Worked research guide · Sources and examples checked 17 October 2026 · AllMarkets Trading
A useful screening process needs more than a working API response or a list of matching symbols. It needs a defined universe, comparable metrics, timestamps and an appropriate use entitlement. A daily price file also needs its own quality checks before it is used in a chart or simulation.
Use the financial-data API directory to inspect documented quota units and use boundaries, the stock and ETF screener comparison for research workflows, and the original local tools to inspect files that you are entitled to use. This guide does not download or redistribute provider data.
Define the job before choosing a feed
A current stock snapshot, ETF holdings, historical daily prices and an economic release are different datasets. Write down geography, instrument identity, frequency, history, units, timestamp and adjustment basis. Public display or customer downloads can also be a different use from personal research.
Record the quota unit. A request may consume several credits because of endpoint weight or symbol count. Daily budgets, minute ceilings, bandwidth and streaming subscriptions are not interchangeable. Confirm the selected account entitlement before scheduling refreshes, and do not interpret an unknown limit as unlimited.
Map source fields deliberately
The local snapshot screener accepts a canonical CSV with user-declared comparable columns. A P/E ratio can be trailing or forward; growth can describe different periods; dividend yield can use different definitions. Rename or map columns only after reconciling those meanings.
Keep plain units explicit. A market capitalization of 5000000000 means five billion units of the dataset currency. Percentage fields use percentage points: 3 means 3%. An ETF’s assets under management are not automatically the same concept as equity market capitalization. The local screener uses one currency and performs no FX conversion.
Preserve missing values in a simple filter
Consider an invented minimum-yield threshold of 3%. A row with a supplied 3% yield passes that inclusive condition. A supplied 0% fails it, while a blank is unknown and also fails the active condition. With the yield filter inactive, that blank row remains available. The tool never turns missing data into zero.
A positive-P/E condition excludes negative, zero and missing P/E values. This can be appropriate for a specific research question, but a passing row is not an investment recommendation. Inspect financial statements, fund documents and the actual metric definition after narrowing a list.
Inspect a daily price file separately
The daily OHLCV inspector works on a single daily series. It reports impossible high/low relationships, malformed cells, repeated dates and source-order issues. It retains every accepted raw row and blocks sorted export while critical structural findings remain.
Calendar warnings are deliberately limited. A missing Wednesday label is a reason to investigate, not proof that an exchange was open. Weekday mode excludes weekends only and does not know holidays or suspensions. A large close-to-close move can span a gap or reflect a corporate action; the tool does not insert bars or rescale prices automatically.
Keep a portable decision trail
Save the applied filters, source definitions and the version/retrieval context that the applicable terms permit. Use full JSON restoration to preserve original strings and settings. A page showing 100 rows can still export all matching rows; read the printed range before assuming a PDF contains the entire dataset.
Public source access is not a blanket licence to cache, redistribute, display or use data in every AI workflow. Check the current product terms for the intended distribution. The local tools need no key and make no authenticated provider request, but that does not grant permission to obtain or republish a third-party file.
Do not rename screening as backtesting
A present-day match does not show what was known at a historical decision date. Financial revisions, delisted instruments, publication timing and the simulated execution model matter. Keep the separate historical-testing workflow distinct from a current filter result and from a structural data-quality report.