WorkSurface-Bench: Benchmarking Enterprise Agents on Multi-Surface Knowledge Routing

2026-07-28

AI Summary

This work addresses the lack of evaluation frameworks for assessing enterprise agents’ ability to route queries across heterogeneous knowledge sources—such as documents, tables, and dependency graphs. The authors propose WorkSurface-Bench, the first fine-grained and auditable multi-surface task benchmark, comprising 1,151 atomic tasks that formalize knowledge source selection as “surface routing.” Answer provenance is enabled through DuckDB queries, textual snippet matching, and graph annotation tracing. Experiments reveal that under golden-tool constraints, agents achieve surface routing F1 scores of 98.7–99.8%, yet answer accuracy remains limited at 56.1–75.3%. Surface-aware prompting substantially improves performance across three model families, and pruning irrelevant tools enhances routing efficiency. A sample of 200 tasks was independently reviewed by three annotators, all meeting six predefined quality criteria.

Abstract

Enterprise agents often need to integrate heterogeneous knowledge sources: documents for narrative facts, tables for computation, and dependency graphs for file relationships. Existing benchmarks typically evaluate retrieval or tool use without distinguishing whether an agent first selects the appropriate knowledge sources. We introduce WorkSurface-Bench, a benchmark for evaluating this capability as surface routing. It contains 1,151 atomic tasks derived from persona-scoped Workspace-Bench-Lite workspaces, spanning document, table, graph, and cross-surface questions. Its reference answers are auditable: table answers are reproduced through executed DuckDB queries, document answers are grounded in verified text spans, and graph answers are traced to source dependency annotations. We evaluate four model backbones across six controlled agent settings, yielding 27,624 protocol-error-free trajectories. Under gold-constrained tool access, agents achieve 98.7-99.8 Route F1, while Answer remains only 56.1-75.3 percent, showing that correct surface selection is necessary but insufficient for task completion. Matched interventions further show that surface hints improve Answer for three of four models, whereas removing irrelevant tools primarily improves routing and efficiency. In an independent three-annotator audit, all 200 sampled tasks pass all six quality criteria by majority vote, with 192 receiving unanimous judgments on every criterion. We release the dataset, construction pipeline, scoring code, and agent harness at https://github.com/haolpku/WorkSurface-Bench.

Problem

• enterprise agents
• knowledge routing
• multi-surface
• benchmarking
• heterogeneous knowledge sources

Innovation

• surface routing
• enterprise agents
• multi-surface knowledge
• benchmarking
• auditable answers