Memory-Gym SFT improves over the untrained Researcher on all four benchmarks and gives the best supervision-source scores on three. MemAgent-derived data gives the best HotpotQA score but, like Memory-R1-derived data, reduces NarrativeQA F1. Full Memory-Gym supervision exceeds each single-family variant on all four benchmarks; the best single family varies by benchmark.

Memory-Gym SFT improves over the untrained Researcher on all four benchmarks and gives the best supervision-source scores on three. MemAgent-derived data gives the best HotpotQA score but, like Memory-R1-derived data, reduces NarrativeQA F1. Full Memory-Gym supervision exceeds each single-family variant on all four benchmarks; the best single family varies by benchmark. · CiteArk

Measurement results

0/28 assessed

0 supported0 disputed or conflicting28 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Metric and conditions	Reported	Latest observation	Evidence judgment
variant: w/o SFT · category: overall · benchmark: locomo	48.59 F1 points (0–100)	—	Not assessed
variant: w/o SFT · category: overall · benchmark: longmemeval	48.8%	—	Not assessed
variant: w/o SFT · category: overall · benchmark: narrativeqa	34.98 F1 points (0–100)	—	Not assessed
variant: w/o SFT · category: overall · benchmark: hotpotqa	45.44 F1 points (0–100)	—	Not assessed
variant: MemAgent data · category: overall · benchmark: locomo	48.6 F1 points (0–100)	—	Not assessed
variant: MemAgent data · category: overall · benchmark: longmemeval	55.2%	—	Not assessed
variant: MemAgent data · category: overall · benchmark: narrativeqa	31.98 F1 points (0–100)	—	Not assessed
variant: MemAgent data · category: overall · benchmark: hotpotqa	54.53 F1 points (0–100)	—	Not assessed
variant: Memory-R1 data · category: overall · benchmark: locomo	48.9 F1 points (0–100)	—	Not assessed
variant: Memory-R1 data · category: overall · benchmark: longmemeval	57.2%	—	Not assessed
variant: Memory-R1 data · category: overall · benchmark: narrativeqa	32.42 F1 points (0–100)	—	Not assessed
variant: Memory-R1 data · category: overall · benchmark: hotpotqa	49.81 F1 points (0–100)	—	Not assessed

Reported and observed measurements

Assessments (0)

Runs (0)

Measurement results