Knowledge encoded directly into an AI model’s neural weights during training — what the model ‘knows’ without needing to retrieve external documents. Parametric memory is temporally frozen at the model’s training cutoff date and cannot be updated without retraining. Entities absent from parametric memory are mechanically disadvantaged in zero-shot queries where the model answers without retrieval. Building parametric presence requires temporal depth: years of consistent, corroborated signals across high-authority sources ingested during the model’s training window. Parametric memory is the harder-to-reach but higher-value component of full spectrum dominance because it persists across retrieval contexts.