Find errors and anomalies in a dataset

The data (pasted below or attached):
"""
{{data}}
"""

What each column should contain (type, allowed values or range, which column should be unique, which must never be blank):
"""
{{column_rules}}
"""

Where the data came from: {{data_source}}
What I will use it for: {{intended_use}}
App I am working in: {{app_and_version}}

Task: Check this data for errors and anomalies before I use it for {{intended_use}}. I need to know whether the numbers can be trusted, so that problems are fixed at the source rather than discovered in the results. This is a check, not a cleanup: flag what looks wrong and leave every value as it is.

Check for:
- Duplicates: exact repeats, near-repeats that differ only in spaces, capitals or punctuation, and repeated values in the column that should be unique.
- Missing values in required columns, including blanks disguised as "N/A", "0", "-" or a single space.
- Invalid values: outside the allowed range or list, negative where that is impossible, dates in the future, end dates before start dates.
- Signs of import damage: default dates such as 1900-01-01 or 1970-01-01, dates showing as numbers, long IDs shown in scientific notation, codes that have lost their leading zeros, numbers stored as text.
- Inconsistencies: totals that do not equal their parts, mixed units or currencies in one column, the same thing spelled several ways.
- Outliers: values far outside the rest of their column. Say how you judged "far", for example more than three times the median. An outlier is a value to check, not proof of an error.
- Test or dummy records, such as "test" or "asdf" in a name field.

Format:
1. Summary: rows and columns checked, and whether the data looks safe to use as it is, in up to 3 sentences.
2. Issue log: a table with the columns Issue, Column, Rows affected (row numbers or IDs), Example values (copied exactly), Severity (stops the analysis, distorts results, or cosmetic), How to confirm.
3. Formulas or filters for {{app_and_version}} that find each type of issue in the full dataset, with one line on what each does.

Do not correct, delete or fill any value, and do not guess what an error should have been. If you could see only part of the data, say how many rows you checked. If a column's meaning is unclear, say so instead of judging its values. Keep the text outside the table under 300 words.