A controlled factorial test separates the evidence a model receives from an instruction about which retailer is reachable.
Same listings. Different access instruction.
In the carry-on fixture, Samsonite appeared in 25 of 25 answers when Amazon was marked reachable. It appeared in 0 of 25 when Amazon was marked unreachable. That pattern held whether direct Amazon listings were included or removed.
Samsonite brand inclusion
Observed counts from the existing Access Shock II public dataset. Lines show marginal Wilson 95% intervals, not simultaneous intervals. Neither 0/25 nor 25/25 establishes a certain population probability.
Scope of the finding
These are observed inclusion counts in controlled fixtures, not consumer recommendation probabilities or actual web-browsing behavior. Three product recommendations were required in every response. Category and brand responses varied.
Neither extreme count establishes a zero or certain underlying probability. The chart’s marginal Wilson intervals make this sampling uncertainty visible.
Evidence and reproducibility
This page uses the existing public Access Shock II report and dataset. The full report describes collection integrity, fixtures, bootstrap analysis, and limitations.