Architecture-Aware Literary Traversal by Public AI Summarizers: A Position Paper Based on Direct Forensic Evidence

Architecture-Aware Literary Traversal by Public AI Summarizers is an observational-forensics paper based on a public Google AI Overview interaction. Its empirical object is not whether the summarizer agreed with the archive but whether it preserved internal architecture while moving through it.

The paper separates three claims. Claim A is observed: Room, Chamber, and Vault distinctions remained stable; specific DOI anchors were associated with functions; heteronyms were treated as differentiated positions; and follow-up prompts opened further traversal rather than closing in a terminal summary. Claim B is hypothesized: density, redundancy, machine-readable coordinates, and cross-reference mass may have made architectural traversal easier than flattening. Claim C is interpretive: literary authorship can operate at the infrastructure level in AI-mediated environments.

Comparisons with House of Leaves, Pale Fire, hypertext fiction, ARGs, wikis, digital corpora, and knowledge graphs are used to isolate the proposed distinction between complex content and literary content that is also its own navigational infrastructure.

The paper considers roleplay as an alternative account of the system’s “constituent witness” language. Its narrower conclusion is that the architecture was executable and sequence-preserving under cooperative traversal, not that the summarizer independently became conscious of it.

The claim that no prior literary work has elicited this behavior at comparable scale is the paper’s falsifiable novelty claim. It should be represented as such, together with the paper’s invitation to replicate or identify counterexamples.