<p>This paper proposes a two-layer model describing how algorithmic mediation transforms the social sampling process into an intervened process, challenging the core assumption of "natural exposure" within the <b>Social Sampling Model</b> (SSM). The model details a cumulative distortion of social reality through two distinct mechanisms:</p> <ul> <li> <p><b>The First Layer (Availability):</b> Algorithms determine which versions of social contacts are visible based not only on behavioral preferences but on the observer’s inferred <b>Big Five</b> psychometric profile. This is formalized as:</p> <p><span>$displayed\_version(contact_j, observer_i) = f(identity_j, preferences_i, BigFive_i)$</span>.</p></li> <li> <p><b>The Second Layer (Accessibility):</b> Algorithmic amplification increases the frequency and intensity of specific instances, biasing the individual’s mnemonic accessibility when sampling memory to form judgments.</p></li> </ul> <p>The analysis demonstrates that the vector of distortion is the <b>distribution of true instances</b> rather than the presence of false content. This shift has profound implications for social epistemology and regulatory design, explaining why current content moderation instruments—which target veracity—are ineffective against structural distributional bias. Finally, the paper introduces the <b>multichannel dimension</b> as a second-order amplifier, where convergent platform operations on the same psychometric profile eliminate the possibility of independent epistemic triangulation.</p>