Map and match source metadata

This page is about the catalog, which the free edition does not include. Connections, imports, and the physical tables and columns they create are part of the catalog surface, so a free organization has nothing here to work against. Everything below is correct for an organization that has the catalog. See Pricing for what each edition includes.

Mapping is what makes governance flow. The Map & Match workbench, opened from Business > Map & Match in the left navigation (/business/map-match), proposes which business entity each source table represents, and which attribute each column maps to, so the classifications and definitions on your model reach the catalog. Matching runs in both directions: start from your tables to find their entities, or stand on a business entity and ask what in the catalog should map to it. You review every proposal before it sticks.

Starting a match

From tables

Pick a scope at the top (a project or workspace) to choose which source tables are in play. The grid lists them; tick the tables you want to match (the header checkbox selects everything in view), then run Suggest matches. Until at least one row is ticked the button stays disabled and says “Select the tables you want to match.”, so a run always covers exactly the tables you chose, never a whole scope by accident. The matching AI skill reads your tables, columns, entities, attributes, and any glossary documents you’ve uploaded, then produces a candidate match per table with a confidence score and a short rationale.

The skill behind this workbench is called Match source tables: it maps whole source tables onto your business entities. Matching individual columns runs as a separate step (the column matching described below); you do not launch it as its own skill from the Ask AI list.

From a single table’s catalog detail, Suggest match opens a matching panel on the right side of the table you are viewing. The panel suggests a match for that table automatically and walks you through the same guided review, including the column matching step, without leaving the table. Choose Open full workbench in the panel when you want the full page instead.

From a business entity

Standing on a business entity, ask the reverse question: what in the catalog should map to this? Choose Find source tables in the entity’s detail header, or right-click the entity in the Business entities tree or on a model canvas and choose Map and match. A panel opens, finds up to 12 unmapped tables that resemble the entity by name and by meaning, and runs one match over exactly those candidates. The cost stays fixed no matter how large your catalog grows, because the run is bounded to the candidates it found.

An entity-seeded run never proposes a new entity: you already named the target, so a table with nothing to offer comes back as a skip with the reason. When no unmapped table resembles the entity at all, the panel says so plainly and no run starts.

Where you can start a match

You do not have to leave a canvas to map the table you are looking at. Right-click a table and choose Map and match. The table’s detail opens in the docked pane beside the canvas with the matching panel already on it, and the guided review runs there exactly as it does from the catalog.

The action is on every canvas that draws source tables: a project or workspace canvas, the catalog’s relationship canvas, and the Lineage tab of a table’s detail. On a lineage graph it works on the neighbors too, so you can map an upstream table without navigating to it first. The catalog tree offers the same action: right-click a table or view row and choose Map and match to open its detail with the matching panel already running.

The entity side mirrors it. Right-click an entity row in the Business entities tree, or an entity node on a model canvas, and choose Map and match; or use Find source tables in More actions of the entity’s detail panel, on the Glossary or a model canvas. All of them open the same entity panel described above.

Pick one table: with several nodes selected the action is not offered, because the panel handles a single table and batch matching belongs in the workbench with its scope picker. When your organization does not have AI matching enabled yet, or you are on the main branch, every one of these actions stays in its menu and tells you why it is unavailable instead of disappearing. If the entity’s detail panel holds an unsaved edit, the Unsaved changes question comes first: cancel it and nothing moves.

What the matcher considers

Which entities the matcher compares your tables against follows the scope you launch from:

A workspace-scoped run compares against the entities in that workspace’s own business models.
A project-scoped run compares against the entities in that project’s own business models.
An entity-seeded run (started from a business entity) compares only against that entity and its group members.

On a workspace-scoped run, an Include project models toggle (off by default) widens the pool to the models of every project inside the workspace too. Next to the scope picker, a summary line states the pool in plain terms (for example, “Matching against 6 models in this workspace” or “Matching against all models in the organization”).

If the chosen scope has no candidate entities, Suggest disables. The workbench offers a one-click way to widen the pool: Include project models (on a workspace scope that has not already turned it on) and Match against all models, which drops the pool to every model in your organization. From a model’s own Map & Match tab there is no wider pool to fall back on: if that model has no member entities yet, Suggest stays disabled and asks you to add entities to the model first.

Widen the pool when nothing matches

The matching panel that opens from a single table, whether from Suggest match or from a canvas, follows the page you launched it from: inside a workspace it compares against that workspace’s models, inside a project against that project’s models. When that pool turns up nothing, the panel offers the next scope out rather than leaving you to work out what went wrong. Two findings, because the remedy differs:

No business models are in scope here. There was nothing to match against, so no run starts. Widen the pool first.
No match found among this workspace’s models. The run finished and found no existing entity in the pool it searched.

