# Impact analysis works only if lineage is fresh, reaches the columns you will change, and can be queried before you merge

GUIDE, read 10 Nov 2026

**The short answer.** Impact analysis lists what breaks downstream of a change. Foundational runs it on pull requests and can block merges; Select Star offers a GitHub Actions app; SYNQ exposes an impact API. Freshness varies: Euno syncs daily by default, SYNQ several times a day, Monte Carlo's dbt context can lag 24 hours.

## Impact analysis asks what breaks downstream

Impact analysis is the question you ask before a change: if I rename this column, change this filter or drop this table, what stops working? Collibra describes its lineage as tracing root causes and downstream impact at table, column and report level. The same graph answers both directions. Upstream shows where a wrong value came from. Downstream shows who will see a change.

The answer is only as good as three things. The graph must contain the edge, which is a coverage question. It must contain the edge now, which is a freshness question. And it must be reachable at the moment you decide, which is an interface question. A perfect graph you open a day after the merge is a post-mortem tool.

## A routine that fits in an hour

The steps below work with any tool in this comparison. They are ordered so that the cheap checks come first.

- Name the exact column, not just the table. Table-level impact lists every consumer of the table, including those that never read your column.
- List the downstream columns two or three hops out, then the dashboards and reports those columns feed.
- For each consumer, find the owner. A lineage list without owners is a list of strangers.
- Check the edges that parsers miss: columns used only in joins or filters. See the blind spots table below.
- Check when the lineage was last refreshed. If the last sync predates yesterday's merge, rerun it.
- Post the list in the pull request, so the reviewer sees the same blast radius you did.

## Checks that run before the merge

Most lineage tools are reached through a web page. A few can be reached from the pull request. Foundational says it is triggered by pull-request webhooks and keeps a versioned lineage graph, and that its CI runs impact analysis on pull requests and can block merges. Select Star offers a GitHub Actions app that it says prevents data breakage before it happens. SYNQ exposes lineage and impact through a public API, with LineageService and ImpactService, which SYNQ says suits CI checks or AI assistants.

dbt adds a caution. dbt says CI jobs do not update Catalog. A change under review is therefore not in dbt Catalog's lineage until it runs in production or staging, so Catalog shows the state before your change, not after. That is correct for impact analysis, since you want the current graph and your diff. But you must compare the diff yourself.

## How current the graph is

Lag changes the answer. A column added on Monday and consumed on Tuesday is invisible to a graph that refreshes weekly. The table lists what each vendor documents.

| Tool | Refresh behavior documented | Source |
| --- | --- | --- |
| Euno | Sources sync daily by default, configurable, with manual syncs possible | Euno FAQ |
| SYNQ | Extracts lineage from the warehouse and BI tool several times a day | SYNQ docs |
| Foundational | Pull-request webhooks plus scheduled scans of BI tools and warehouses | Foundational docs |
| Monte Carlo | dbt context in lineage can take up to 24 hours because of batch processing | Monte Carlo docs |
| Coalesce Catalog | Graph uses the last 30 days of SQL; first Catalog-managed BI syncs can take up to 48 hours | Coalesce docs |
| Snowflake | Built from its own query history; lineage began in November 2024 | Snowflake docs |
| Databricks | Captured automatically; system tables hold a rolling 395-day window | Databricks docs |
| Select Star | Refresh frequency: none found on 10 Nov 2026 | Select Star product page |

## How far back the graph remembers

Impact analysis also looks backward. If a report runs quarterly, a graph that remembers 30 days will not show its dependencies in the other eleven weeks. Coalesce Catalog says its graph uses the last 30 days of SQL. Snowflake says lineage is retained for one year, and that data-movement lineage captured before November 2024 is not available. Databricks says its lineage system tables keep a rolling 395-day window.

A versioned graph helps here. Foundational says it keeps a versioned lineage graph, which lets you ask what the graph said at an earlier date. Whatever the tool, record the date of the lineage you relied on in the pull request.

