Continue merge PR if Dr. CI AI verdict deems breaking signals are unrelated

[georgehong]

1. We need to estimate how many of the existing AI verdicts are incorrect: this is a correlation of signals within the PR to ensure output is consistent.
   a. Ivan: run for a week and increase verdict rates
   b. (other option): incremental rollout like greenlight

[izaitsevfb]

draft PR: #8671

[izaitsevfb]

UI landed: #8671

mergebot soon to land: pytorch/pytorch#195503

[izaitsevfb]

final PR (replicator schema) #8825

before enabling the feature

[izaitsevfb]

looks like the suppression is globally broken: #8840

[georgehong]

Action Item/Follow-Ups: in 2 weeks, check on. the PRs that were part of this workflow cross-referenced with auto-revert cases. Add a metric in the dashboard (HUD/Grafana) that matches this.

• Centralized place to check follow-ups like this?
  • HUD, Grafana, agent analysis --> preferring a specific permanent dashboard.
• Existing mergebot monitoring/bypasses.
  • "tag PRs that are bypasses and landed."

[izaitsevfb]

dashboard added: https://hud.pytorch.org/metrics/ai_suppression