Escalation ladder

A design pattern, developed in Section 3K, in which a single action type carries more than one checkpoint tier, each with its own explicit trigger condition and its own approval form, so routine and unusual instances of the same action type are not treated identically. A well-designed ladder reserves its highest-friction rung for outlier behaviour (Trigger 3) rather than relying solely on named thresholds.

The graduated sequence of responses to an objectionable directive, from a private clarifying question, to a voiced concern with an alternative, to escalation above your boss or a parallel report to a specialist function (compliance, legal, data-protection officer, AI governance), to a principled refusal, to exit; you start at the lowest rung that fits the severity and climb only as forced.

The tiered path a conflict takes when it cannot be resolved at the leader's level: things you resolve, things the governance group resolves on its cycle, things needing the top approver, and a fast lane for genuine emergencies. A working ladder resolves most conflict below the CEO, so the rare escalation lands with weight instead of as noise.

A hybrid in which the agent handles a task at a light pattern and automatically moves to a heavier one, and ultimately to a human, when a measured signal (rising error rate, repeated failure, uncertainty) turns against it. It encodes the demonstrated-reliability override as a live control that does not assume the agent stays reliable.

A documented, graduated set of responses to confirmed content violations, mapped in advance to violation severity, satisfying Article 18's requirement of warning, feature restriction, service suspension, account termination, and authority reporting for the most serious cases.