Under a frozen FP32 dense-transformer regime, this report evaluates replay structure on teacher-forced Gate8 candidate-execution bundles using a fixed Gate12A transport-and-loop audit surface. The main evidence consists of 9 closed runs (3 models × 3 rendering families) under transformers 4.57.3. The closed lines are Qwen/Qwen2.5-3B-Instruct, meta-llama/Llama-3.2-3B-Instruct, and Qwen/Qwen3-4B. All 9 runs receive a quartet-positive structural replay verdict under the frozen rule. By contrast, the present first-pass phenotype layer remains family-conditioned and model-specific. The clearest localized concentration within the current closed 3B/4B evidence set appears only in the archive high-tension band (6 packets per model). The report also includes 4 supporting transcript-only dense-transformer extensions, 2 frozen-protocol admission failures, and 1 non-transformer sidecar (state-spaces/mamba-2.8b-hf under an altered observation surface) that does not preserve the current dense-transformer transcript-path pattern. Included files:- paper.pdf: frozen public report PDF- paper-source.zip: LaTeX source bundle for the paper- selected-manifests.zip: selected archived manifests and status files for provenance support- README.txt: release overview- CHECKSUMS-SHA256.txt: checksums for the uploaded files Frozen mainline replay summaries and structural quartet verdicts are archived under commit:084eb7878d8cb016243950e1cf4b4bd7379daaba One upstream Gate8 candidate-execution artifact exception:Qwen/Qwen2.5-3B-Instruct transcript bundle under58d06742f23a0bc7ba25c6ecde790e2e03b4324e Repository:https://github.com/Udonburo/pale-ale
Building similarity graph...
Analyzing shared references across papers
Loading...
Aoi Kawasaki
Building similarity graph...
Analyzing shared references across papers
Loading...
Aoi Kawasaki (Thu,) studied this question.
www.synapsesocial.com/papers/69d9e60578050d08c1b76534 — DOI: https://doi.org/10.5281/zenodo.19483161