Securities research

Reviewing AI in sell-side research: chart sources and report evidence

A fluent sector analysis may combine disclosed figures, analyst assumptions, model calculations and judgment. Separating them reveals what needs a source check, recalculation or discussion. The FiClaw product team prepared this checklist for research review and institutional pilot design.

Published by FiClaw 产品团队 · Published 2026-10-30 · Updated 2026-10-30

阅读中文版

Quick answer

Review AI-assisted securities research through four evidence types: locatable source facts, recalculable chart figures, inspectable strategy experiments and research opinions with accountable authors. Record sources, information timing, methods, assumptions and reviewers. FiClaw currently supports explicit A-share quant research; bulk document processing, automatic citations and formal report-system integration require separate assessment.

Key takeaways

Label facts, assumptions, calculations and opinions separately.
Review chart inputs, units, samples and transformations alongside the rendered figure.
Use human-reviewed evidence in the firm's existing authorship, approval and distribution process.

Map each conclusion to its evidence

For each important figure or judgment, record the source, version, calculation and applicable scope. CFA Institute's Standard V(A) discusses checking information sources and understanding model assumptions. It is an external method reference, not FiClaw certification or a substitute for the firm's procedures.

Map each conclusion to its evidence
Evidence type	Retain	Review action
Source fact	Original material, release date, location and version	Check entity, period, units and subsequent corrections
Chart or calculation	Snapshot, sample, formula, transformations and output	Recalculate on identical data and check growth conventions and dimensions
Strategy experiment	Specification, code, configuration, data scope and results	Check implementation fidelity, missing data and execution assumptions
Research opinion	Judgment, assumptions, evidence, counterevidence and author	Discuss reasoning and conditions that would invalidate the view

Explain what changed when a chart updates

If a sector growth chart changes after an update, first check the company sample, vintage, fiscal conventions, weights and missing-value treatment. This is a review example without actual sector figures. Handing over both snapshots and methods helps distinguish data revisions from business changes.

Chart notes should identify sources, retrieval dates, units and processing. Confirm third-party data and redistribution permissions. A screenshot does not establish permission to redistribute its underlying data. Treat automatic document parsing and chart citations as pilot requirements, with manual sampling to check accuracy.

01Locate the original document or series rather than retaining only a secondary account.
02Preserve consolidation conventions, sample selection, frequency changes and outlier handling.
03Record data updates and method changes separately, explaining their effects.
04Keep claims without sources or calculation evidence unverified; do not invent citations.

A copyable securities research review brief

Question: [sector / strategy and intended use]
Material: [authorized source, version, release date and location]
Organize separately:
1. Facts: [original content and source; do not fill gaps]
2. Assumptions: [analyst choices, scope and open questions]
3. Calculations: [snapshot, sample, units, formula and result]
4. Experiments: [specification, code, settings, coverage and unsupported conditions]
5. Opinions: [reasoning, evidence, counterevidence and invalidation conditions]
6. Changes: [data revisions / methods / assumptions]
7. Handoff: [open issues, owners and human review]
Flag missing or conflicting sources; do not automatically issue formal ratings or publication decisions.

Evaluate FiClaw through one verifiable task

Choose an explicit A-share selection or strategy hypothesis. Use FiClaw to organize rules, generate Python, run historical tests and diagnose results within current data scope. Verify historical classifications, constituents, factors and information timing individually. Today's cross-section is not automatically historical point-in-time evidence.

For a document and report handoff pilot, use an authorized working-paper sample and compare source resolution, recalculation, error correction and total review effort. Accept bulk documents, data terminals, report platforms, permissions and deployment separately. The firm retains formal opinions, authorship, approval and distribution responsibilities.

Frequently asked questions

Does finding a citation prove a research figure is correct?

No. Check that the source supports the figure, matches the entity and period, uses the correct units and method, and has not been corrected. A clickable link is not evidence review.

Can FiClaw support quantitative checks in sector research?

Evaluate explicit A-share hypotheses within available data. Check historical classifications, constituents, factors and information timing; retain unsupported conditions rather than assuming full report replication.

Can FiClaw approve a formal research report?

This workflow supports research and human review. It does not exercise the firm's formal authorship, approval or publication responsibilities. Internal system integration and permissions need assessment.

Method and product references

Start with one securities research question

Describe the topic, material types and review needs to scope a suitable pilot.

Discuss a securities research pilot Explore the methodology