Addendum A2 to the pre-registration of a replication of CNet (arXiv 2610.08592): rerun of the tiny-shakespeare arm at the authors' current code

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).

Frozen addendum (29 Oct 2026) to the private pre-registered replication of the character-level language-model claim of arXiv 2610.08592 (CNet, complex-valued causal FNet on tiny-shakespeare). A first run at the commit available on 11 Oct 2026 did not reach the paper's reported loss. The authors' repository has since changed (including a new default weight initialisation on 22 Oct 2026), so the run is repeated at commit 4d69f7af1c3d1b0cd7fddb3dbf2ca4b0ef554b64 with the flags fixed here, with the new default initialisation (primary) and the old initialisation as a descriptive pair, against the pre-registered gate (final smoothed validation loss within 0.05 of 1.75 and 2.03 reached by step 5,500). Includes the Kaggle kernel. No result from either arm exists at the time of deposit.