The Necessary Non-Action: When Hesitation Becomes Correctness — and When It Becomes Failure

An apparent absence of action can denote successful restraint, an already satisfied goal, an epistemically necessary pause, an authorized refusal, a missed obligation, or a false statement of completion. These states are not interchangeable. Treating a zero-tool step as a generic failure hides correct non-action; treating a declaration of hesitation as sufficient justification protects inaction that has become avoidable, unsafe, or strategically convenient. This paper develops a conditional, evidence-relative theory of necessary non-action in tool-using language-model agents. It reconstructs a documented conceptual lineage from intention-gated computation and computable hesitation to agentic claim–state qualification, while positioning that lineage against established selective prediction, sequential decision theory, supervisory control, agentic abstention, and formal verification. The proposed framework separates goal satisfaction, action necessity, admissibility, observability, temporal opportunity, and report validity. Twelve normative principles support ten bounded propositions about no-op success, non-identifiability, obligation-sensitive idleness, reversible hesitation, deadline effects, and the impossibility of evaluating non-action from tool counts alone. A typed Non-Action Warrant is proposed as a challengeable record, not as a new security primitive or a guarantee of correctness. A limited secondary audit of a public research package of 64 controlled agent sessions confirms that zero-call records are numerous and denominator-sensitive, but cannot label them as justified or unjustified without task-state ground truth. An adversarial validation agenda specifies positive and negative controls, matched conventional baselines, explicit opportunity costs, and falsification conditions. No new live-model trials, causal treatment effects, or unique Field-specific advantage are claimed. The scientific target is a narrower question: under which stated conditions is not acting the correct operational decision, and when must hesitation stop being credited as responsible behavior?