E089 Quantified-Chunk New-Post Cohort: Pre-registration with In-Flight Amendments (v1.2). Primary analysis re-scoped to P3; primary contribution a query-validity and namespace-drift taxonomy.

Abstract

E089 is a pre-registered field study testing how generative AI search engines cite cold, newly published content. The study used 12 posts arranged as six topic-matched A/B pairs. One page in each pair opened with a quantified-chunk template: a front-loaded definition, named entities and at least one hard number already deposited in a public source. The other used the standard template. I checked citation and chunk reconstruction at D14 and D28 across Google Search, AI Overviews, AI Mode and Perplexity Sonar. This version finishes the analysis re-scope that began in amendment v1.1 and changes what the study can reasonably claim. V1.1 had already narrowed the confirmatory analysis from all six pairs to the two coined-term pillar pairs, P3 and P4. The informational half of the cohort failed query validity because the target queries hadn't been screened for search demand or namespace ownership, and every pair was compromised in a different way. P4 later collapsed too. Both arms were namespace-invalid, although for different reasons. That left P3 as the only primary pair. P3 produced a measurable negative result on the registered dependent variable: reconstruction of the engineered opening chunk. None of the four answer surfaces reproduced the treatment page's front-loaded statistic. They did, however, reproduce the plain declarative definition almost word for word on both arms. So the number being placed at the front wasn't what got the content selected. That result comes from one pair at one timepoint, with no D14 baseline and a sample below the minimum-N standard. I'm therefore reporting it as a directional negative, not as a confirmatory finding about the template. The study's primary contribution is now a taxonomy of query-validity failure and namespace drift. Five of the six pairs were compromised through different mechanisms: foreign-entity namespace capture, product-brand ownership, dense-corpus collision, authority saturation, topic-invalid queries and within-pair namespace asymmetry. This extends my earlier work on namespace drift. Two incidental extraction positives remain in the record. Two cold control pages, published without amplification, earned citations in AI Overviews and Perplexity with their engineered openings reproduced. One was still being cited in an off-protocol observation roughly three weeks after its last confirmed read. Nothing about the experiment itself has changed. Treatment assignments, the randomisation seed, publishing schedule, hygiene baseline, measurement offsets and coding rules remain as registered in v1.1. This version changes only the analysis and interpretation plan. Deposited under CC BY 4.0.