Structured Data Extraction & Audit Workflows

Cómo usar esta colección

This collection guides you through a rigorous data extraction workflow using three distinct prompt templates. It is designed for researchers and analysts who need to convert messy text into structured, verifiable data without hallucinating facts. **Step 1: Scientific Metric Extraction** Use the **Scientific Data Extractor Prompt** (UUID: c7226f4e...) to process academic abstracts. This tool forces the AI to extract raw numbers, units, and p-values into a Markdown table. *Prerequisite:* A scientific abstract containing numerical data. *Limitation:* It does not interpret statistical significance; it only records what is written. If the abstract lacks numbers, it will state so explicitly. **Step 2: Financial Date Auditing** Use the **Utility Bill Date Extractor** (UUID: 0f1c93cc...) to handle financial or scheduling text. This prompt builds a due-date table but strictly flags ambiguous dates (e.g., '07/22/2025') and forbids any payment actions. *Prerequisite:* Text copied from bills or invoices. *Limitation:* It cannot calculate dates based on terms like 'due upon receipt' unless the specific calendar date is present. It serves as a review list, not a payment scheduler. **Step 3: Evidence Indexing** Use the **Question-to-Source Index Builder** (UUID: 6a73a4e6...) to map research questions to exact source excerpts. This ensures your final report relies only on evidence actually present in your materials. *Prerequisite:* A list of numbered research questions and a reading list with pasted excerpts. *Limitation:* It does not browse the live web. It works solely on the text you paste, marking any gaps where sources do not answer your questions. **Workflow Summary:** Start with Step 1 to get raw metrics. Use Step 2 if you are dealing with administrative dates or financial constraints. Finish with Step 3 to ensure your conclusions are traceable to specific source texts. All outputs are draft tables requiring human verification before use in production reports.