Signals: autonomous schema monitoring

Autonomous agents cannot operate on stale logic. Signals are not just logs. They are the real-time feedback loop that keeps your semantic layer deterministic even as your underlying warehouse schemas evolve. Scheduled, threshold-aware queries compile through the same semantic execution layer as everything else in Colrows. Alerts you can trust. Grounded in governed metrics. Not stitched together with cron jobs and webhook handlers.

Standard data logs vs. Colrows Signals

The architecture gap

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 Signal advantage

Colrows Signals 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. Signals compile through the same governed execution pipeline as every other Colrows query: intent resolution, constraint validation, join path proof, compile-time governance. When a Signal 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 Signal is

A Signal 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 Signals 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 Signal has a single owner. Owners receive operational notifications (compilation failure, schedule skew).
• Persona - Signals 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 Signals. It prevents a noisy Signal from paging on every small fluctuation.

Result - how Signals 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 Signals fit into the Semantic Control Plane security architecture.

Schedule

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

• Schedules are timezone-aware. Use the workspace timezone or override per Signal.
• 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 Signal 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.