Goal: Human moderators can review escalated reports, see AI confidence scores, take moderation actions, and have every decision recorded in the audit trail.
→ "A moderator can open the admin dashboard, see all pending and escalated reports with AI scores, take action (Remove / Warn / Ban / Dismiss) with a confirmation step, and see the report resolve and the queue counter decrement in real time."

Features in this milestone:

• #34 — Admin dashboard layout (sidebar, topbar, metrics row)
• #35 — Moderation queue with filter chips (All / Escalated / Spam / Hate speech etc.)
• #36 — ReportCard component (collapsed + expanded states, pipeline stepper, audit trail)
• #37 — AI confidence score bar (color-coded: red ≥ 0.85, amber 0.60–0.85, green < 0.60)
• #38 — Moderation action buttons (Remove content / Ban author / Warn author / Dismiss)
• #39 — Confirmation modal for destructive actions (with action summary)
• #40 — Moderator claim-lock (show "Moderator X is reviewing" to prevent double-handling)
• #41 — AI screening trigger (manual "Run AI screening" for unscreened reports)
• #42 — Metrics row (Pending count, Escalated count, AI auto-resolved, Escalation rate)
• #43 — Mobile-responsive moderation queue (bottom nav, card-based layout)
• #44 — useModerateAction hook + optimistic UI update

Definition of Done:

• All features merged to main
• Destructive actions (Ban, Remove) require confirmation before firing
• Optimistic UI updates queue count immediately on action — no waiting for server round-trip
• Claim-lock prevents two moderators from acting on the same report simultaneously
• Report status transitions match backend state machine exactly (no invalid transitions possible from UI)
• No TypeScript errors
• Tested on mobile viewport
• Backend team sign-off on moderation action endpoint and claim-lock mechanism

Target Date: YYYY-MM-DD
Owner: @username