## Blind spots to check by hand

Every tool documents some case it cannot see. Read these before you trust an empty impact list.

| Tool | Documented blind spot | Why it matters for a change |
| --- | --- | --- |
| Atlan | Column lineage excludes JOIN, FILTER and GROUP BY transformations | Changing a join key may show no impact |
| Snowflake | Lineage excludes tables used only for joins or filters; unavailable for shared databases | A lookup table can look unused |
| dbt | Column lineage reflects select statements only, not joins and filters | Filter columns are not traced |
| DataHub | Manually edited lineage may conflict with ingested lineage and be overwritten | Hand-added edges can vanish on the next sync |
| Euno | Daily default sync | Same-day changes need a manual sync |

## Reading an empty result

An empty impact list has three possible meanings. The column has no consumers. The column has consumers the tool cannot see. Or the lineage has not caught up. Only the first is good news. To tell them apart, check one known consumer: pick a dashboard you know reads the column and see whether the tool lists it. If it does not, the tool has a gap for your stack, and every empty result from it is suspect.

The same logic applies to a long list. A list of fifty tables may include tables that only read the table, not your column. If the tool is table-level, narrow the list by hand. If it is column-level, check whether it draws filter and join edges. Atlan, dbt and Snowflake each document that they leave some of those out, so a short list from them is a lower bound, not a total.

Teams that make impact checks routine tend to keep a short list of canary consumers: three or four dashboards or tables that must always appear. If a canary goes missing from the lineage, the sync is broken, and you find out before a real change depends on it.

## Who owns the consumers

A lineage list is only useful when it turns into messages. For each consumer in the list, name the owner, the change and the date. Give a rollback plan if the consumer is a finance report or anything an executive opens on Monday. Where an owner cannot be found from the tool, that is itself a finding: either the metadata is missing, or the dashboard has been abandoned, and both are worth fixing.

Keep a record of the lists you produce. After a few months you can see which tools of yours miss edges, which consumers always appear, and which changes tend to cause incidents. That history improves the next impact check more than any new feature does.

Impact analysis ends with a message to people. Where the tool shows no usage, ask the dashboard owner directly. A short list of heavily used consumers is easier to act on than a long list of every possible one.

Do this even when the tool reports a long list. A long list can still miss the one consumer that matters, and the check costs a few minutes. Before a risky change, compare two outputs: the tool's impact list and a grep of your repository for the column name. Where they disagree, you have found a blind spot, and you should record it. [The matrix](/matrix/) shows how far each tool's column lineage reaches, and the [method page](/method/) explains how freshness is scored.

## Sources

- Collibra: www.collibra.com/products/data-lineage: https://www.collibra.com/products/data-lineage (read 10 Nov 2026)
- Foundational: docs.foundational.io/en/articles/12802881-solution-architecture.md: https://docs.foundational.io/en/articles/12802881-solution-architecture.md (read 10 Nov 2026)
- Select Star: www.selectstar.com/product/data-lineage: https://www.selectstar.com/product/data-lineage (read 10 Nov 2026)
- SYNQ: docs.synq.io/lineage/lineage.md: https://docs.synq.io/lineage/lineage.md (read 10 Nov 2026)
- dbt: docs.getdbt.com/docs/explore/explore-projects: https://docs.getdbt.com/docs/explore/explore-projects (read 10 Nov 2026)
- Euno: docs.euno.ai/support-resources/faq.md: https://docs.euno.ai/support-resources/faq.md (read 10 Nov 2026)
- Monte Carlo: docs.getmontecarlo.com/docs/dbt-integration: https://docs.getmontecarlo.com/docs/dbt-integration (read 10 Nov 2026)
- Coalesce Catalog: docs.coalesce.io/docs/catalog/navigate/lineage.md: https://docs.coalesce.io/docs/catalog/navigate/lineage.md (read 10 Nov 2026)
- Snowflake: docs.snowflake.com/en/user-guide/ui-snowsight-lineage.md: https://docs.snowflake.com/en/user-guide/ui-snowsight-lineage.md (read 10 Nov 2026)
- Databricks: docs.databricks.com/aws/en/admin/system-tables/lineage: https://docs.databricks.com/aws/en/admin/system-tables/lineage (read 10 Nov 2026)
- DataHub: docs.datahub.com/docs/features/feature-guides/lineage: https://docs.datahub.com/docs/features/feature-guides/lineage (read 10 Nov 2026)
- Atlan: docs.atlan.com/product/capabilities/lineage/best-practices/capture-and-troublesh: https://docs.atlan.com/product/capabilities/lineage/best-practices/capture-and-troubleshoot-lineage (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://datalineagecompare.com/guides/impact-analysis-before-a-change/
