Comparing waveform files made by different conversion routes: a packed struct can lose its history in one route, so diff content before comparing sizes

Field result: two copies of the same simulation dump were made by different routes, one written directly from the simulator's native dump by a reader, the other via a vendor dump-to-VCD tool piped into vcd2fst. The second file looked like a fair 'same content' competitor, but about 5.6% of the distinct changes were missing. The VCD tool expanded packed-struct signals into per-member variables that carried only an initial x and no later changes, while the direct route kept each struct as one wide vector with its full history. Neither file format was at fault; the conversion route was. Lessons: (1) before a size or latency comparison, match signals by content (per-stream change count and hash, independent of names) across files, and report any streams present on only one side; (2) record exactly how each competitor file was produced (tool, version, flags, pipeline); (3) when a gap shows up, confirm the cause with a tiny toy input through each stage separately (a two-member packed struct through the dumper and VCD tool, a hand-written VCD through vcd2fst) instead of re-running the big conversion; (4) report size claims as 'smaller while carrying more' only after that, and never as a property of the competitor format.