Skip to main content

Module paper_a

Module paper_a 

Source
Expand description

Paper A fixture benches: LongMemEval_S, MemoryAgentBench, MemConflict.

Writes go through admit_seat_write. A raw line is a refusal. Hit@1 is lexical overlap of the question against admitted claims only. These rows are fixtures, not a published SOTA table.

Structs§

BenchRow
PaperATable

Constants§

BENCHES
The three named benches the paper eval gate asks for.