integer-dictionary.parquet was written with parquet-mr 1.13.1 ExampleParquetWriter, PARQUET_1_0, dictionary encoding enabled, and no compression. Its schema is message schema { required int32 x; }.

It contains three row groups of 100 rows each. The groups alternate between [-10, -1], [0, 10], and [5, 10], respectively. It was written with withRowGroupSize(1) and withPageSize(1024); parquet-mr checks the row-group size every 100 rows.

Both dictionary and data pages use PLAIN_DICTIONARY in the footer‘s page encoding statistics. This exercises Auron’s legacy dictionary extraction path; modern Rust writer output does not enter that path.

For x = 5, min/max statistics prune the first group, dictionary filtering prunes the second group, and the third group is retained. The regression test checks these decisions, the metrics, and matching rows against an unpruned read.