* disparity variation algorithm

2 files changed, 9 insertions, 1 deletions

diff --git a/doc/ideas.txt b/doc/ideas.txt
index ec8f999..3c7f773 100644
--- a/doc/ideas.txt
+++ b/doc/ideas.txt
@@ -7,6 +7,7 @@ Tables:
 Charts:
  - rolling average and deviation of rankings
  - average ranking at the beginning of each championship season
+ - average ranking at the end of each championship season
  - peak season ranking (MAX, MIN, AVG, STDDEV GROUP BY YEAR(date))
  - top 5 of every season (on average) by decade
 
diff --git a/doc/results.txt b/doc/results.txt
index c353368..13a56f3 100644
--- a/doc/results.txt
+++ b/doc/results.txt
@@ -21,14 +21,21 @@ Specific rules regarding Elo implementation for this application's purposes:
  * no minimum rating limit for driver is enforced
  * drivers are rated from their first entry (there's no initial grace period)
  * shared drive entries have the effective ranking equal to the average ranking of drivers sharing the drive; pending ranking points from duels of such entries are divided equally between drivers sharing the drive
+ * driver group disparity is varied to accommodate for dynamic shifts of relative performance within the F1 field (caused by technical changes) - see below
 
 Other than that, standard Elo rating conventions apply:
  * drivers start with identical initial rating
  * duels between high-ranked drivers change their rankings by fewer points than duels between low-ranked drivers
 
+Field disparity change:
+ * higher disparity leads to more rating inflation - as disparity is a measure of rating difference that yields a certian probability of driver's victory, so higher disparity leads to more attribution for a victory to a driver (and less attribution to the shift of performance, e.g. car performance change)
+ * higher rating deviation in the months prior to a race suggests a shift in relative performance (some drivers gain a lot, some divers lose a lot), so there's a need for damping further changes a bit - by lowering disparity
+ * once the ratings are stabilized (meaning relative performance within the field had settled), disparity can be increased
+ * this helps with the initial "rolling start" phase of the ranking (1946-1949 races)
+
 Specific parameters which are configurable:
  * initial driver ranking
- * disparity factor (ranking difference which drops the possibility of lower-ranked driver's win by the factor of 10)
+ * initial disparity factor (ranking difference which drops the possibility of lower-ranked driver's win by the factor of 10)
  * duel importance (base, i.e. for drivers ranked below importance thresholds) for all race types
  * importance thresholds (for 50% of base importance and 75% of base importance)