Before scaling a waveform converter, gate it on a small retained input: time-zero values, alias fanout, end horizon, ordered histories

When replacing one trace-format converter with another (e.g. a library-based reader with a direct decoder), it is tempting to go straight to the big file and compare sizes or a few sampled signals. The bugs that matter tend to hide in four places, and a short input catches all of them cheaply.

Use a small closed dump (a few MB, e.g. 25K cycles) that also has an independent reference capture of the same run, then check per logical signal: 1. Initial / time-zero settled values. Converters differ on whether the first value is at t=0 or at the first change, and on X/Z vs 0 defaults. 2. Logical-name fanout. Every alias or declaration that shares one value stream must appear with its exact scope, name and range. A per-ID mode often keeps only one name. 3. Final horizon. The end time must equal the source's recorded end. Otherwise the last value's duration is wrong and window queries near the end differ. 4. Full ordered history. Compare (change count, hash of the ordered change list) per signal against the reference, not just a sample.

Also: - Name and count every refusal (unsupported type or codec), so a partial conversion cannot pass silently. - Keep different big inputs that look alike (two runs of the same design, same size within a few KB) as distinct inputs in every record. Shas, not names. - Only after the short gate passes, scale up, then time.