A useful retrospective starts with something that actually happened. For this first entry, the example is the October 10, 2026 work on this website. It is a publishing and configuration review, not an AI benchmark or a report of CommonGround experiments.
A repeated opening
The welcome article displayed a short excerpt above its body, followed by the same opening in full. Inspection showed that the article used a custom single-post template containing a 30-word Post Excerpt block. Removing that block corrected the single-post display while leaving the article body and archive excerpts intact.
The lesson was practical: find where the repetition is produced before rewriting the prose. A template-level display issue and an article-content issue require different changes.
Review instructions belong outside the page
During AI-assisted preparation, editorial review notes appeared inside a Projects draft preview. I caught them while watching the editing. They were removed from the page body and kept in a separate local review record. The Projects page remained a draft.
That distinction should have been clear from the beginning. Reader-facing copy and instructions for an editor have different purposes. Checking the rendered page helps catch mistakes that a successful save message cannot.
A selection is not a verified change
An initial AboutPage schema selection did not persist after saving. A separate save, reload, and inspection of the public structured data confirmed the corrected value. The visible selection alone would have been incomplete evidence.
The broader lesson applies to my software work: check the state that matters to the user. A local preview, an editor setting, and the published page can each show a different part of the process.
Keep the unfinished part visible
At the time of the review, Search Console reported the homepage indexed but also reported a sitemap fetch problem. Public retrieval of the sitemap succeeded. Those observations did not establish a root cause or prove that the Google report was resolved.
I want this kind of entry to preserve what changed, how it was checked, and what remains uncertain. The value is a record that helps the next decision. Future research retrospectives will need actual experiment records before they report findings.
Keep reading
Why I’m building in public · The welcome article · Testing an AI system
About Tim Hauptrief · Research notebook
Prepared with AI assistance. Featured illustration generated with AI.
Leave a Reply