Accuracy issue in predicting the performance of bottom participants

[conlacda]

Stat

Contest	predicted perf	correct perf	number of bottom users	delta
agc067	1778 (299.5)	1712 (rank: 309.5)	94	perf: 66 & rank: 10
abc367	64 (9640.5)	53 (rank: 9788 -> 9801)	1370	perf: 11 & rank: 148 -> 161
abc366	60 (9703)	50 (9834 -> 9847)	1371	perf: 10 & rank: 131 -> 144
abc365	48 (10003.5)	40 (10107 -> 10120)	977	perf: 8 & rank: 104 -> 117
abc364	59	49
abc363	60	49
abc362	47	39
abc361	54	45
abc360	70 (7540)	42 (7819 -> 7828)	707	perf: 28 & rank: 279
abc359	42	35
arc182	404 (2131)	337 (2204)	1480	perf: 67 & rank: 73
arc181	221 (2655.5)	183 (2735.5)	1317	perf: 38 & rank: 80
arc180	272	230

Observation

• In ABC/ARC contests $$\frac{predicted perf}{correct perf} \approx \frac{6}{5}$$