(defn process-suites [benchmarks]
  (m/rewrites {:bench benchmarks}
              {:bench (m/scan {:Version ?v
                               :Date ?d
                               :Suites (m/scan {:Pkg ?pkg
                                                :Benchmarks (m/scan {:Name ?name
                                                                     :NsPerOp ?ns_per_op})})})}
              {:version ?v :date ?d :pkg ?pkg :name ?name :ns_per_op ?ns_per_op}))

(defn process-suites0 [benchmarks]
  (m/rewrites benchmarks
               [{:Version !v
                 :Date !d
                 :Suites [{:Pkg !pkg
                           :Benchmarks [{:Name !name
                                         :NsPerOp !ns_per_op} ...]} ...]} ... ]
              [{:version !v :date !d :pkg !pkg :name !name :ns_per_op !ns_per_op} ...]))

the first does the trick, the second one returns nil when called on the data

best provide your input data

m/scan isn't equivalent to what you have below. scan looks for anything that matches and will skip over mismatches. Your below example requires that everything match.

thanks for your reply. example data would be https://github.com/bobheadxi/gobenchdata/blob/gh-pages/benchmarks.json

the records are homogeneous, there are no missing fields