Scheduled reports need a verified deployment
Daily Stock Analysis documents a useful workflow: analyze a watchlist, generate a decision dashboard and deliver it through a web workspace or notification channel. Our recorded run did not verify that workflow end to end. The distinction matters when deciding whether to depend on its daily schedule.
The English README describes mainland Chinese, Hong Kong, US, Japanese, Korean and Taiwan stocks, plus ETFs. Reports can include quotes, technical indicators, news, announcements, fundamentals, scores, entry or exit levels and risk alerts. These are documented capabilities, not workloads validated by our run. Market coverage and available fields depend on the underlying sources.
The project says its AI-generated analysis is not investment advice. Its buy, watch and sell labels should remain inputs to human research.
What our recorded run establishes
The stored run began on September 29, 2026, at 03:42:28.280 UTC, using commit d3fee51a9e5ebec756fe8184c1d13b2accf5e39f. The recorded configuration was 3 CPUs, 8 GB of memory and image ghcr.io/astral-sh/uv:python3.12-bookworm. The install output identifies a Python 3.12.12 environment.
The harness marked the install step successful after 49 seconds. However, the raw tail contains wheel metadata errors and an unsatisfiable Longbridge dependency requirement. It reports that longbridge>=4.0.5,<5 could not be used because available wheels did not match the recorded platform. This supports an environment-specific dependency-resolution problem, not a conclusion that the product is broken on every platform.
The dependency summary records 35 packages and 37 MB. Those figures do not establish a complete dependency installation. The build step returned exit code 0 after 15 seconds, with an empty retained tail. Without the command or build artifacts, that result cannot establish that the application or an installed binary worked.
The test command exited 4 after 10 seconds while loading tests/conftest.py. It attempted to import anyio.to_thread and raised ModuleNotFoundError: No module named 'anyio'. This was a test-setup failure; the record does not establish executed test cases or a passing-test count. It also does not establish a completed market-data request, model analysis, notification delivery or measured financial workload.
Pip-audit recorded 0 advisories, with unknown package coverage. That result cannot support a claim that the dependency tree is safe.
The repository parser recorded 1,167 files, approximately 434,404 source lines, 95.6 MB, 10 CI workflows and a tests directory. It detected neither a Dockerfile nor a Compose file. Those scan results describe this checkout and detector; they do not invalidate the README's Docker deployment documentation.
Configuration is part of the evaluation
The README's Actions instructions call for a watchlist, at least one model-provider configuration and at least one notification channel. They recommend a news-search service. Ollama is documented for local inference, so a cloud API key is not a universal requirement for every deployment route.
AkShare, Baostock and YFinance are documented free market-data sources that need no additional data-source credentials. The README warns that rate limits, upstream contract changes and network conditions can affect them. It recommends token-based sources such as Tushare or Longbridge for scheduled or batch work and steadier quotes. That recommendation is a project claim, not a reliability comparison measured here.
The current README lists Anspire, SerpAPI, Tavily, Bocha, Brave, MiniMax and SearXNG for news search. The README at our measured commit also listed TickFlow among market-data options; the supplied current README does not. Match configuration instructions to the revision you deploy.
The documented default Action runs at 18:00 Beijing time on weekdays and skips non-trading days. Verify the schedule against your markets and desired delivery time rather than assuming it follows each exchange's close.
Two issue reports need careful attribution
Issue 2355 describes a Windows 11 desktop v3.32.0 run in which capital-flow collection recorded 45,000 ms under a 45-second fundamental-stage budget. The reporter said increasing the fetch timeout did not change that duration. We did not reproduce it. The issue was automatically closed as not planned on October 9, 2026, after inactivity; closure does not demonstrate a fix.
A reply displayed under the owner account explicitly identifies itself as generated by OpenReview Bot. It suggests blocking AkShare calls and says an outer timeout may stop waiting without terminating the background request. Treat that as attributed diagnostic guidance, not our verified implementation finding.
Issue 2357 originally alleged that scheduled monitoring could not see rules created through the web interface. Its bot-generated reply disputes that diagnosis: it says the worker reads enabled database rules and merges legacy environment rules, with database rules taking precedence on duplicates. The same reply warns that deleting a database rule does not automatically delete an independent legacy environment rule.
The suggested check is to clear AGENT_EVENT_ALERT_RULES_JSON to an empty array, restart, and verify a rule created through the Web/API path. We have not verified that procedure or the cited fix in source code. This issue also closed automatically as not planned on October 9, 2026. It supports checking legacy-rule visibility, not stating as fact that the scheduler ignores database rules.
Features do not establish operational reliability
The README documents manual analysis, task progress, history, Markdown reports, backtesting, portfolio management, settings and strategy chat with 15 built-in strategies. The v3.32.0 release notes describe expanded index support, structured research artifacts, data-source quality interfaces and hard timeouts for scheduled analysis. These are upstream descriptions, not features exercised in our sandbox.
Before adopting the schedule, resolve dependency installation in the target environment, record the exact commands and revision, and verify one familiar symbol through data collection, report generation and delivery. Inspect missing fields and provider status, then check that scheduled alerts use the rules you intended. This is a proposed evaluation, not a test we performed.
How to read the scores
These are editorial evidence ratings, not benchmark results. The scale is 1 for major unresolved evidence or setup gaps, 2 for partial support with substantial verification still needed, 3 for adequate documented support with limits, 4 for detailed support across the relevant task, and 5 for strong documentation plus direct verification of that task. Unknown adoption, compatibility and runtime behavior receive no assumed credit.
Setup scores 1/5 because installation remains unverified. Documentation scores 4/5 for the supplied English setup, provider, scheduling and limitation guidance, without claiming every linked guide was inspected. Community scores 2/5 because the extracts show reports, replies and release contributors but do not establish response quality or sustained support. Maturity scores 2/5 because the documented workflow is broad while our runtime evidence is incomplete.
For agent-role experiments, TradingAgents is the relevant framework comparison. For building a financial-data foundation, OpenBB is the relevant component comparison. Neither comparison establishes an easier deployment or better investment results.