Monitors: autonomous schema monitoring

A Monitor is a saved, scheduled query with thresholds and a delivery channel - the way Colrows watches your data and semantic layer for change. Because monitors compile through the same governed pipeline as every other query, their alerts are grounded in governed metrics and carry a full trace of what changed and why. This page covers what a monitor is and how to configure one.

Standard data logs vs. Colrows Monitors

Why Monitors run through the compiler

The drift trap

Traditional data monitoring watches for broken pipelines: failed ETL jobs, null columns, schema errors. It misses the far more dangerous problem: semantic drift. A column name stays the same, but the business logic inside it changes. A dimension value shifts meaning across regions. A metric definition is updated in the warehouse without updating the analytics layer. Standard logs report "all systems healthy" while your AI agent generates silently wrong answers from stale definitions.

The Monitor advantage

Colrows Monitors track the semantic contract, not just the pipeline status. If an upstream schema change invalidates a metric definition, the system alerts or auto-corrects before the agent generates an inaccurate query. Monitors compile through the same governed execution pipeline as every other Colrows query: intent resolution, constraint validation, join path proof, compile-time governance. When a Monitor fires, it carries the full trace of what changed, why it matters, and what the system did about it. This is how you move from reactive monitoring to autonomous semantic integrity.

Do not monitor your data for uptime alone. Monitor for semantic integrity. Fix the context, not the model.

What a Monitor is

A Monitor is a saved query plus a schedule, optional thresholds, and a delivery channel. When the schedule fires, Colrows compiles the query through Consensus, evaluates the threshold against the result, and routes a structured payload to your channel of choice.

Because Monitors run through the compiler, they inherit every Colrows guarantee: governed metrics, persona-scoped access, point-in-time reproducibility, and an audit trail per invocation.

General settings

• Name & owner - every Monitor has a single owner. Owners receive operational notifications (compilation failure, schedule skew).
• Persona - Monitors run as a persona, not as the user who created them. This is how you keep schedule-triggered jobs reproducible after a teammate leaves.
• Source - a saved query, a metric, or a natural-language question.
• Tags - used for routing and dashboards.

Stopclock - debouncing & back-off

Stopclock is the configurable circuit breaker for Monitors. It prevents a noisy Monitor from paging on every small fluctuation.

Result - how Monitors deliver

Every fire produces a structured payload:

{
  "signal_id":   "sig_8a2c…",
  "fired_at":    "2026-04-29T08:15:00Z",
  "trigger":     "threshold_breach",
  "metric":      "net_revenue",
  "value":       1842300,
  "threshold":   { "op": "<", "rhs": 2000000 },
  "trace_id":    "trc_e1f9…",
  "persona":     "regional_analyst",
  "scope":       { "region": "EMEA" },
  "context":     { "top_drivers": [...] }
}

Built-in delivery channels: email, Slack, Microsoft Teams, PagerDuty, Opsgenie, generic webhook. The webhook signs payloads with HMAC so your receiver can verify authenticity. Learn how Monitors fit into the Semantic Control Plane security architecture.

Schedule

Cron expressions, simple intervals (every 15 minutes), or event-driven triggers from upstream Monitors or external webhooks.

• Schedules are timezone-aware. Use the workspace timezone or override per Monitor.
• Schedule skew (the gap between intended and actual run) is captured in the audit log and alertable.
• Failed runs are retried with backoff; persistent failures eventually page the Monitor owner.

Anomaly detection

For metrics with enough history, Colrows can derive thresholds automatically - seasonality-aware baselines from the same metric you'd otherwise threshold by hand. Combine with Stopclock sustained mode for high-quality alerts that don't page on weekend dips.