Foveance Benchmark: Performance Results for Long-Horizon LLM Agents

AbteeXAILabs created this dataset of benchmark results for Foveance, an anticipatory context-allocation layer for long-horizon LLM agents. It contains evidence files for six different allocation policies tested on a buried-fact recall agent loop. The dataset was last updated on July 3, 2026.

Use Cases

• Compare the performance of different context-allocation policies like full replay and recency.
• Benchmark the Foveance allocation method against reactive AFM and budget-aware truncation.
• Validate the results reported in the project's README and technical report using the raw CSV data.

Strengths

• All numerical results in the project's documentation are traceable to the raw CSV files, with no hand-entered data.
• Benchmarks six distinct context-allocation policies on a specific agent task.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

AbteeXAILabs

Collection Method

Likely contains results from automated benchmarking runs of the Foveance layer.

Freshness

Last updated 2026-07-03 10:15:19; freshness should be verified.