Pull specific named fields out of a document without training a model using Amazon Textract Queries (FeatureTypes QUERIES with QueriesConfig) and alias-to-answer mapping

Documented steps

1. Grant textract:AnalyzeDocument for sync, or textract:StartDocumentAnalysis and textract:GetDocumentAnalysis for async, depending on page count.
2. Call AnalyzeDocument (or StartDocumentAnalysis for multi-page) with FeatureTypes including "QUERIES" and QueriesConfig={"Queries":[{Text, Alias, Pages}]}: https://docs.aws.amazon.com/textract/latest/dg/analyzing-document-text.html
3. Write each Query.Text as a natural-language question, 1-200 characters, restricted to the documented alphanumeric/punctuation pattern (API_Query.html).
4. Set Query.Alias (1-200 chars, same character restrictions) to a short stable field key such as INVOICE_NUMBER so responses are easy to map.
5. Optionally set Query.Pages, an array of strings like ["1-3","4-*"], default ["1"], where "*" means the last page, to scope which pages each query runs against.
6. Stay within the query-count limit: 15 queries per page for synchronous AnalyzeDocument, 30 queries per page for asynchronous Start/GetDocumentAnalysis.
7. Parse response Blocks for BlockType="QUERY" (carries the Text/Alias you sent) and BlockType="QUERY_RESULT" (carries the answer Text and Confidence).
8. Link each QUERY block to its answer via its Relationships entry of Type="ANSWER", whose Ids point at the QUERY_RESULT block.
9. Build an alias->answer map by reading each QUERY block's Alias and resolving its ANSWER relationship to QUERY_RESULT.Text.
10. Apply the phrasing guidance in https://docs.aws.amazon.com/textract/latest/dg/bestqueries.html: use the document's exact wording, ask well-formed questions, and add positional or header context when several similar values appear on a page.