Data Quality Checker reads a market analysis document or blog article and looks for five kinds of mistakes: prices with the wrong number of digits for the instrument (for example an ETF quoted at a futures level), inconsistent ticker notation, dates paired with the wrong day of the week, allocation tables that do not add to about 100 percent, and mixed units.
It handles English and Japanese text. Findings are advisory: the script flags them for a person to review and exits normally even when it finds issues. It runs offline on a local file.
What You Get
- A JSON report with severity, category, message, line number and context for each finding
- A Markdown report grouped by severity
- Date and weekday mismatches confirmed by calendar computation
- Explanations drawn from a notation standard and a catalog of common data errors
Common Use Cases
- A final check on a weekly strategy blog before it is published
- Reviewing an automated market summary for scale mistakes between ETFs and futures
- Checking a translated document for date and weekday accuracy
- Verifying that an allocation table adds up
What Makes It Unique
Price scale checks count digits before the decimal point and so keep working as prices change. Allocation checks look only inside allocation sections, so percentages in body text such as RSI or growth rates are ignored. Full-width Japanese characters and Japanese weekday labels are normalised before checking, and the year for yearless dates is inferred in a stated order of priority.
Under The Hood
The agent runs one Python script that uses only the standard library, reads the markdown file named on the command line and makes no network calls. It writes a timestamped JSON file and Markdown file into an output folder (reports/ by default), so it creates files on disk. No API keys are needed. The agent also reads two reference files to explain the findings.