digna Data Reconciliation: Do Your Targets Still Match Their Sources?

Every data platform is full of tables built from other tables: daily aggregates summed from transaction lines, data marts assembled from a dozen source systems. Each of those loads can finish without an error and still produce a target that no longer agrees with the data it came from, because a filter changed, a late file was skipped or a join quietly dropped rows. Whether the numbers still add up is the question that matters, and it is usually answered by hand, if at all.

digna is a Data Quality & Observability platform, and digna Data Reconciliation, new in Release 2026.11, asks that question for you every day. It compares a target data source with the sources it was built from and alerts you when they no longer match. This short video shows both kinds of rules in digna's demo projects, with fictional data and the real product.

Configured on the target

Reconciliation starts from the table whose correctness you care about. You open the target data source, go to its Data Reconciliation tab and define the rules there, using as sources any data sources in the project that have reconciliation enabled.

There are two kinds of rules, because some relationships are simple and exact, while others involve many sources and transformations that nobody wants to rewrite as a check.

Exact rules for the reconciliations you already know

An exact rule works like a data quality rule for one source and one target. You pair checks by statistic and attribute, for example a row count on the target against a row count on the source, or a sum of amounts against the same sum one level down, and set an absolute or relative tolerance. The timeline then shows each day's difference against that tolerance, so you can see whether a gap is a one-off or has been growing for weeks.

In the retail demo, the daily sales aggregate pos_sales_daily is compared with the sales lines it is built from, pos_sales_lines, pairing gross revenue with the revenue summed from the lines at a tolerance of ± 0.01 %. On 26 August the aggregate showed €51,747 while its own lines added up to €66,205, which left the target 21.8 % short and far outside the tolerance.

Finance, risk and controlling teams already run checks like this as standard reconciliation procedures; as exact rules, they are written down once and run on every inspection.

AI rules for targets built from many sources

Exact rules stop being practical once a target is assembled from many tables through joins, filters and aggregations, because checking it that way means maintaining a second copy of the pipeline's logic.

AI rules take a different route. You combine checks from several sources with one target, with no joins, filters or aggregations to define and no SQL to write. A model learns from history how the target follows its sources and alerts you when the target falls outside the band of the rule's past errors. Each AI rule has three settings, Sensitivity, Memory and Model Complexity, so you can tune it for that particular target.

In the demo, an order mart is built from ten source tables. On 6 October it held 1,742 orders while its sources predicted about 2,771, roughly a thousand fewer, and the rule caught it without anyone having written down how the mart is assembled from those ten tables.

On the Dashboard and in your notifications

Reconciliation checks appear on the Dashboard with a timeline for each rule. They count towards the overall inspection status, next to your other checks, and they trigger notifications, so the right people hear about a mismatch on the day it happens.

Who benefits

Data engineers get a direct way to validate ETL results: not only that a load ran, but that its output still matches its inputs. Finance, risk and controlling teams get their standard reconciliations as exact rules with a visible history. And because digna runs on-premises or in your private cloud, your data never leaves your infrastructure while it is being checked. The Data Reconciliation documentation explains how to set up both kinds of rules.

Take the product tour to see digna Data Reconciliation and the rest of the platform at work.

Frequently asked questions

What does digna Data Reconciliation compare?

It compares a target data source with the sources it was built from and alerts you when they no longer match. Rules are configured on the target's Data Reconciliation tab, and any project data source with reconciliation enabled can be a source.

What is the difference between exact rules and AI rules?

An exact rule connects one source and one target, pairs checks by statistic and attribute, such as row counts or amount sums, and applies an absolute or relative tolerance. An AI rule combines checks from several sources with one target, learns from history how the target follows them and alerts when the target falls outside the band of the rule's past errors.

Do I need to write SQL or define joins for AI rules?

No. An AI rule needs no joins, filters or aggregations, and there is no SQL to write. You can tune each rule with its Sensitivity, Memory and Model Complexity settings.

How are reconciliation results reported?

Each rule has its own timeline on the Dashboard. Reconciliation checks count towards the overall inspection status and trigger notifications, like digna's other checks.