Repeatability
High
The same fixed set of fields must be extracted from each PDF, making this structurally identical across all 200 documents. Inconsistent formatting adds noise but doesn't change the underlying task logic.
Ambiguity Tolerance
High
Success criteria are explicit: nine named fields, one output CSV, and a quality flag for missing or estimated values. The agent can objectively determine when each row is complete or flagged.
Data & Tool Availability
High
The PDFs are the sole input required, and PDF parsing plus LLM extraction pipelines are mature and readily available. No external APIs, logins, or live data sources are needed.
Error Cost
Medium
Incorrect field extraction (e.g., wrong price or square footage) could mislead downstream decisions, but the output is a CSV that a human can audit, and the quality flag system surfaces uncertain rows for review. Errors are reversible with a spot-check pass.
Human Judgment Required
Low
Field extraction is largely pattern-matching with no taste, ethics, or relationship context required. Edge cases like ambiguous lot size units or merged fields are handled adequately by flagging rather than guessing.