Iceberg 129 keeps that list as a tree of immutable files. After books.py (Snapshots) inserted the six books and changed one price, the books table's directory holds:
# Every object of the books table after two commits (UUIDs and snapshot IDs shortened).
U='[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}'
mc ls -r lake/warehouse/booknest/books/ | sed -E "s/^\[[^]]*\] *//; s/$U/<uuid>/g;
s/snap-([0-9]{4})[0-9]+/snap-\1.../"2.0KiB STANDARD data/00000-1-<uuid>-0-00001.parquet 2.0KiB STANDARD data/00000-5-<uuid>-0-00001.parquet 970B STANDARD metadata/00000-<uuid>.metadata.json 2.0KiB STANDARD metadata/00001-<uuid>.metadata.json 3.1KiB STANDARD metadata/00002-<uuid>.metadata.json 7.2KiB STANDARD metadata/<uuid>-m0.avro 7.2KiB STANDARD metadata/<uuid>-m1.avro 7.2KiB STANDARD metadata/<uuid>-m0.avro 4.4KiB STANDARD metadata/snap-1796...-1-<uuid>.avro 4.3KiB STANDARD metadata/snap-9156...-1-<uuid>.avro
A metadata file (00002-....metadata.json) is the table's root at one moment: schemas, partition specs, properties and the snapshots, each pointing to its manifest list (snap-....avro). The manifest list names the snapshot's manifests with partition ranges so planners can skip whole manifests, and each manifest (...-m0.avro) lists data files with partition values, row counts, column statistics and a status (added, existing or deleted). Every commit writes a new metadata file; 00000 is the empty table of CREATE TABLE.
The update never touched file A: it wrote file B, and snapshot 1796's manifests record A as deleted and B as added, while snapshot 9156 still points at A.