How each AI analyst behaves over the published weeks: gap to the consensus, stability, corrected scores and invalid answers, recomputed from the public reports.