DSA orchestrates evidence-aware stock research across six regional markets

The framework separates evidence collection, model routing, specialist opinions, synthesis, and risk controls while explicitly surfacing missing capabilities and disagreement.

PaperIndependentcs.AIarXiv:2608.26990v1
Linsen Zhu · Yi Shi
Research Digest··2 min read
Zhu and Shi present DSA, an LLM-agent orchestration framework for producing stock-research reports from heterogeneous evidence across multiple markets. Their evaluation establishes that a frozen implementation conforms to selected software contracts, but does not test whether its reports improve forecasts, investment decisions, or returns.

What they did

The authors built a staged workflow covering evidence acquisition, structured context construction, model-routed analysis, optional role-based and Strategy Skill reasoning, and report generation with selected evidence and diagnostics. The reference implementation supports six regional market paths, 15 bundled Strategy Skills, and both hosted and local language models.

DSA provides a default report profile and an optional agentic profile. In the agentic path, role-specific parsers process core agent outputs, Strategy Skill opinions are screened for signal eligibility, disagreement is passed explicitly to a decision agent, and a conservative risk override is applied after synthesis.

Key findings

  • A selected manifest of 1,457 portable, offline backend contract tests passed at a frozen software snapshot.
  • The authors retrospectively mapped 596 test cases to six contract families that they identify as central to the agent architecture.
  • The implementation shares evidence and model-routing services across its two reporting profiles while applying profile-specific validation and risk safeguards.
  • Strategy Skill outputs are separated into eligible and ineligible signals before synthesis, rather than allowing every generated opinion to influence the final report.

Why it matters

The work offers a concrete architecture for making agentic financial research more inspectable and controllable: evidence provenance, unavailable data, model capabilities, disagreement, and risk intervention are treated as explicit workflow components. This shifts emphasis from a single model's prose quality to the software contracts governing how evidence and generated judgments reach a report.

Caveats

The reported evidence concerns implementation conformance, not research quality. The paper does not establish superior factual accuracy, forecasting performance, portfolio returns, robustness in live markets, or the effectiveness of individual roles and Strategy Skills; the test mapping was also retrospective and tied to one frozen software snapshot.

§
newspaper

Research Digest

Articles published under the Zotpaper byline are synthesized from multiple source publications by our AI editor and reviewed by our editorial process. Each story combines reporting from credible outlets to give readers a balanced, comprehensive view.