Repeatability
High
Every invoice follows the same extraction pattern: vendor, date, amount, line items. The structure repeats across all 280 files, making this highly amenable to a single automated pipeline.
Ambiguity Tolerance
Medium
Output columns are clearly defined, but category assignment (software/hardware/services/other) requires interpretation of line item descriptions that may be vague or vendor-specific. Success is mostly crisp, with a narrow gray zone.
Data & Tool Availability
High
Files are self-contained PDFs totaling under 15 MB — easily ingested by OCR and PDF-parsing tools. No external APIs, logins, or live data sources are required.
Error Cost
Medium
Errors in extracted amounts or dates could cause accounting mistakes, but the output is a CSV that a human can spot-check before use. Mistakes are reversible and detectable with a basic review pass.
Human Judgment Required
Low
The task is data extraction, not interpretation. Category assignment is the only judgment call, and a four-bucket taxonomy is simple enough for an LLM to handle with high accuracy on most invoices.