Either finding comes with the rungs you have left: Include project models, on a workspace that has not already turned it on, and Search the whole organization. Choosing one re-runs matching immediately at the wider scope. Once you reach the whole organization the panel states the result plainly, “No match found anywhere in the organization”, with nothing further to try.

Widening never discards what the matcher already proposed. A new-entity suggestion is often the right answer, so it stays on screen beside the widening options rather than being replaced by them.

Match again re-runs matching at the scope you are on, from a project page as well as a workspace one.

Watching a run

While a run is in progress, a step timeline shows exactly where it is:

Preparing candidates. The matcher gathers the entities in scope and the evidence it will compare your tables against.
Table batches. Tables are matched in batches, one timeline step per batch, each labeled with the tables it covers (for example, “Tables 1-20 of 45”).
Recording results. The suggestions from every batch are saved and the run’s summary comes together.

Each step shows its own status: a spinner while it runs, a checkmark once it succeeds, an X if it fails. A failed batch does not stop the run: the remaining batches still run, and the completion summary at the end reports how many batches failed. Only when every single batch fails does the whole run fail.

Once a run finishes, a one-line summary reports the outcome: how many tables were scanned, how many table suggestions and column suggestions came out of it, and, if any batches failed, how many (for example, “18 tables scanned, 15 table suggestions, 42 column suggestions, 2 of 6 batches failed; some tables were not scanned”). The suggestions that did come back are ready to review; the tables that were not scanned are worth another Suggest run once you are ready.

Reading the suggestions

Each table row shows its suggested entity, a confidence percentage, and the reasoning behind it. Every suggestion in this workbench is grounded in evidence rather than naming similarity alone, so you can see why a match was proposed; what counts as evidence differs by tier. Table suggestions weigh name and business-term similarity, the table’s own column names and declared data types, its row count, and whether sibling tables in the same schema are already mapped, together with the Business Modeling and Glossary & Naming Standards foundations the skill cites. Column suggestions start from the same name and data-type signals, then add the evidence that lives at column level: the profiling statistics stored on the column (null ratio, distinct count, minimum and maximum values, value lengths, and most frequent values), the classifications the candidate attribute already carries, and the columns already matched on the same table; that skill cites Data Quality & Profiling alongside the same two foundations. The rationale names the signals that supported each match, so you can judge a suggestion on its evidence instead of taking the percentage on faith; when only names were available, the rationale says so, and the confidence reflects the thinner evidence.

Confidence reads the same way on every review screen in this workbench: a whole percentage (for example, 94%), with a Low confidence warning icon beside any score below 50%. The 50% mark is fixed: it is separate from your organization’s Auto-accept confidence floor (see Acceptance settings) and stays at 50% whatever the floor is set to.

A suggestion is one of two kinds:

An existing entity: the matcher recognized the table as your Customer, Order, or Shipment entity.
A new entity: nothing fit, so it proposes a skeleton. The row shows a non-editable new: label beside an editable name field. Rename the skeleton inline before you accept it.

A run started from a business entity only ever proposes the first kind: the target is already named, so a table with nothing to offer comes back as a skip instead of a new entity.

Four filter chips control the grid: Unmapped (the default, so you only see work left to do), Suggested, Mapped, and All. A mapped table whose mapping was accepted where the AI suggested it shows a small sparkle icon beside its entity name until someone points the table at a different entity (see How accepted AI values stay marked). Remap re-runs matching for a table whose suggestion you want to redo.

Matching across a table family

Source data usually arrives in layers: a raw landing copy, a staging copy, and a persistent-staging copy, all derived from one source table. Map & Match recognizes that chain and matches the source table on behalf of its whole family. In the grid, a landing, staging, or persistent-staging copy whose chain leads back to a source table has no checkbox of its own: its Entity cell reads Will follow and the source table’s name until the source is mapped, and Mapped via family once it is.

When you accept an entity for the source table, every other layer in its family is mapped to the same entity in the same step, with nothing further to confirm. A layer that already maps to a different entity keeps its own mapping. A source table and its landing, staging, and persistent-staging copies end up pointing at one entity without re-deciding the match for each.

Columns work the same way. Choose Columns on a copy that follows its source and the panel works on the source table’s columns instead, with a notice naming the source table and saying that column decisions apply to it and cascade to its whole family.

