elasticsearch custom_score multiplication is inaccurate -

i've inserted documents identical except 1 floating-point field, called a.

when script of custom_score query set _score, resulting score 0.40464813 particular query matching fields. when script changed _score * a (mvel) same query, a 9.908349251612433, final score becomes 4.0619955.

now, if run calculation via chrome's js console, 4.009394996051871.

• 4.0619955 (elasticsearch)
• 4.009394996051871 (chrome)

this quite difference , produces incorrect ordering of results. why be, , there way correct it?

if run simple calculation using numbers provided, result expect.

curl -xpost 'http://127.0.0.1:9200/test/test?pretty=1'  -d ' {    "a" : 9.90834925161243 } '  curl -xget 'http://127.0.0.1:9200/test/test/_search?pretty=1'  -d ' {    "query" : {       "custom_score" : {          "script" : "0.40464813 *doc[\u0027a\u0027].value",          "query" : {             "match_all" : {}          }       }    } } '  # { #    "hits" : { #       "hits" : [ #          { #             "_source" : { #                "a" : 9.90834925161243 #             }, #             "_score" : 4.009395, #             "_index" : "test", #             "_id" : "lpesz0j6rt-xt76aatcfow", #             "_type" : "test" #          } #       ], #       "max_score" : 4.009395, #       "total" : 1 #    }, #    "timed_out" : false, #    "_shards" : { #       "failed" : 0, #       "successful" : 5, #       "total" : 5 #    }, #    "took" : 1 # } 

i think running here testing little data across multiple shards.

doc frequencies calculated per shard default. if have 2 identical docs on shard_1 , 1 doc on shard_2, docs on shard_1 score lower docs on shard_2.

with more data, document frequencies tend out on shards. when testing small amounts of data either want create index 1 shard, or add search_type=dfs_query_then_fetch query string params.

this calculates global doc frequencies across involved shards before calculating scores.

if set explain true in query, can see how scores being calculated