Paper B: Field Baseline Pre-Registration. Does presentation format affect blind entity retrieval and brand attribution in production generative engines?

Pre-registration, deposited before any confirmatory session. With the same fact set present in every arm, does changing how those facts are presented affect (a) whether a production engine retrieves the entity from a blind prompt, and (b) whether it attributes the retrieved facts to the correct brand token? This is specifically a retrieval and attribution study. It is not testing density against citation, and it is not a comprehension study. The scope is production consumer engines on the dates we run the experiment. Four virgin entities are deployed simultaneously on kusimanta.es, one for each format arm. Every entity carries the same canonical eight facts. Presentation format is the only intended difference: B1 Sypelozy (labelled list), B2 Xykelovy (structured prose), B3 Vyremosy (narrative prose), B4 Klyxemory (dense prose). Engines: ChatGPT and Gemini, desktop web UI, one query per fresh session, at least five clean sessions per entity per engine. Primary DV is brand attribution per fact; retrieval and recognition gating are coded separately. Freeze verified 15 Sep 2026 by a second path. The deposited artefacts are the four raw HTML source files uploaded to the server. SHA-256: sypelozy.html 73152c86b795a24e097732b59e7de8a94d5b7a2070cbb2924745faa29b1dbb14 xykelovy.html 558d425c7240674d7fff02ba87d60e4c1a2a1468aca5eb964ac0c50636886b57 vyremosy.html 8e72e2c739543cb1ddd53ce63192dc730cd7046ebd3a8c61558c9777148df3d1 klyxemory.html 329aa000d4dece5d2f122d0cf6174d8f1a9aa91abbbbac7bbef944c67dab25a6 The blind scoring workbook deposited alongside this document is the coding instrument. It fixes the DV rules and rollups before any outcomes are observed. Competing interests: Julio Arévalo Piedra founded and operates KusiGEO, a commercial GEO-audit platform. This experiment does not evaluate KusiGEO or any other commercial tool, and KusiGEO tooling is not used in the design, deployment or measurement of the study. Artur Ferreira declares no competing interest in the outcome. Drafting disclosure: parts of this pre-registration and its scoring workbook were drafted with AI assistance using Anthropic Claude. The authors subsequently reviewed, corrected and finalised both documents and take full responsibility for the design, controls and registered analysis. The four deposited source files are the co-author's page source and were not AI-generated.