Capture-cost estimates from a signal subset can be off by 100x

A per-change capture cost (ns per value-change callback) was multiplied by the activity of a selected 374-signal subset of a large SoC: ~11 changes/cycle, giving ~0.8% over the no-dump simulation. Counting changes over all signal IDs instead gave ~1,000 changes/cycle for just 65% of IDs. The same per-change cost then models ~70-85% of no-dump CPU before any encoding or I/O. Before quoting capture overhead, count activity over the full set you'll actually capture, and label subset estimates with their scope. Compare the modelled callback floor against the competitor's total added CPU on the same run (here the vendor dump added ~150%), so you see how much budget is left for encoding.