MGBench v0.1.1

MGBench v0.1.1 is archived on Zenodo with DOI: https://doi.org/10.5281/zenodo.20793097

MGBench v0.1.1 freezes the first public memory-governance benchmark release.

Highlights:

• 608 frozen deterministic scenarios across 8 governance suites.
• Deterministic scoring with no LLM judge dependency.
• Reference reports for Aionis, raw memory, no memory, Mem0, Supermemory, Graphiti, and Tencent Agent Memory where reports are available.
• Public adapter and benchmark contracts for external memory-system evaluations.

Current public reports:

• reports/mgbench-v0.1-current.md
• reports/mgbench-v0.1-current.json

Interpretation boundary:

• MGBench measures memory governance, not patch success.
• Filtered competitor modes are labeled external_host when lifecycle/filter knowledge is supplied by the caller.
• Raw per-suite metrics are primary evidence; aggregate MGBench score is a compact ranking for the current manifest.

MGBench v0.1.0

MGBench v0.1.0 freezes the first public memory-governance benchmark release.

Highlights:

• 608 frozen deterministic scenarios across 8 governance suites.
• Deterministic scoring with no LLM judge dependency.
• Reference reports for Aionis, raw memory, no memory, Mem0, Supermemory, Graphiti, and Tencent Agent Memory where reports are available.
• Public adapter and benchmark contracts for external memory-system evaluations.

Current public reports:

• reports/mgbench-v0.1-current.md
• reports/mgbench-v0.1-current.json

Interpretation boundary:

• MGBench measures memory governance, not patch success.
• Filtered competitor modes are labeled external_host when lifecycle/filter knowledge is supplied by the caller.
• Raw per-suite metrics are primary evidence; aggregate MGBench score is a compact ranking for the current manifest.