03 · Writing
The best code is the code you never wrote.
Judgment case studies and field notes from production — what was removed or avoided, why, and the principle that carried out of it.
- I Built My Chatbot's Knowledge Like a Lakehouse The chatbot is a data product — so I gave it Bronze, Silver, Gold, and tests. −1 text dump 4 min
- Deleting a Dependency Instead of Duplicating It A cleanup ticket that ended in removing an SDK entirely. −1 dependency 3 min
- Dropping Security Code Because the Threat Didn't Exist Defensive code that protected against nothing. −1 attack surface 3 min
- Validate Once, Then Trust the Types Defensive code scattered everywhere is fear. A typed boundary is a decision. −scattered checks 3 min
- Counting Rows Without Paying for Them df.count() after a write is a second job. The numbers you want already exist. −1 full scan 3 min
- The Writer That Refuses to Guess saveAsTable will invent a table from a typo. writeTo won’t. 0 surprise tables 3 min
- Changing a Column Type Without Rewriting the Table With type widening, int → bigint is a metadata change, not a migration. 0 rewrites 3 min
- What mergeSchema Actually Merges It adds columns. It does not change types — that’s a different feature. +columns, not types 3 min