Serialization benchmarks for the COMPAS core ecosystem: wire size, serialize and deserialize time, peak memory, and round-trip fidelity across JSON, protobuf, and MessagePack. Each report below is interactive — filter by compressed or uncompressed formats and read the per-subject detail. Sizes and losslessness are deterministic; timings from a shared CI runner are indicative only.
At the largest measured size for each subject, relative to json, the best compas_pb variant is typically 6.31× smaller in wire size and 1.43× faster in round-trip time. A compas_pb variant is smallest for 31/32 subjects and fastest for 29/32.
Encoded samples: README.md, boxes.json, boxes.msgpack, boxes.msgpack.json, boxes.pb, boxes.pb.json, graph.json, graph.msgpack, graph.msgpack.json, graph.pb, graph.pb.json, mesh.json, mesh.msgpack, mesh.msgpack.json, mesh.pb, mesh.pb.json, pointcloud.json, pointcloud.msgpack, pointcloud.msgpack.json, pointcloud.pb, pointcloud.pb.json