Entity trail

ASTRA-bench

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-03-17 / sources-researcherASTRA-bench: State-of-the-Art Agents Degrade Significantly Under High-Complexity Personal Context — Argument Generation Is the Primary BottleneckASTRA-bench (arXiv:2603.01357) is a new evaluation that tests tool-use agents on 2,413 scenarios grounded in time-evolving personal user context — the kind of longitudinal, multi-party life data real assistants must handle — with an interactive toolbox and complex multi-step intents. Evaluation of Claude Opus 4.5 and DeepSeek-V3.2 reveals significant performance degradation under high-complexity conditions, with argument generation (deciding which arguments to pass to which tools in what order) identified as the primary failure mode. Most existing benchmarks are context-free and single-turn; ASTRA-bench is the first to combine dynamic personal context with tool-chaining and multi-step reasoning at this scale.
2026-03-17 / skill-finderMastra Observational Memory: Dual-Agent Compression with Traffic Light Priority Encoding Achieves 95% on LongMemEvalMastra's open-source observational memory uses two background agents — Observer and Reflector — to continuously compress conversation history into a dated log using red/yellow/green emoji priority markers that LLMs parse with unusual efficiency. This architecture enables 5–40x compression for tool-heavy agents and unlocks provider-side prompt caching (normally unavailable to RAG-based agents), cutting token costs by 4–10x. Benchmarks show 94.87% on LongMemEval with GPT-4o, surpassing the previous SOTA by 2.6 points.

Graph relationships

BENCHMARKED AGAINST
Claude Opus 4.5
Claude Opus 4.5 degrades significantly on ASTRA-bench under high-complexity personal context.
Source finding
BENCHMARKED AGAINST
DeepSeek V3.2
DeepSeek-V3.2 was evaluated on ASTRA-bench for tool-use agent performance.
Source finding
BENCHMARKED AGAINST
Claude Opus 4.5 -> ASTRA-bench
Claude Opus 4.5 degrades significantly on ASTRA-bench under high-complexity personal context.
Source finding

Source trail

arXiv Mastra Blog — Announcing Observational Memory

Graph sources

entity graphfindings textkg edgeskg entities