Validating a compressed/closed-form waveform index (clock descriptor) against the real reader

Pattern that caught real gaps when reviewing a 'periodic grid + exception intervals' index that answers edge/search queries on FSDB/VCD signals without the reader:

1. Real dumps are often perfectly regular clocks (1 interval), so an on-vs-off comparison on them only exercises the closed form. Build an adversarial fixture (VCD -> vcd2fsdb): x before the first edge, a clock that stops, x/z landing exactly on grid edges, edges 1 tick late, missing periods, several changes in one time step (last wins), first change being a fall. Probe every deviation at -6..+11 and the dump end at T-1, T, T+1, T+2, across every edge kind, value search and strict-wildcard search, both directions. Re-run with several segment counts so segment and chunk boundaries move through the deviations. 2. Compare responses with timing/routing fields removed and binary payloads hashed. Also check that the off-by-default path is identical to the previous release. 3. Look for intervals that split at segment boundaries. They are still exact, but they inflate 'deviation' counters that may gate the feature. 4. For a red/green commit pair, run the red's own test harness on its fix. If the harness pieces (stub binaries, env vars) only arrive in the fix commit, the red crashes rather than FAILs. 5. RSS-capped child jobs: set the cap just below and just above the measured peak (check > vs >=), confirm the parent worker pid is unchanged, and check for orphans after the parent exits. Children that die only on SIGPIPE linger about one write interval.