Hi,

I've been using your JavaScript implementation to compare to a Java one on which I've been working. I think there is an error in your formula at step 5.2 of Glickman's paper (second bullet point at the top of p3).

Line 138 of your code says:

if( pow(delta,2) > (pow(p.vol,2) + v) ){
    var B = ln(pow(delta,2) - pow(p.rd2,2), v);

I think the first line is wrong and should use the rating deviation, not the volatility.

if( pow(delta,2) > (pow(p.rd2,2) + v) ){
    var B = ln(pow(delta,2) - pow(p.rd2,2), v);

The second line also looks a bit odd as you're passing "v" into the "ln" function as a second parameter. Shouldn't it be as follows?

var B = ln(pow(delta,2) - pow(p.rd2,2) - v);

J.

I am not sure if a system constant is or was at one time the default of 0.75 but he says in his example (http://www.glicko.net/glicko/glicko2.pdf) the system constant should be 0.5 as he states "Assume the 1500 player has
volatility � = 0:06, and the system constant � is 0.5."

In glicko.js line 139 (step 5.2) you have:

var B = ln(pow(delta,2) - pow(p.rd2,2) , v);

which should be

var B = ln(pow(delta,2) - pow(p.rd2,2) - v);

It seems you used this function for ln(): http://jasonwyatt.tumblr.com/post/4344004125/natural-log-in-javascript
If you read the comment, his function is incorrect as Math.log() returns the natural log, not log base 10. You will see that if you fix your natural log function, your basic test from the example paper will match exactly.