0mn1

Where the moat is built. Per-tenant feedback tunes its own agents automatically (Lessons, below); cross-tenant lessons are promoted here — and ONLY here — after you abstract & sanitize them by hand. Owner feedback never auto-generalizes.

Two tenant rules that overlap, contradict, or sit in tension with each other — flagged by distill, promote, sweep, or load time. Dispose of each: keep as-is, replace the loser, split by scope, or kick to the owner.

Where each promoted directive actually landed — per-destination status across the fleet.

Product signals clustered across tenants — the second pipeline in this workbench, next to rule-promotion above. Breadth (tenants asking) outranks depth (times asked); move each cluster through triage as it's handled.

Every note S4ge distilled sits here as a pending candidate until you promote, edit, or reject it — scoped to its own tenant, not the fleet. Compliance-quarantined items need human/legal review and can never be promoted.

Below-threshold observations — visible, never applied. A situational note needs 3 concept-matched instances (or one deliberate confirmation) before it becomes a real candidate.

This tenant's graduation dial is at Rung 1 for this rule class — the owner confirms in their own dashboard, not here. You can still pull a proposal back before they see it.

Every tenant, always glanceable. Opening a tenant's dashboard is a plain new-tab link to that tenant's own live surface — there is no reason-bound, time-boxed, or audited drill-in session today; that's a real gap, not a design choice, flagged for a future build.

Agent health, the governed-action log, the R0n sandbox, and live-surface launchers — condensed onto one page. Read-only except R0n (synthetic sandbox only, no real sends).

Last full health sweep and the platform-wide error rate it measured. Auto-resolves server-side; nothing to act on here.

Drives the real agents against synthetic tenants. No live SMS/email, no real charges — every run auto-archives.