Give your plan a room that disagrees.

The source checks four fixed keyword lists for evidence, edge cases, reversibility and trust. Each matching list gives one point, so two matching lists produce two out of four and all four produce four out of four. Bars use one hundred twenty pixels per matching list. Despite the interface wording independent review, this is literal regular expression presence, not an AI critic, actual test run, implemented safeguard or semantic evaluation of your plan. The single word rollback matches both the edge cases and reversibility keyword lists. One word therefore yields two points and Needs sharper defenses. Bars compare one entered word with two matched lists at one hundred fifty pixels per count. The literal fragment rate can also match inside an unrelated word such as celebrate. A PASS label means only a regex match and its accompanying text remains a question. It does not prove that a rollback, permission boundary or test actually exists. There are five possible totals, zero through four. Zero or one show Consensus risk is high, two or three Needs sharper defenses, and four Strong challenge coverage. Bars show two totals in the first bucket, two in the second and one in the third at one hundred fifty pixels per total. These phrases are source heuristic labels, not calibrated risk. Review stores the trimmed plan, four checks, points and a physical creation timestamp. Editing the draft alone leaves the last review and its Markdown export unchanged until Review runs again.