Catch late and broken streaming data first.

What it helps your organisation achieve

LionPact Watch is built for data platform, analytics and ML teams that depend on real-time data.

• Find problems in minutes

  Know about a late or broken stream before a business user or customer reports it.

• Trust real-time numbers

  Show how fresh each dataset is, so people know when a dashboard can be relied on.

• Protect downstream teams

  Schema contracts catch breaking changes at the source, before every consumer fails.

• Know the impact straight away

  Lineage shows which tables, reports and models a failure affects, so the right people are told.

• Spend less time debugging

  Each alert names the stage, the check that failed and when it started, so on-call engineers start in the right place.

• Keep data inside your walls

  It runs in your environment with read-only access. Your data is never sent anywhere else.

How it works

Watch sits beside your pipelines. It reads signals your systems already produce and never sits in the data path, so it cannot slow down or break a job.

How to use it

Watch is in development, so the steps and configuration below show the planned setup. Details may change before the first release.

1. Run it next to your pipelines

   Deploy Watch as a container in your own cloud account or Kubernetes cluster. Nothing is installed inside your jobs, and no pipeline code changes.

2. Connect your sources, read-only

   List the topics, jobs and tables you want watched. Watch reads offsets, metrics and a small sample of messages. It never writes to your systems.

   # watch.yaml (planned format)
   sources:
     - name: orders
       type: kafka
       bootstrap_servers: kafka-1:9092
       topic: orders
     - name: orders_clean
       type: iceberg
       table: analytics.orders_clean

3. Say what healthy looks like

   For each stream, set how fresh it must be and which fields and types consumers rely on. Volume and value ranges are learned automatically from normal traffic.

   checks:
     - stream: orders
       freshness:
         warn_after: 60s
         fail_after: 5m
       contract:
         required: [order_id, customer_id, amount, created_at]
         types: { amount: decimal, created_at: timestamp }
       volume:
         baseline: learned   # learns the normal hourly pattern

4. Send alerts to the right people

   Route warnings and failures to Slack channels or email by stream or team. Each alert says which stage, which check and since when.

   alerts:
     - channel: slack
       webhook_env: SLACK_WEBHOOK_URL
       streams: [orders, payments]
       on: [fail]

5. Review and tune

   Use the dashboard to see the health of every stream. After the first few days, tighten thresholds where alerts are too loose and relax them where they are noisy.

Where organisations can use it

Any team that makes decisions on data that is seconds or minutes old.

Questions