 1 Round 9 (of 10 prelim?) pairings -

11 Alba - - - Anglian Avengers 1
12 Atticus A - - - Sussex Social Isolators
13 Battersea Horses - - - Blackthorne Russia
14 Celtic Tigers - - - Broadland Kestrels
15 ChessPlus Alpha - - - Gonzaga
16 Guildford Young Guns - - - Arkell's Angels
17 Kent KJCA Kestrels A - - - Bon Accord A
18 King's Head - - - Grischuk Grizzlies
19 Northumbria Vikings - - - Wood Green Monarchs
20 Oxford 1 - - - Sussex Auguries
21 Shropshire 1 - - - Spirit of Atticus
22 Surbiton 1 - - - Watford 1
23 Warwick University A - - - Barbican 4NCL
24 West is Best 1 - - - Gonzaga B Sharps
25 Wimbledon 1 - - - Dundee
26 Wood Green - - - Chessable White Rose 1

Spot the forumites playing in this, and the other divisions, via the 4NCL website -

http://www.4ncl.co.uk/

[Paul McKeown]

if the computer analysis of the games is sufficiently (5 sigma? 10 sigma?) away from 'typical', then the ban is automatically triggered

There is nothing known to humankind by inductive reasoning within 10 sigma bounds. I tried to calculate the error function for 10 sigma, but my calculator blew a fuse. I gave up and used *oogle: http://www.aleph.se/andart/archives/200 ... nance.html.

5 sigma is a huge hurdle to cross in experimental physics, with enormous levels of automation applied to both experiment and to analysis. Some manufacturing processes claim to be "6 sigma", whether that is marketing bluff or reality is open to question. Many would suspect the triumph of hope over experience. To get to 10 sigma, you had better just ask the divine, you'll be taking observations for a significant part of the lifetime of the universe.

If some server claims to know if a player has cheated to 5 sigma limits, you can expect reasonably that to be their management putting a positive gloss on their analysis. If 5 sigma was applied to judicial proceedings, then you wouldn't convict anyone.

[John McKenna]

UK Gov. would not agree - they are planning on ramping their propaganda sigma up to 11, if necessary, before the end of days.

[NickFaulks]

There is nothing known to humankind by inductive reasoning within 10 sigma bounds.

Winning any lottery with over 100 ticket purchasers and only one winner is a 10 sigma event.

[Paul McKeown]

What inductive reasoning are you claiming is involved in winning the lottery jackpot?

I thought that the odds on winning the UK Lotto were 1 in 45 million, by the way, whereas the value of the error function at 10 sigma was on the order of 10 to the minus 23. Perhaps I am wrong.

[Paul McKeown]

To be clear here, are you claiming that someone knows, to within a set confidence limit, by a process of induction what the winning numbers for the Lotto will be or are or were on a particular day based on some theory of the Lotto, without consulting the published results or results published by a third party with access to the published results?

[Joseph Conlon]

if the computer analysis of the games is sufficiently (5 sigma? 10 sigma?) away from 'typical', then the ban is automatically triggered

There is nothing known to humankind by inductive reasoning within 10 sigma bounds. I tried to calculate the error function for 10 sigma, but my calculator blew a fuse. I gave up and used *oogle: http://www.aleph.se/andart/archives/200 ... nance.html.

5 sigma is a huge hurdle to cross in experimental physics, with enormous levels of automation applied to both experiment and to analysis. Some manufacturing processes claim to be "6 sigma", whether that is marketing bluff or reality is open to question. Many would suspect the triumph of hope over experience. To get to 10 sigma, you had better just ask the divine, you'll be taking observations for a significant part of the lifetime of the universe.

If some server claims to know if a player has cheated to 5 sigma limits, you can expect reasonably that to be their management putting a positive gloss on their analysis. If 5 sigma was applied to judicial proceedings, then you wouldn't convict anyone.

My meaning is different - what '10 sigma' means here is, if I assume the null hypothesis (which in this case would be 'the moves are those played by a single human player without any outside assistance'), then the frequency that the observed games would occur is more than 10 sigma outside the centre of the normal distribution. It is not a statement, as such, that the player cheated.

In the context of a null hypothesis 10 sigma is not that out of the ordinary - if my null hypothesis is 'This coin is biased and will land 75% heads.', I can disprove it at 10 sigma by approx 1,200 coin tosses.

Likewise with chess although the statistics are not as clean - if I have a sample of 40 games, twenty of which are Carlsen-Caruana and twenty of which are drawn from the local U7 tournament, I would expect these to be 10 sigma incompatible with the hypothesis that the same opponents at the same ages were involved in all games.

[NickFaulks]

My point was much simpler than any of this.

If 101 people each pay £1 for a lottery ticket, winner takes all, then their profit is distributed with mean = 0, standard deviation = approx £10. The winner gains £100, which is a 10 sigma outcome.

Nothing to get excited about.

[Joseph Conlon]

My point was much simpler than any of this.

If 101 people each pay £1 for a lottery ticket, winner takes all, then their profit is distributed with mean = 0, standard deviation = approx £10. The winner gains £100, which is a 10 sigma outcome.

Sorry, but it's not - the resulting profit isn't a normal distribution.

[NickFaulks]

Who is saying anything about a normal distribution?

When you plan for 6 sigma events you are recognising that in real life things are not normally distributed. That's the whole point.

[Joseph Conlon]

If you call it a 6 sigma event, or a 10 sigma event, you are using the language of the normal distribution.

Indeed many things do not follow normal distributions, but there is still the notion of events that are extremely uncommon (e.g. a Richter scale 10 earthquake). One can use p-values or any other methodology to make this more formal.

[Roger de Coverly]

It is not a statement, as such, that the player cheated.

What the Regan software is identifying is where there's a greater match between the player's moves and the moves recommended by an engine than might be expected based on the player's rating.

There are a number of conclusions that could be drawn from that observation, not least that the rating is incorrect. Only some of them being that the player was using outside assistance during the game and that was against the rules of the competition.

[Michael Farthing]

If you call it a 6 sigma event, or a 10 sigma event, you are using the language of the normal distribution.

Not quite. Standard deviation is calculable for any distribution and would still be indicated by a sigma. What is different is the percentage of the population that falls in particular bands. Thus, in a normal distribution about 95% of the population will be within two standard deviations of the mean. That would not be the case for other distributions. However, in this discussion I accept that a normal distribution has been (heroically) assumed.

Actually, given that the exercise is to some extent about assessing the quality of a chess game, if some such assessment were possible and all games played included (in the playground and on train journeys) it would be much more likely to be a poisson distribution - an initial bulge of poor games tapering quickly to a long tail of increasing quality. As in Nick's example, such a distribution might well allow ten standard deviations to be realistically acheivable. [Indeed - that is probably what a grading list produces].

[Roger de Coverly]

[Indeed - that is probably what a grading list produces].

Whenever a grading list is produced in graphical form, it invariably comes out looking Normal. The ECF one of Clarke grades peaks at around the 135 to 140 area. Whilst it's plausible for the upper end to be Normal or have a strong resemblance, an explanation of the left hand side is that extremely weak players don't play in graded or rated chess. Thus there's a minimum standard which enough participating players exceed to get a Normal shape.