AI-Written pandas: Two Missing IDs Can Make a False Match
A pandas merge can match missing IDs and duplicate order rows. See a tested example and a way to retain every order while keeping uncertain matches visible.
Tag: Practical AI
AI-Written Python: Test Negative Numbers Before You Trust the Buckets
Floor division and truncation agree on many positive examples. Test both sides of zero before trusting AI-written Python to group events into time buckets.
AI File Inventories: A Symbolic Link Is Not Another Copy
A Python file check can follow a symbolic link and count its target again. Make an AI-assisted inventory distinguish files, links and unavailable destinations before handover.
AI-Written Time Reports: Two Clock Hours Can Mean Three Elapsed Hours
Time-zone-aware timestamps can still produce a surprising duration. A tested Python example shows why elapsed-time reports need an explicit calculation rule.
AI-Written Python: An Empty List of Checks Can Still Pass
Python’s all() returns True for an empty collection. Give missing validation results their own status before an AI-written script declares success.
AI-Written CSV Imports: Two Columns Can Become One Field
Duplicate CSV headings can overwrite a value when Python builds a dictionary. Check column names before an AI-written importer processes the records.
AI-Written Python: Counting a Stream Can Leave Nothing to Total
A report can count three records and still show a zero total if an earlier step consumed its iterator. Test the complete reading path, including previews.
AI-Written Date Imports: A Missing Year Needs a Decision
A day and month can become a complete-looking date with an unintended year. Define what the record means before asking AI to fix the import.
AI-Written SQLite: Check That the Relationship Is Enforced
A relationship in a SQLite schema is not enough. Test that every writing connection rejects records linked to a customer who does not exist.
AI-Written Python: When a JSON Export Contains a Number JSON Cannot Read
Python can write NaN and Infinity by default. Decide what an unavailable calculation means, then test that the receiving tool can read your JSON export.