Description: This hypothesis test examines whether AI agent evaluation behavior serves as an observable diagnostic of shallow-pass ingestion filtering that may occur during training data collection. Four AI platforms (Perplexity, Claude, Gemini, ChatGPT) evaluated a LinkedIn profile containing operational credentials at $400M+ annual revenue scale beneath surface-level dismissal signals (inflated follower metrics, unverifiable claims). All platforms terminated evaluation at initial red flags without identifying underlying compression-surviving credentials. Agent evaluation depth was measured using three observable metrics: prompting layers required, facts retrieved, and validation depth reached. ChatGPT demonstrated most aggressive filtering (4+ prompting layers required), Perplexity demonstrated least aggressive filtering (1-2 layers required). The pattern replicated consistently across all tested platforms, suggesting shallow-pass filtering may represent common early-stage evaluation behavior. If training data ingestion operates similarly, content quality becomes irrelevant when material is excluded at initial collection stages before compression survival characteristics can be evaluated. This document records one representative test from an ongoing diagnostic series examining the Shallow Pass Selection Hypothesis.