1id,representation,question_access,training,compatibility,composable,status_in_wp302A0,none (task only),n/a,none,any,n/a,measured3A1,raw text,blind,none,any receiver,concatenation,measured4A2,text head/tail,blind,none,any,concatenation,measured5A3,extractive sentences (BM25/centrality),blind or aware,none,any,concatenation,measured6A4,abstractive summary,blind or aware,none (prompted sender),any,concatenation,measured7A5,typed facts + source spans + unknowns (JSON),blind or aware,none (prompted sender),any,record merge,measured8A6,reference index + metered retrieval,"blind index, aware retrieval",none,any + source access,shared memory,measured9A7,"raw text, zstd transport codec",blind,none,any,concatenation,measured10A8,LLMLingua-2 token dropping,blind,pretrained compressor (official),any,concatenation,measured11B0,native bf16 KV cache,blind,none,same checkpoint,no,measured12B1,serialized KV cache (manifest-checked),blind,none,same checkpoint + revision,no,measured (process + 2-node)13B2,packed quantized KV (8/4/2 bit),blind,none,same checkpoint,no,measured14B3,sink+recent retention (StreamingLLM-inspired),blind,none,same checkpoint,no,measured15B4,random retention (control),blind,none,same checkpoint,no,measured16B5,attention-informed retention (SnapKV rule),aware (blind variant tested),none,same checkpoint,no,measured17B6,pyramid layer budgets (PyramidKV rule),aware,none,same checkpoint,no,measured18B9,selection then packed quantization,aware,none,same checkpoint,no,measured19C0,sender cache fed as receiver cache,blind,none,incompatible (negative control),no,measured (control)20C2,"linear KV mapper (ridge, un-rotated K/V)",blind,ridge on train caches,trained pair (Qwen 1.5B→3B),no,measured21C3,mapper + attention selection,aware,ridge on train caches,trained pair,no,measured22—,SwaT cross-family mapper,blind,~300M mapper (not public),Granite→SmolLM,no,not run: no checkpoint/licence23—,Attention-Matching compaction,blind,none (optimization),same model,no,not run: CUDA/vLLM code; MLX port out of scope24—,C2C fuser,blind,released fusers (Qwen3),Qwen3 receivers,no,not run: CUDA; receiver still prefills source25—,H2O eviction,cumulative,none,same checkpoint,no,not run: requires attention scores during whole prefill26—,KIVI per-channel key quantization,blind,none,same checkpoint,no,not run: CUDA kernels; motivation confirmed by B2 Qwen result27