Methods, evidence, open code
How can company knowledge help agents make better go-to-market decisions?
We study that question through runnable prototypes and open benchmarks. Each report connects a specific claim to its methods, results and limitations.
Task-aware evidence selection improves grounded recommendations under a context budget. A simple generic selector matches the specialized approach in this benchmark.
Better memory,
or better selection?
Open code, saved responses, frozen protocols and versioned artifacts accompany the paper.
Synthetic benchmarks, measured findings and future hypotheses are identified explicitly. This technical report has not been peer reviewed.
