Track 3 · #3
Errata
Awarded to Kshitij
OSS-engine (OpenCypher)
150 questions across 3 seeds with McNemar significance and a validated judge.
Judges’ citation
Best eval methodology: 150 Qs x 3 seeds, McNemar significance (p=0.049/0.016), validated judge (8.3% false-accept), deployed + MCP, publishes its own failures.
Track 3 · Agent memory — LongMemEval, BEAM, MedLoCoMo