Different formats.
One clear destination.
Compare CSV, XLSX, JSON, HTML, documents, images, and prompt records—and identify the decisions that a file format cannot make for you.
Start with the source’s meaning
A filename extension identifies a likely representation, not a complete contract. Choose the parser and settings for the formats you deliberately accept, then define what the destination expects. Preserve identifiers, source references, and useful structure before applying business validation.
| Source | Useful for | Decide explicitly |
|---|---|---|
| CSV | One rectangular set of records | Delimiter, quoting, headers, encoding, and field types. |
| XLSX | An agreed worksheet in a workbook | Sheet selection, the data region, formulas, and date interpretation. |
| JSON | Structured records and nested values | Object shape, required fields, identifiers, and unexpected properties. |
| HTML | Authorized web content | Main content boundaries, links, assets, sanitization, and revision identity. |
| PDF / DOCX | Document content and metadata | Reading order, headings, tables, extraction uncertainty, and review. |
| PNG / JPEG / WebP | Image assets and variants | Accepted types, inspection limits, original identity, and derivative recipes. |
| Prompt records | Reusable application instructions | Message roles, declared variables, version, ownership, and activation status. |
CSV: agree on a dialect
The CSV description in RFC 4180 documents a commonly used quoting convention. Test embedded commas, doubled quotes, and quoted newlines. Do not assume that splitting every line on a comma is an adequate parser. Explain whether a header is required in your own workflow.
JSON: define the record contract
The JSON specification provides the interchange structure; the application still chooses its fields and semantics. Keep identifiers as strings when their exact representation matters. Separate absent values, explicit nulls, and empty strings according to the destination’s needs.
Documents and web pages: retain context
Text extraction and content approval are separate stages. Preserve useful heading relationships and source revisions, and flag layout that the extractor cannot interpret reliably. A successful parse does not prove that the resulting content is complete or ready for publication.
Images and prompts: model the lifecycle
An original image, its web derivative, and its page description are related but distinct artifacts. A prompt definition, its rendered representation, and an actual execution are similarly different. Keep the relationships explicit so replacement, evaluation, and rollback remain understandable.
Next step
Choose the relevant import category and use a small representative file to establish a destination contract. The local quickstart illustrates mapping and validation with fixed JSON records; it is not a general-purpose file parser.