Control Tower

The Control Tower is Beacon’s operations view: a single, auto-refreshing screen that shows the real-time health of every subscription — success rates, run and failure counts, open tasks, and anomaly activity — so you can spot failing or silent monitors before your users do.

Table of contents

Overview

Every Subscription in Beacon runs a query on a schedule. The Control Tower aggregates the execution history of all subscriptions over a selectable time window and answers three operational questions at a glance:

  • Is it working? — success rate and run/failure counts per subscription, plus a health status (Green / Amber / Red / Stalled)
  • Does it need attention? — open (unresolved) task counts per subscription
  • Is the data behaving? — anomaly counts and per-day sparklines from Anomaly Detection

Open it at Control Tower in the left navigation (/control-tower).

Health Statuses

Each subscription gets exactly one status, computed from its execution history within the selected time window (default: last 30 days).

Status Meaning Rule
Green Healthy Success rate ≥ 90%
Amber Warning Success rate ≥ 70% and < 90%
Red Critical Success rate < 70%
Stalled Silent Zero executions in the window, and the subscription was created more than 1 day ago

Two edge cases are worth knowing:

  • New subscriptions stay Green. A subscription with zero executions that was created within the last day is treated as Green (“waiting for first run”), not Stalled — a 1-day grace period prevents false alarms right after creation.
  • Stalled is independent of the past. A subscription that ran perfectly last quarter but has zero executions in the selected window shows as Stalled. Silence is a failure mode.

What counts as a successful execution

The success rate is successful executions / total executions × 100, where an execution counts as successful if its notification status is one of:

  • NotificationSent — the query ran and a notification went out
  • NoResults — the query ran and returned nothing to alert on
  • NotificationSilenced — the query ran but the notification was deliberately silenced

Every other outcome (Failed, Timeout, Created, BelowThreshold) counts against the subscription. The failed count shown in the UI is simply total − successful.

Both the health list and the KPI statistics use exactly the same status calculation, so the summary cards always agree with the rows below them.

KPI Cards

The top of the page shows six KPI cards, all scoped to the current filters and time window:

Card Value
Total Number of subscriptions matching the current filters
Healthy Green subscriptions, with the overall success rate (all successes ÷ all executions) as the subtitle
Warning Amber subscriptions
Critical Red subscriptions
Stalled Subscriptions with no runs in the window
Open tasks Total unresolved tasks across matching subscriptions, with the total anomaly count in the window as the subtitle

Statistics are cached server-side for 30 seconds per unique filter combination, matching the UI’s refresh interval. When there are no executions at all, the overall success rate reports 100% rather than 0% — an empty system is not a failing system.

The Health Table

Below the KPIs, every matching subscription appears as a row:

Column Content
Status pill Green / Amber / Red / Stalled
Subscription Query name, folder path, plus badges for an assigned AI actor and enabled anomaly detection
Success Success rate in the window (— when there are no runs)
Runs Total executions, with the failed count highlighted when > 0
Tasks Open (unresolved) task count as a warning pill, or the total task count when nothing is open
Anomalies Sparkline of anomaly events per day across the window, with the window total next to it
Last run Outcome pill (Sent, Failed, Timeout, No results, …) and relative time of the most recent execution — never if it has not run

The table is paginated at 50 rows per page.

Sorting

By default the table sorts worst-first: most open tasks first, then lowest success rate, then name — the subscriptions that need attention float to the top. Clicking a column header switches to explicit sorting by name, success rate, runs, open tasks, anomalies, or last run.

Row detail

Clicking a row opens a detail panel for that subscription showing:

  • The query name, folder path, and cron schedule
  • Recent executions — the last 20 runs with outcome, timestamp, row count, execution time in milliseconds, and the error message for failed runs
  • Open tasks — up to 20 unresolved tasks with priority, assignee, and snooze state
  • Recent anomalies — up to 20 anomaly events in the window with severity, detected value, explanation, and acknowledged state

From the panel you can jump straight to the underlying query.

Filtering

All filters combine, and apply to both the KPI cards and the table:

Filter Options Behavior
Search Free text Matches against the query name
Folder Any query folder Shows only subscriptions whose query lives in that folder
Health All / Green / Amber / Red / Stalled Shows only subscriptions in that status bucket
Time range 24h / 7d / 30d / 90d The execution window used for all counts, rates, statuses, and sparklines (default: 30d)
Open tasks only Checkbox Shows only subscriptions with at least one unresolved task

Changing any filter resets the table to the first page. A Clear action resets everything back to the defaults.

The time range changes what “healthy” means. A subscription that failed once yesterday might be Red over 24h but Green over 30d. Use a short window for incident triage and a long window for trend review.

Auto-Refresh

The Control Tower is designed to be left open on a screen:

  • Data auto-refreshes every 30 seconds and also refetches when the browser tab regains focus
  • A manual Refresh button forces an immediate reload
  • If a selected subscription drops out of the refreshed result set (for example, it no longer matches the active filters), its detail panel closes automatically

API

The Control Tower is backed by three read-only endpoints under /beacon/api/control-tower, using the same cookie authentication as the rest of the REST API.

Endpoint Purpose
GET /beacon/api/control-tower/statistics Aggregate KPI counts for the current filters
GET /beacon/api/control-tower/health Paginated per-subscription health rows
GET /beacon/api/control-tower/subscriptions/{id}/detail Executions, open tasks, and anomalies for one subscription

Query parameters

Both /statistics and /health accept the same filter parameters:

Parameter Type Default Description
timeRangeDays integer 30 Size of the execution window in days
folderId integer — Restrict to one query folder
dataSourceId integer — Restrict to one data source
healthStatus enum — Green, Amber, Red, or Stalled
hasUnresolvedTasks boolean — true = only subscriptions with open tasks
searchKeyword string — Substring match on the query name

/health additionally accepts page (default 0), pageSize (default 100), and sortBy (Name, SuccessRate, Executions, OpenTasks, Anomalies, LastExecution, or the default WorstFirst).

Example

Illustrative request and response (values are examples, not real data):

curl "https://your-beacon-host/beacon/api/control-tower/statistics?timeRangeDays=7&healthStatus=Red" \
  --cookie "Beacon.Auth=..."

{
  "totalSubscriptions": 3,
  "healthySubscriptions": 0,
  "warningSubscriptions": 0,
  "criticalSubscriptions": 3,
  "stalledSubscriptions": 0,
  "totalUnresolvedTasks": 5,
  "totalAnomalies30Days": 2,
  "overallSuccessRate": 41.7,
  "timeRangeDays": 7
}
  1. Keep the default sort. Worst-first puts subscriptions with open tasks and low success rates at the top — start there.
  2. Triage Red and Stalled first. Red means runs are failing; Stalled means nothing is running at all. Click the row and read the last error message in the recent executions list.
  3. Use “open tasks only” for follow-up. After an incident, filter to subscriptions with unresolved tasks and work the list down.
  4. Watch the sparklines. A rising anomaly sparkline on a Green subscription means the runs succeed but the data is drifting — see Anomaly Detection.
  5. Widen the window for reviews. Switch to 90d during a periodic health review to catch slow degradation that a 30d window smooths over.
  • Subscriptions — schedule the queries the Control Tower monitors
  • Tasks — the open (unresolved) tasks surfaced in the table and detail panel
  • Anomaly Detection — the source of the anomaly counts and sparklines
  • Notifications — the delivery outcomes that feed the success-rate calculation
  • Queries — the queries behind every subscription