Repeatability
High
The task runs on the same CSV schema every month across the same 18 clients and three platforms. The logic — aggregate, calculate MoM growth, compute engagement-per-post, flag anomalies — is structurally identical each cycle, making it highly automatable.
Ambiguity Tolerance
Medium
Aggregation and growth-rate calculations are crisply defined, but 'anomaly' thresholds (what counts as a suspicious spike or sudden drop) require a human to set the rules once. After that initial calibration, success criteria are clear enough for an agent to execute without further guidance.
Data & Tool Availability
High
All inputs are structured CSV files with a known schema, and the output is a clean CSV plus a flagging report — no live API access or account permissions needed. As long as the agent receives the monthly file drop, it has everything required.
Error Cost
Low
The output is an internal reporting artifact reviewed by humans before client delivery, so a calculation error is catchable before it causes harm. Mistakes are reversible by re-running the pipeline with corrected logic.
Human Judgment Required
Low
The task is arithmetic and rule-based pattern detection — no taste, relationship context, or ethical judgment is needed. A human should review flagged anomalies before client-facing use, but the detection and aggregation work itself requires no human intuition.