Financial AI Articles · Research Backlog
EDITORIAL RESEARCH · October 22, 2026 · Not a product roadmap
The curated finance topics show what is already worth reading. This backlog records questions still needing evidence, method, cases and opposing views. It is the only editorial topic register: open an Issue when a topic is actively researched, rather than opening fifteen empty tasks.
Research states: evidence needed → researching → thesis formed → editorial decision → deepen / publish separately / defer. A completed study need not become an essay. Product changes, if any, belong to ETF Agent's separate engineering OpenSpec. Source dates, financial definitions, jurisdiction and uncertain findings must be preserved.
P0 · Priority Research
| Task | Evidence needed | Existing work / state |
|---|---|---|
| F01 · Financial research quality (B3) | Separate factual correctness, suitable methods, counterevidence and decision help through a full portfolio example. | Trustworthy financial evidence chain (Chinese original)、Financial Python result consumption；Issue #12 opened; evidence needed |
| F02 · Financial error ledger and evaluations (D1) | Classify errors and near misses, preserve data/policy versions and hold out unseen cases. | No directly matching essay yet；Evidence needed |
| F03 · Human review and accountability (C4) | Distinguish confirmation, professional review and compliance; examine independent opinions, rejection and workload. | Research, advice and trading；Evidence needed |
| F04 · Financial audit evidence package (C4) | How can a reviewer lawfully reconstruct sources, authority, method, proposal, execution, delivery and missing facts? | Trustworthy financial evidence chain (Chinese original)；Evidence needed |
| F05 · From summaries to decisions (A2) | Compare holding, adjusting and waiting given horizon, liquidity and exposure; include objections and revision triggers. | No directly matching essay yet；Evidence needed |
| F06 · Thesis, execution and outcomes (A3) | Separate original advice, adoption, trades, costs and market effects without hindsight rationalization. | Investment Thesis history；Evidence needed |
P1 · Research Pipeline
| Task | Evidence needed | Existing work / state |
|---|---|---|
| F07 · Financial data freshness and quality (B1) | Verify NAV, quote and holdings timeliness, coverage, source conflicts and provider responsibility. | Financial data: dates and reproducibility；Evidence needed |
| F08 · Backtesting and strategy validity (B4) | Study look-ahead and survivorship bias, leakage, tuning, rolling/OOS tests, slippage and liquidity. | No directly matching essay yet；Evidence needed |
| F09 · Brokerage-app AI features and UX (A1) | Verify diagnostics, monitoring, conditional orders, natural-language orders and reviews, including UX misinterpretations. | Research, advice and trading；Evidence needed |
| F10 · Financial AI vendor governance (D4) | Examine data purpose, model/market vendor changes, privacy, takeover and exit. | Market-data MCP responsibilities；Evidence needed |
| F11 · Business ROI of financial AI (D3) | Compare useful work, missed risks and verification/review/maintenance costs against existing workflows. | No directly matching essay yet；Evidence needed |
| F12 · Convergent judgments and regime change (B4) | Test shared sources/factors across models, rule correlation and applicability across changing regimes. | No directly matching essay yet；Evidence needed |
| F13 · Employee-facing financial AI (D2) | Investigate adviser consoles, report production, compliance screening, approvals, roles and review costs. | No directly matching essay yet；Evidence needed |
| F14 · Financial AI approvals and lifecycle (C1) | Verify filing/registration, suitability, service authorization, retention and change processes per jurisdiction. | Research, advice and trading；Evidence needed |
| F15 · Investor AI reliance and biases (A3) | Study outsourced judgment, automation and confirmation bias, and investor learning/review processes. | Investment Thesis history；Evidence needed |
Initial Research Queue
F01 → F02 → F09 → F13 is the editorial priority, not a dependency chain or a commitment to publish four essays.
The first actionable item is F01 · When is an AI investment study genuinely useful? (Issue #12). Use a worked case: whether multiple AI-sector funds cause overlapping exposure and whether the evidence supports holding, adjusting or waiting. A valid deliverable is a sourced research note with methodology, counterexamples, limits and an editorial decision—not a new publication by default.
What We Took from Three Industry Essays
- “What AI Features in Brokerage Apps Can and Cannot Do” raises F09/F14/F15: product functionality, service boundaries and investor reliance.
- “AI Trading Losses: Tools Without Work Processes” raises F02/F06/F08/F12/F15: domain error sets, guards, evidence, investor reviews, strategy tests and correlated judgments.
- “Why AI Adoption in Brokerages Is Harder Than Saying Yes” raises F03/F09/F11/F13/F14: business ROI, domain evaluation, substantial human review, employee workflows and approvals.
These user-provided industry essays are research starting points, not verified original statistics or regulations. Do not repeat claimed investor or institutional win rates, universal retention periods, filing rules or guaranteed monitoring accuracy without primary-source verification.
Exit Criteria for Each Task
Deliver a self-contained note with the question, dated references, financial definitions, competing explanations, one worked example, verified versus unknown claims and overlap with existing essays. End with an explicit deepen / new essay / defer decision. If a new essay is warranted, produce and validate both complete language versions under the existing publishing workflow. Tests or tool-call counts do not replace financial reasoning.