Calibrate video screening against feedback (Beta)

(Beta) Re-judges every creator marked good or bad fit with the automation's current video rules, then tries each hard filter at thresholds 0.1-1.0 with the other rules unchanged. summary.bad_fit.passes is how many bad fits the rules still approve; summary.good_fit.blocked is how many good fits they would stop. Each filter's suggested_min_share is a threshold that lets fewer bad fits through without losing a good fit (none until 3 good fits can be judged). Judged on each creator's latest analysed videos today, not those at decision time; creators at or above screen_below_gmv_30d are approved on numbers and not covered by video rules. Covers the newest 2,000 verdicts. Read-only: apply a change with automation_update.