15 commits

Author	SHA1	Message	Date
jinli.yl	3d31c660a0	docs: add Chinese counterparts for all compaction benchmark guides	2026-10-17 00:39:37 +08:00
jinli.yl	9cc6dad5a7	docs: retain only quality and context-length experiment comparisons	2026-10-16 22:11:19 +08:00
jinli.yl	7686be7baa	docs: compare compaction-only costs and omit Full from cost panel	2026-10-16 18:34:50 +08:00
jinli.yl	8f796715ab	docs: price total compaction and answering cost across Top N	2026-10-16 17:49:06 +08:00
jinli.yl	e1a3b42579	docs: plot Jev token overhead by retained Top N	2026-10-16 17:40:09 +08:00
jinli.yl	65ef735841	docs: specify reproducible compaction settings and Hermes summary scope	2026-10-16 17:36:19 +08:00
jinli.yl	d2638b166f	bench: compare Jev fusion across frozen summary strategies	2026-10-16 17:21:53 +08:00
jinli.yl	11317634e5	bench: evaluate fixed-summary fusion across Hermes fixture tasks	2026-10-16 16:47:41 +08:00
jinli.yl	26a83c1904	bench: focus on frozen-summary Jev Top-N payment evaluation	2026-10-16 14:53:20 +08:00
jinli.yl	a013411457	bench: evaluate gold-labelled HotpotQA trajectories collected with Hermes	2026-10-16 13:49:41 +08:00
jinli.yl	de6f80bd1a	refactor: replace compaction studies with one matched-input benchmark	2026-10-16 12:47:30 +08:00
jinli.yl	4c72814fae	docs: unify compaction strategy comparison	2026-10-16 12:22:52 +08:00
jinli.yl	ba57fc5cdb	benchmark: evaluate fixed-summary evidence additions and summary-aware selection	2026-10-16 12:07:35 +08:00
jinli.yl	cf55876810	docs: visualize compaction benchmark results with English charts	2026-10-16 10:47:38 +08:00
jinli.yl	49f76bfcc4	refactor: preserve recoverable tool evidence and add compaction benchmarks	2026-10-16 01:12:37 +08:00