Column-level lineage
Alkera’s lineage engine merges dependencies from each connection. Each plugin traces columns from what its tool exposes, such as Fivetran’s column mappings, Databricks’ lineage tables, or the SQL behind a dbt model. Every edge shows where it came from.
Before a change, the agent traces it through the column graph and labels it breaking, potentially breaking, or non-breaking. It lists the affected downstream assets and the tests worth rerunning.
When a change would break something downstream, the agent stops and asks for approval in the chat, with the affected assets listed.
I checked the downstream dependencies. fct_orders.customer_id reads staging.customer_id, so removing it would break fct_orders.
When ingestion runs in your own code, the agent writes a plugin that reports its inputs and outputs to lineage, monitoring, and the knowledge base.
ingest.pyimport pandas as pdfrom snowflake.connector.pandas_tools \ import write_pandas def ingest_daily_bars(source, connection): rows = pd.read_json(source) rows = rows.dropna( subset=["ticker", "session_date", "close"] ) rows["ticker"] = ( rows.ticker.str.upper() ) rows["adj_close"] = ( rows.close * rows.factor ) bars = rows[["ticker", "session_date", "adj_close", "volume"]] write_pandas( connection, bars, "DAILY_BARS", database="RAW", schema="MARKET", ) return { "table": "RAW.MARKET.DAILY_BARS", "rows": len(bars), "columns": list(bars.columns), } if __name__ == "__main__": ingest_daily_bars(source, connection)I’ll trace the processing code and connect its output to Snowflake.
I found the source fields and transformations. I’ll create a Python plugin to trace and monitor this job.