SentinelGrid Supervisory Console — node 11 (Z3, human only)

audit streamconnecting… VeltoraCore… data source… models… feed runidle emergency stop…
binreading…
no case is waiting on a human decision.

Supervisory floor — five desks, one tape. Each reads the same bins through its own mandate, derives under its own login, and hands over on its own record.

reading the floor…

Agent mesh — 16 nodes, lit by audit events as they land (click a node to filter the trail)

no events yet
Z1 market (0 · 14 · 15)Sentinel swarm (1–4) / pipeline (8 · 9 · 10 · 12) 5 leader6 audit7 simulation only (dashed) 13 twin (one derivation / bin)11 human (Z3) approved / broadcastrejected / anomaly

Live bins — what node 6 relays to node 8: recorded marks, flow, and the four declared alert screens (D4 DEFAULT_THR)

Amber/red lines are editable only while the selected case is AMBER or RED; every edit is audit-logged as screen_threshold.edited and applies to the next tick.

VeltoraCore engine — one running service; the twin derives on a frozen bin, on demand, never per bin

…

Recent twin derivations

CaseCcySizeVerdictBracketSolvesSource

Cases — one case per tick; Approve/Reject appears at the human gate

Trigger a case by hand (no market feed)
CaseSevStatusOpenedActions

Case detail — select a case

Nothing selected.

Audit trail — click a row for its payload

TimeEventActorBin

Agents — what this process runs, and which model each LLM agent is bound to

#AgentPromptModelLast seen

Broadcast stability signals

SignalCaseSevTypeAssetLiquidityConf

Baseline vs SentinelGrid — §24-A scoreboard, last sentinelgrid demo run

MetricBaselineSentinelGrid