The rubric's changelog

13 September 2026

Each finding now says which one of the published tests it is about, so a grill can show what passing that test would be worth to the score. A finding may also carry a second proposal when there is a genuinely different way to fix the problem.

9 September 2026

Findings are split into the problem (the words on your page and what they do to a stranger) and a proposal (what you could change; you decide). Each half is capped at twenty-five words, after early feedback that findings read as long and as orders.

7 September 2026

The critic no longer gives an axis a number. It records pass, partial or fail for each test, and the score is computed from those results the same way every time: the share of points earned, squared. One fail of five takes an axis to 64, two to 36. The same page now scores more consistently from one grill to the next.

6 September 2026

The one piece of web vocabulary the rubric relied on is now defined in plain words the first time it appears.

6 September 2026

The published part of the rubric no longer mentions data formats or field names. The two special results are named in English, could not see it and does not apply, and the rubric explains the scoring in the same words a grill prints.

6 September 2026

The grade, from Raw to Well done, is the answer to the question a founder asks: is this ready to launch? The one-line summary on every grill is the evidence for that grade, not a second opinion.

6 September 2026

A section describing an output layout the app never used was taken out of the rubric. Nothing about how pages are judged changed.

5 September 2026

LaunchGrill stopped asking for the launch post and stopped grading it. That axis's weight was spread across the other four in proportion, so a page scored without a post gets the same number as before.