Pdf Document Structured Json Extraction

Deterministically extracts and validates structured JSON records from inline documents against a customer-supplied JSON Schema (a 2020-12 subset). The Actor never makes a network request: it reads only the INPUT you provide and writes to the dataset and the OUTPUT key.

Input

Set mode to inline (the only supported value) and provide jsonSchema and documents (1-100). Each document has an id, a contentType and exactly one of text, json, records or base64.

• text/plain, text/html, text/csv: Label: value lines (or the first CSV row) are matched to schema property names or your fieldConfig aliases.
• application/json: keys are mapped directly onto schema properties (case-insensitive, aliases supported).
• application/vnd.pdf-layout+json: a records array of { text, page, boundingBox } pre-extracted from a PDF; page and bounding box are carried into the evidence.
• application/pdf (base64): not parsed in this build; returned as UNSUPPORTED. Supply already-extracted text, json or records instead. OCR is not available.
• Other options: fieldConfig, includeEvidence, minimumConfidence, failOnInvalid. validationOnly and templateId are accepted but informational only.

Supported schema keywords: type, properties, required, additionalProperties (boolean), enum, const, format (date, date-time), pattern, string/numeric/array limits, items, uniqueItems, anyOf, oneOf, allOf.

Output

Each document yields one dataset row: documentId, fileName, status (SUCCEEDED, INVALID, UNSUPPORTED, FAILED), record, schemaValid, validationErrors, missingFields, ambiguousFields, fieldEvidence (confidence, sourceText, method, page, boundingBox), warnings, processing. The OUTPUT record holds { total, succeeded, invalid, unsupported, failed, items }, where items equals the dataset rows.

Pricing

No pay-per-event pricing is configured in this build; you pay standard Apify platform compute and storage. Work is in-memory string/JSON processing with no browser, proxy or external API, so typical runs are very cheap.

Limits

• Max 100 documents, 100 MB total input, 25 MB per document, 256 KB serialized schema.
• Nesting depth 8, 100 keys per object, 1000 items per array (applied to schema, fieldConfig, json and records).
• Zero network requests.

Failure behaviour

• Invalid input (wrong mode, missing or oversized schema, malformed documents, unknown properties) fails validation before processing and exits non-zero; see sample-invalid.json.
• One bad document (malformed base64/JSON, missing content) becomes a FAILED row and does not stop the others. PDFs become UNSUPPORTED rows.
• Missing, ambiguous or type-invalid fields make the row INVALID with details. The run fails only when failOnInvalid is true, after all results are stored.