What the frozen film taught me
While making this site, a change left the camera journey frozen behind its posters. The video still had frames and a duration. Its file checks were accurate, but they could not show whether scrolling the page moved through those frames. I reported the frozen result, and the implementation was repaired.
The follow-up opened the page and scrolled it. That reached the failure where a visitor would encounter it. It also gave the reviewer a specific question to answer: did the camera move with the page?
When reviewers disagree
A reviewer who inherits the assumption behind that film check could confirm the same mistake. I give a fresh reviewer the artifact and criteria without a persuasive account of why I think it is already good. Another model can offer a different perspective, but its findings still need evidence. If reviewers disagree, I examine the disagreement; a finding that does not hold up should not create another round of work.
The Receiver Reliance review had a different task: independently recompute all six estimates and bounds and examine twelve predetermined traces. Those internal checks helped me inspect the research. They were not external peer review, and the published record says so.
Checking the result people will use
I direct the work through AI and remain responsible for what I say about it. The film needed a browser check; the study needed a path from each claim to its source and limitations. In both cases, a convincing summary was something to verify.
