Repeatability
High
The task structure is fixed: write regex patterns, build a Python script, process PDFs, output CSV, write tests. Each execution follows the same engineering pattern, even if invoice formats vary.
Ambiguity Tolerance
Medium
The output spec is fairly crisp (15-column CSV, 5 edge case tests, batch of 200+), but 'semi-formatted' PDFs are inherently variable and the exact regex patterns needed depend on real sample data the agent may not have upfront.
Data & Tool Availability
Medium
A code agent can generate the script and tests without sample PDFs, but producing robust regex patterns really requires access to representative invoice samples. Without them, the agent is writing to a hypothetical format.
Error Cost
Low
The output is a script and CSV — both easily reviewed, re-run, and corrected. No irreversible actions are taken; a bad regex just produces wrong output that a human can spot and fix.
Human Judgment Required
Low
This is a technical engineering task with no taste, ethics, or relationship context required. The main judgment calls — which fields to extract, how to handle malformed entries — are specified in the task brief.