Engineering Journal

Public decision records, not notes.

Every entry follows the same six steps. The structure is the point: it forces a decision to state what it saw, what it traded, and whether it would hold up again.

1. 1Question
2. 2Observation
3. 3Tradeoffs
4. 4Decision
5. 5Outcome
6. 6Would I do this again?

1. 01How should humans operate AI systems?Running 13 sites on a 1970s mail protocol→
2. 02Manuscript in progressKeeping LLM output quality flat: harness engineering→
3. 03RPM on the biggest site dropped from $8 to $4. Nothing was deployed... or was it?Rendering is a revenue variable: the ISR incident→
4. 04Manuscript in progressA first-party analytics platform for 13 properties→
5. 05Manuscript in progressThe migration: platform risk, canonicals, and an index incident→
6. 06Why can't good models reach production?2020: a sentiment model that never shipped→
7. 07Manuscript in progressIndexNow and crawl budget→
8. 08Why doesn't SEO reward effort?Six months of AdSense failure→