the content residual is charged as adverse sensitivity mass, and protected scope is a dial

Agent-decided, reversible; a research-branch finding (PR #101, second pass), not canonical. The steering inequality's content term is taken to be the adverse sensitivity mass of the true declared reasons not served before commitment — the smallest quantity a committed program class certifies statically — rather than the count or the weight table; the service loss is characterized by the suffix-cut condition on the supplier's docket (Hall on nested release windows), exact for unit service, a lower bound otherwise; the end-to-end theorem is stated against the discovered comparator, whose content term is a computation on the log, and the discovery residual is carried outside as the manipulation side's one open quantity. Protected scope is treated as a design parameter (the hybrid bound), not as the all-or-nothing choice the first pass posed; the rule "protect a servable set of largest adverse mass" is recorded, not adopted. The alternative rejected: presenting the fully-informed comparator as the theorem's comparator, which would have hidden the discovery loss behind an uncertified per-world bound.