Data Reconciliation Recipe FAQ
Common questions about the rule type behind the recipe, left and right, tolerances, what it writes, AgentQ, and resuming.
The Recipe and the Check
Is the Data Reconciliation recipe a new rule type?
No. The recipe is a guided flow around the existing Data Diff rule type. It creates an ordinary Data Diff check between the two assets, activates it, scans with it, and helps you act on its anomaly. The check behaves exactly like one created by hand.
Does the check stay active after the recipe finishes?
Yes. The check is created as a Draft when you leave Review, becomes Active when Validate finishes, and stays Active, so scheduled scans keep reporting drift between the two assets. Reconcile in Recipe on their anomalies brings you back to the Reconcile step.
Can I load a check I created by hand into the recipe?
An active Data Diff check can be taken to the Scan step with Scan in Recipe. There is no Open in Recipe option for a Data Diff check that is not active yet; activate it on the datastore page first.
Is there a dry run?
No. Validate saves the check and activates it; the platform accepting the configuration is the validation. The first real results come from the full scan.
Left and Right
Which asset goes on which side?
Put the asset you would correct on the Left, the asset to inspect: mismatches are reported on it, and the keep and remove sets describe what it should end up with. Put the system of record on the Right, the reference treated as correct. The reference is never asked to change.
What do Added, Removed, and Changed mean?
Always from the left asset's point of view: Added rows exist only on the left, Removed rows are missing from the left, and Changed rows exist on both sides with different compared values.
Can the two assets come from the same datastore?
Yes. Either side can come from any datastore you can access, including the same one.
Why is a field I expected missing from the compare choices?
Fields are paired by name and type. A field that exists on one side only, or shares a name but not a type, is excluded. The Schema overlap panel on Select Assets lists which case applies. Left-only fields can still ride along as Passthrough.
Tolerances
Can I set a tolerance for one field?
No. Tolerances are set per field type: one numeric comparator, one date and time comparator, and one text comparator for the whole check. Each card on the Tolerances step lists the compare fields it covers.
What does Passthrough do?
Passthrough fields are never compared. They ride along so the keep set and the mismatch explorer carry their context.
Outputs
What is written, and where?
A keep set and, when needed, a remove set, to the enrichment destination linked to the left datastore. The keep set holds the values the left asset should end up with and which asset each one came from, as <prefix>_mat_dr_<table>. The remove set lists the rows the left asset holds that the reference does not, as the same name plus _removals. Inside the enrichment datastore they carry the display names <TABLE> / Data Reconciliation (<datastore>) and <TABLE> / Data Reconciliation / Removals (<datastore>).
Why was nothing materialized?
Because every mismatch was resolved in favour of the left asset. The outputs are corrections only, so when the left asset is already right there is nothing to write, and the recipe completes with the anomaly archived as Discarded.
What happens to the anomaly?
The full scan raises a Data Diff anomaly holding every mismatched row with both sides' values. When the keep set is written, the recipe acknowledges it automatically.
Does the recipe change my assets?
No. Neither asset is changed. The keep and remove sets are new outputs in the enrichment destination for you to act on.
AgentQ
Can I use the recipe without AgentQ?
Yes. Every step works manually. When no AI provider is configured, the AI banners are simply absent.
Why do the field suggestions work but the resolution suggestions do not?
The row identifier and compare column suggestions and the scan analysis send field names and statistics, which the Metadata Shared data sharing level allows. Resolution recommendations send both sides' row values, so they need Source Data Shared. Below that level the step reports that no suggestions arrived and you pick the sides yourself. See AgentQ Access Controls.
Why does the Reconcile step ask before using AI?
Because that step is the only one that sends your data values to the model. Nothing is sent until you click Get AI suggestions.
Resuming and Rerunning
Can I rerun the recipe on new data?
Yes. The check stays active, so open it on the datastore page and click Scan in Recipe to start at the Scan step. If a scheduled scan already ran, click Reconcile in Recipe on the anomaly it raised.
Does Start Over delete anything?
No. It clears your progress and returns to the first step. The check stays active, the scans and anomalies remain available, and the outputs already written to the enrichment destination stay as they are.
Can I resume on another computer?
Not from the Recipes page. Progress is kept in your browser session, so another browser or computer shows Get Started instead of Continue. Use Scan in Recipe on the check or Reconcile in Recipe on the anomaly to pick up at the right step.