Yorkshire Chess Association





Developments in the ECF Rating System

and some “bugs”


a)  Swathes of Players Becoming Unrated


Some players, especially those who have not yet resumed over-the-board chess since the original cessation caused by Covid, noticed in May or before that they no longer had a current ECF rating.  On the ECF rating database inquiry screen, which differentiates between “Rated” and “Unrated” players, these players had been cast into the abyss of the undead, i. e. the “Unrated” list.  As most had been out of action for less than the traditional three years period of grace, some evidently were a bit grumpy about this turn of events.


This may have been in part due to the introduction of a system of rating categories, which hitherto had been absent, being later than planned.  (At this point the reader should visualise the cavalry beginning to appear over the horizon.)


b)  Introduction of Grading Categories


In the old 3-digit grading system, each person’s grade had affixed to the end a letter, A, B, C, D or E, and later sometimes X.  This letter encapsulated a combination of how many games contributed to the grade and how long ago they were played.  Essentially A grades were likely to be more accurate than B grades, which in turn were probably more accurate than C grades and so on.


In the new 4-digit rating system there are 4 categories, A, K, P and H, and most of the post-Covid non-returners will find they are now in category K or category H, according to circumstances.


So, what do these category affixes mean?  To understand this, it helps to understand that two different rating calculation methods (“algorithms”) are used as the circumstances require.  Most calculations follow the Elo-style approach in which, broadly, the difference between the two opponents’ ratings is used to determine an amount to be added or subtracted to a player’s rating as it was at the start of the game, giving a new rating which will feed into the next game.  This method is called the “K rating algorithm”, and yes, there is a link there to the K category suffix.


For players without a rating, a calculation method is used which does not include a player’s previous rating. This very much resembles the recursive method use by graders of old in producing an “estimated” grade for an ungraded player before doing the main calculation run.  (Recursive methods produced an “estimated” initial grade for an ungraded player which was the same as the final grade calculated for that player, “proving” the estimate was “correct”.)   This method is called the “P rating algorithm”, and yes, there is a link there to the P category suffix.


Meanings of the categories:


A category is ascribed to players whose rating has been calculated by the K rating algorithm, and who have at least 30 games included from the previous 12 months - essentially the same as for an old-style A grade.


K category is ascribed to players whose rating has been calculated by the K rating algorithm, but who have less than 30 games included from the previous 12 months and at least 10 games in total – much like the old-style B, C and D grades, though without the graduated degrees of “accuracy”.  (There could have been new K, L and M categories on a similar basis as for B, C and D.)

[The qualifying period for “at least 10 games” seems to have been extended (from 3 years to 5 years?) to allow for COVID putting players out of action for a long time.]


P category is ascribed to players whose rating has been calculated by the P rating algorithm, and who have less than 10 rated games in the system.  The “P” actually stands for “partial”, but it could perhaps more usefully be thought of as standing for “provisional”.  If after a rating calculation a player who had a P grade has then got 10 or more rated games in the system, they acquire a K grade, or if they had played 21 or more rated games in the last month (very, very unlikely) they would get an A grade.


H category is ascribed to players who “have limited recent activity under many different circumstances.”  The “H” stands for “historic”.  If a player with an H rating has at least one game rated in the next 12 months then their rating with become a K rating, but on the other hand a further 12 months of dormancy will result in them becoming “unrated”.  Thus the H category seems only a temporary one, leaving just A, K and P being used in the long term.


When an unrated player who Has had a rating in the past eventually returns to being rated, they are treated as a “new” player for the purposes of rating calculation, and so will initially be rated using the P algorithm.


A and K ratings are regarded as “full ratings” and are regarded as fully fit for normal use.


P and H ratings are regarded as “partial ratings”.  The accuracy of P and H grades in terms of a player’s present strength is unquantifiable, and they can only be an approximate “ballpark” guide and are best taken along with any known recent results which as yet have not been rated.


Details of ECF rating calculation (though not yet including references to categories) can be found on the ECF rating website at https://www.ecfrating.org.uk/v2/help/help_rating.php.


c)  Known Bug in Congress Result Cross-Tables


If you take a half-point bye in round 1 of a 5-round weekend congress, then in the remaining four rounds score successively a draw, a win, a loss, and a draw, then in the tournament result chart generated by the ECF rating website you will be credited with a final total of 2.  This is because byes and wins by default are not being added in, even if the rating officer involved has reported them.


Where the rating officer has reported it, the half-point bye will appear in the list of games for the player concerned, though it has little relevance in that list, yet it is not added into the players total in the tournament result chart where it is needed.


This is irksome when a player complains in the belief that the rating officer is at fault.  The specification for the submission files, both “text” and “Excel” formats and presumably also “Jason” format, include the facility for recording byes and defaults, even though they were/are not relevant to grading/rating calculation.


The ECF rating manager has been informed, but whilst a review of such things is planned it is “low priority” in the present context of the operation of rating categories still being tested.


d)  Known Bug in Event Listing – just cured!  (Original present tense changed to past tense.)


Results of all sections of a congress can be submitted in a single file.  When this was done, it resulted in a tournament cross-table for each section, proving the system has recognised the sections which thus were correctly identified as specified in the submission file format specification.  However, in the list of events, the event was not identified by the “Event Name” supplied in the header records, but instead was identified by the Event Name compounded with the name of the last section.  On going back to the website to find an example to quote, it proves that this bug has been cured in the last few days!  Now for the Doncaster Congress all sections are listed rather than just the last one.


e)  Player Look-up (Temporarily) Not Working


When trying to check whether, by any chance, the bug in the cross-tables had also just been cured, it was found the facility to search on a player’s name did not work!  Somebody was obviously tinkering with the system.  (Sure enough, an update was in progress.)  Approaching the matter from a different angle, it transpired the cross-table bug persists, and will probably be there for some time.


All that said, the new ECF rating system seems good, and there are people working hard in the background to keep it running and to make improvements and enhancements.


f)  Rating Audit


The above-mentioned update was to implement a “rating audit” for a player for a specified month.  Now, if you click on an OTB rating (standard rate or rapidplay) in a player’s profile page, the details of rating calculations for the latest ended month are displayed.  You can change the month and/or year (or indeed the player rating reference number).