Safe CSV Import Doctor

Choose one CSV file. The browser strictly decodes it as UTF-8, parses the table, and inspects every parsed header and data cell.

Review the finding count and diagnostic table. High findings deserve immediate attention, Review findings can change import meaning, and Info findings describe notable structure or formatting.

Choose a formula treatment only after deciding whether the export is intended for spreadsheet viewing or machine ingestion.

Choose whether the repaired export should contain a UTF-8 byte-order mark, then download both the repaired CSV and JSON report.

The repair process is deliberately structural rather than semantic.

Headers can be trimmed, named when blank, made unique, or extended when data rows contain additional fields.

Short rows receive empty trailing cells so every exported record has the repaired table width.

No data column is shifted, guessed, deleted, converted to a number, or converted to a date.

Leading-zero identifiers, long integers, scientific-notation-like values, and date-like text are reported but preserved.

The doctor does not decide whether 00123 is a postal code, whether 1E10 is a measurement, or whether 01/02/03 uses day-first or month-first notation.

Those decisions belong to the source owner and the schema of the destination system.

CSV stores textual fields and record boundaries. It does not reliably declare that one column is text, another is a date, and another is a 64-bit identifier.

Quoting controls how separators, quotes, and line breaks are represented inside fields. It does not force a spreadsheet to preserve a numeric-looking value as text.

The same CSV can therefore be interpreted differently by a spreadsheet, database wizard, programming library, or marketplace importer.

The file is read as bytes and decoded with a fatal UTF-8 decoder.

An invalid byte sequence stops the workflow instead of being silently replaced, because replacement can change names, identifiers, addresses, and other meaningful data.

The tool does not guess a legacy encoding. Convert from the known source encoding using a controlled process and retain the original file.

A literal Unicode replacement character can exist inside otherwise valid UTF-8. It may be intentional, but it can also show that an earlier conversion already discarded an unknown character.

NUL characters are reported as High findings because they are unusual in ordinary CSV text and can produce inconsistent behaviour across import tools.

The report gives coordinates without copying the complete data-cell value.

The first parsed row is treated as the header.

Surrounding whitespace is removed. Blank headers become column_n. Case-insensitive duplicates receive _2, _3, and later suffixes.

When a data row is wider than the source header, generated names extend the table so no parsed source field is discarded.

Every header action is recorded. The summary distinguishes affected header columns from the number of repair actions.

A short row contains fewer fields than the original header. Empty trailing fields are added during export.

A wide row contains more fields than the original header. The repaired header expands to preserve all parsed fields.

The doctor never guesses that an extra value belongs in a different position. A width mismatch can result from a legitimate optional field, an incorrect delimiter, an unescaped separator, or malformed quoting, so the source still requires review.

Rows containing no non-whitespace values are reported as informational findings.

They remain in the repaired CSV because deleting a record can alter row alignment, audit references, or destination-specific meaning.

Remove them only when the source owner has confirmed that they are not significant.

Spreadsheet programs can interpret some CSV cells as formulas rather than literal text.

The scanner reports equals, plus, minus, at sign, tab, carriage return, line feed, and supported full-width variants at the start of a cell after ordinary leading spaces are ignored.

The check applies to both headers and data cells because either can be opened in a spreadsheet.

A diagnostic identifies the coordinate and prefix without embedding the complete source value.

No formula-prefix treatment is universally safe for every spreadsheet and downstream parser.

An apostrophe or tab can prevent interpretation in a particular spreadsheet workflow, but it also changes the underlying text.

A file intended for database ingestion may need exact source characters, while a file intended only for human spreadsheet viewing may justify a spreadsheet-specific treatment.

The tool therefore begins in warn-only mode and records every optional formula repair when another mode is selected.

A negative number starts with a minus sign, which is also one of the spreadsheet-active prefixes.

The scanner cannot determine whether −10 is trusted numerical data or attacker-controlled text.

Review the expected column type. Applying a text prefix to a legitimate numerical column can prevent later arithmetic or break a machine importer.

Values such as 00123 are often identifiers rather than quantities.

Spreadsheet software may remove the initial zeros when it interprets the field as a number.

The repaired CSV preserves the original characters, but the destination column still needs to be imported or configured as text.

Common examples include SKUs, postal codes, phone numbers, account codes, employee IDs, and fixed-width reference numbers.

Excel documents a maximum numerical precision of 15 significant digits.

A 16-digit or longer identifier can therefore be changed when it is interpreted as a number, even when every digit appears correctly in the source CSV.

The scanner reports signed and unsigned integer-looking values containing at least 16 digits. It does not add punctuation or change the source string.

Text such as 1E10 can represent a valid numerical value or an identifier that merely resembles scientific notation.

The tool reports the pattern and leaves it unchanged.

Choose a numerical destination type only when exponential notation is intentional. Otherwise import the column as text.

Numeric dates can depend on day-first, month-first, and two-digit-year conventions.

A value is reported as ambiguous when both initial components can represent a month and day or when the year uses two digits.

Unambiguous but date-like numeric and month-name forms are reported as Info because spreadsheet software may still convert them automatically.

No date is rewritten, assigned a timezone, or converted to an internal serial number.

The metadata records CRLF, LF, CR, mixed, or absent line endings.

Mixed line endings are informational because the repaired serializer may use one normal export convention.

The shared parser determines the delimiter. The repaired CSV preserves that parsed delimiter.

A spreadsheet sep= directive is reported because parsers disagree on whether it is metadata or an ordinary first record.

A UTF-8 byte-order mark can help some spreadsheet programs recognize UTF-8.

Some database, shell, and programming workflows expect UTF-8 without a BOM.

The export checkbox initially matches the source file rather than adding a BOM to every output automatically.

Changing the setting is an explicit export decision and is recorded in the JSON report.

High means the pattern can create unsafe spreadsheet behaviour or invalid textual content.

Review means the file may import with a different structure or meaning than expected.

Info describes a property that may be relevant without proving an error.

Priorities organize review work; they do not replace knowledge of the source and destination.

The same text can be harmless in one destination and damaging in another.

The scanner reports patterns rather than declaring a universal corrected value.

Use the examples to choose an import type and review policy.

The report contains source filename, byte count, generation time, parsed delimiter, selected export settings, metadata, summary counts, header repairs, formula repairs, and stored diagnostics.

Data-cell diagnostics contain coordinates, category, code, priority, column name, and explanation without copying the complete source cell.

Header repair entries include original and repaired names because those changes need to be traceable.

When detailed diagnostics reach the browser-safety cap, aggregate counts still include all detected patterns.

The doctor does not know required columns, primary keys, unique constraints, accepted codes, foreign keys, numerical ranges, locale, timezone, or application-specific limits.

A structurally clean file can still contain missing records, incorrect values, duplicated entities, invalid references, or incompatible business meaning.

Run a controlled test import and validate the result inside the destination before replacing production data.

File reading, fatal UTF-8 decoding, CSV parsing, inspection, filtering, repair, serialization, and report generation run locally in the browser.

The source file and parsed values are not uploaded or placed into a shareable URL.

Downloaded files are generated only when the corresponding action is selected.