Observed arrival · 2026-08-27
What Went Wrong? — The AI Output Debugger
A diagnostic guide for deciding whether a bad AI result came from the instruction, the model, the task, the model choice, both sides, or no real failure.
Field notes
The site separates a user’s stated request from the intended constraints, the observable output, and the final verdict. Its six-way framework includes “wrong model,” “task blocked,” and “nothing wrong,” allowing disappointment to be treated as something other than model failure. Visible examples cover a replaced landing page, deleted unrelated code, and a generative image edit that altered the subject. The homepage also links to fuller cases and practical guides about instruction-following and code recovery.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue