Platform · Observability
Every asset.
Every stage.
Live.
The pipeline canvas shows every table in your client's pipeline — ingested, being cleaned, deployed, or stale — as a connected graph. Health metadata sits on every node. You see the shape of your data and how it's doing in one view.
How to read the canvas
Three node types.
One lineage map.
Raw
Source tables exactly as ingested — every record tagged with source identifier and ingestion timestamp. No transformations applied.
Cleaned
Intermediate models — normalised, prepped, quality-checked. Built by agents, proposed for approval, deployed as dbt models.
Mart / Golden
Final mart tables and MDM golden records — fully approved, deployed, and ready to activate or hand off.
Ghost nodes (dashed border, no colour) appear for pipeline stages not yet reached — they show what comes next without implying it exists.
Data observability
The canvas doesn't just show lineage.
It shows health.
Every node tells you whether to proceed. A stale raw table means a downstream mart is running on yesterday's data. A null-rate spike in a key column means the cleaning rule needs revisiting. You see it on the canvas before it reaches the ERP.
Freshness
Time since last run
Detects silent failures and stale data before downstream consumers notice.
Volume
Row count vs. prior run
Catches upstream truncation or duplication that row counts alone can mask.
Null-rate
% null per key column
Surfaces data quality degradation before it propagates into your mart models.
From signal to action
See it.
Trace it.
Act on it.
The canvas is the navigation layer for governance. When something looks wrong, you don't hunt through logs — you click the node and go directly to the decision that needs making.
-
Canvas flags a stale or drifting node
Freshness drops below threshold or null-rate climbs — the node is visually marked.
-
Click → approval queue for that table
One click surfaces the pending proposal for that asset — no context switching.
-
Review, approve or reject
The pipeline resumes on approval. A rejection holds the stage and logs the reason.
Previous
← Target Mapping + ActivateSee all stages
Platform overview →