Dividing a request's total allowed time across its downstream calls, each getting what remains rather than a fixed full timeout.