Repeatability
High
SQL migration scripts follow well-established patterns: ALTER TABLE, UPDATE with string parsing, CREATE INDEX, ADD CONSTRAINT, and transaction blocks. The structure is highly repeatable and AI has seen thousands of examples in training.
Ambiguity Tolerance
Medium
The core requirements are crisp, but the format of the concatenated `customer_info` column is unspecified — the agent must assume a delimiter or pattern, which could be wrong. Success is otherwise verifiable: the script runs, constraints pass, indexes exist, rollback works.
Data & Tool Availability
Medium
The agent doesn't need live database access to write the script, but without seeing actual sample data from `customer_info`, it must guess the parsing logic. If the user provides a sample row or format spec, this becomes high.
Error Cost
High
Running a bad migration on 2 million rows in production can corrupt data or cause downtime. However, the task explicitly includes rollback logic and dry-run mode, which significantly mitigates risk — the real danger is if someone runs the script without reviewing it first.
Human Judgment Required
Low
The technical decisions here — index types, constraint syntax, transaction wrapping, runtime estimation — are well-defined engineering choices with established best practices. A DBA review pass is prudent but the judgment required is minimal.