Dashboard

Verdict Feed

The Verdict Feed is the chronological record of agent activity and enforcement decisions. Use it to audit what agents attempted, what the system decided, and where operator action is required.

Filter quickly

Slice the feed by status and agent identity to isolate a single behavior pattern.

Follow time

Read events in order and correlate spikes with deploys, incidents, or policy pack changes.

Escalate holds

When status is HOLD, the decision is waiting on an operator — resolve it in Approvals.

Verdict status reference

The Status filter groups events by outcome. These labels reflect what the system allowed to execute, what it paused for review, and what it refused outright.

StatusMeaningOperator expectation
ALLOWSafe actions permitted to execute.No action required.
HOLDActions paused for manual human-in-the-loop review.Review and decide in Approvals.
BLOCKActions denied due to policy violations.Investigate the triggering rule/red line; repeated blocks are a signal.
AUDITPassive logging for non-enforcement events.Useful for visibility; does not gate execution.

Anatomy of a feed item

Each row represents a single event and its verdict. Treat it as an audit log entry: the action text, who initiated it, when it happened, and what it cost to evaluate.

Action name

The descriptive intent of the agent — what it was trying to do.

Agent identity

The originating agent’s hex ID (public key) and any friendly label shown in the UI.

Latency

The performance cost for that specific verdict — useful for spotting degraded inference or network paths.

Timestamp

The exact time the event was recorded (useful for correlation and incident timelines).

Filtering logic

The feed is designed for fast isolation. Use the two primary controls together:

Status filters

  • All shows everything in time order.
  • Allow isolates actions that proceeded.
  • Hold isolates actions awaiting a decision.
  • Block isolates refused actions (often red lines).
  • Audit shows passive, non-gating events.

Agent dropdown

Use the Agent dropdown to focus on a single agent’s behavior. This is especially useful when you see repeated HOLD/BLOCK outcomes and want to confirm whether the pattern is isolated to one integration or systemic across the fleet.