paper_id	claim	claim_status	evidence_status	source_depth	source_quote	page_or_section	taxonomy_fit	audit_status
2603.17771v2	The abstract reports: Attention sinks and massive activations are recurring and closely related phenomena in Transformer models.	preliminary-linked	has evidence row	abstract	Attention sinks and massive activations are recurring and closely related phenomena in Transformer models.	abstract	in-scope: taxonomy category match	needs work; preliminary / abstract-derived; report=audit_report.md
2504.20966v4	The abstract reports: We introduce softpick, a rectified, not sum-to-one, drop-in replacement for softmax in transformer attention mechanisms that eliminates attention sink and massive activations.	preliminary-linked	has evidence row	abstract	We introduce softpick, a rectified, not sum-to-one, drop-in replacement for softmax in transformer attention mechanisms that eliminates attention sink and massive activations.	abstract	in-scope: taxonomy category match	needs work; preliminary / abstract-derived; report=audit_report.md
2605.00968v1	The abstract reports: Positional encoding plays a pivotal role in determin?ing the extrapolation and generalization performance of wireless foundation models for channel state information (CSI) modeling, latent characterization, and task-specific prediction.	preliminary-linked	has evidence row	abstract	Positional encoding plays a pivotal role in determin?ing the extrapolation and generalization performance of wireless foundation models for channel state information (CSI) modeling, latent characterization, and task-specific prediction.	abstract	in-scope: taxonomy category match	needs work; preliminary / abstract-derived; report=audit_report.md
2402.17762v2	The abstract reports: We observe an empirical phenomenon in Large Language Models (LLMs) -- very few activations exhibit significantly larger values than others (e.g., 100,000 times larger).	preliminary-linked	has evidence row	abstract	We observe an empirical phenomenon in Large Language Models (LLMs) -- very few activations exhibit significantly larger values than others (e.g., 100,000 times larger).	abstract	in-scope: taxonomy category match	needs work; preliminary / abstract-derived; report=audit_report.md
2509.05218v2	The abstract reports: Positional encoding mechanisms enable Transformers to model sequential structure and long-range dependencies in text.	preliminary-linked	has evidence row	abstract	Positional encoding mechanisms enable Transformers to model sequential structure and long-range dependencies in text.	abstract	in-scope: taxonomy category match	needs work; preliminary / abstract-derived; report=audit_report.md