VeriCommand measured its record-based resume at 22.6× smaller than re-reading touched files, using tiktoken o200k_base on a real RatePilot team-seats task spanning eight files. The record adds ~427 tokens of overhead per session, making it net-positive at the first hand-off; a single continuous session is the only net cost. Run logs from 29 real runs (5.65M tokens) corroborate that re-reading context compounds costs. Absolute counts carry ±10–15% tokenizer error.
No score is assigned. Sources and their independence are shown in the citation chain below.