Same time, same object, different value: probing a small language model for conflicts inside the context window

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

A pre-registered pilot on Qwen2.5-0.5B. A first protocol met all its criteria (AUROC 1.00), but post-hoc controls showed the result was lexical. A second protocol, trained on colours and materials and tested on sizes, an attribute never seen in training, met all its criteria again: AUROC 0.99 pooled, 0.98 against temporal update, with an increment beyond next-token entropy. The note reports a transferring direction that is sensitive to time and object. It does not claim a detector of contradiction as such. Frozen protocols with SHA-256 hashes, scripts and results are included.