When you match a derived copy on its own, for example with Suggest match from its catalog detail, the matcher reuses an entity already mapped to any member of the family; when nothing in the family is mapped yet, the name signal uses the clean source name instead of the decorated layer name, so a staging table named stg_customer is recognized as your Customer entity.

Accepting matches

Select the rows you trust and accept them in bulk. Accepting an existing-entity match records the mapping. Accepting a new-entity match creates the skeleton entity and maps the table to it in one step. Either way, the mapping keeps the AI badge, with the suggestion’s confidence and rationale, until someone points the table elsewhere; How accepted AI values stay marked covers the details.

Skeleton entities are governance without modeling. You get a real entity you can classify and define right away, without stopping to draw it on the board. Accepted skeletons wear a Derived chip on the modeling board, indicating they are derived from source mapping and not yet adopted into the model. When you’re ready to make one a first-class part of your model, open the entity’s panel and click Adopt into model on its Details tab.

Accept all above the floor

Above the grid, Accept all at or above N% accepts every pending suggestion in the run that meets your organization’s configured confidence floor, in one click and without selecting rows. The percentage in the label is that floor. The control is governed: it is unavailable under the Review each acceptance setting, where every suggestion must pass through Guided Review, and while an administrator has locked the Conventions section of Settings (its Lock your categories button), only administrators can use it. Anything below the floor stays pending for you to judge.

Guided Review

Guided Review walks you through pending suggestions one at a time, so you can give each table your full attention without hunting through the grid. For each suggestion you see a decision card with the proposed entity, the confidence score, and the AI rationale. A progress indicator shows how many decisions remain.

Keyboard shortcuts in Guided Review: press A to accept, R to reject, E to refine a new-entity suggestion, and the right arrow key to skip to the next suggestion without deciding, the same as Skip.

How you reach Guided Review depends on your organization’s acceptance setting (see Acceptance settings below):

Under Review each, the workbench opens directly into Guided Review and hides the bulk grid. Every suggestion goes through the decision card.
Under Allow bulk or Auto-accept, the grid stays visible. A Guided review toggle switches the view into Guided Review when you want to work through remaining suggestions one at a time.

Add to business model

When you reach a suggestion where no existing entity fits, Guided Review gives you two options rather than forcing a reject:

Map to an existing entity: search your model and pick the entity manually. The table is mapped immediately.
Create a new entity: name and create a skeleton entity on the spot. After you confirm, the workbench offers to run column matching for that table so you can continue enriching it without switching context.

Create an entity from a source table

A new-entity suggestion creates a bare skeleton you name and enrich later. When you would rather start from the source table’s shape, this option on the decision card builds the entity from the table’s own columns instead. Choosing Create entity from source opens the column picker. Confirm or edit the Business name the entity will get (it is required), then choose which source columns become attributes: business columns start ticked, dv_ system columns start unticked. If the source table has a designated business key column, it is pre-selected, locked, and becomes the new entity’s business key automatically. Without a designation, the picker embeds the define-business-key step: promote one of the selected columns to business key, or derive a pattern-named key attribute from the business name. Creation stays blocked until the key is resolved, so every entity created from a table leaves the flow with its business key in place. If the table has siblings in its family, they are mapped to the new entity in the same step.

Model-scoped Map & Match

Opening a model from Business > Models and switching to its own Map & Match tab gives you the same workbench, scoped to that model. The source picker starts narrow: it offers only the projects and workspaces that carry this model, with Any scope… at the end restoring the full list when you want it. A model that is not in any project or workspace yet falls back to the full picker with a note saying so. What the tab proposes and shows is narrower too: existing-entity suggestions are drawn only from the model’s direct member entities, and a table already mapped to an entity is badged In this model or Outside this model, depending on whether the mapped entity belongs to this model.

The grid is honest about what is worth a run. Tables related to this model sort first and start ticked: a table counts as related when its schema already contains a table mapped into the model, or when its name resembles a member entity. Unrelated tables stay in the grid below a divider, unticked. Nothing is hidden; adjust the selection freely before you choose Suggest matches.

Create entity from source works exactly as described above, with one difference: the new entity becomes a member of this model in the same step, so there’s nothing further to do to add it.

Column matching

Drill into a table to match its columns: choose Columns on the table’s row to open its column panel. The panel has one launcher, Map and match, with an Effort picker beside it. Run it and DeltaVault first matches every column that has no decision yet (system columns are left out) against your existing attributes, then drafts a business name and a definition for each column that matched nothing. Names and definitions generate in your preferred language. Once every column has a decision, the launcher stays disabled and tells you why: “Every column has been decided. Reject a decision to offer it again.”

The Review enriched columns dialog opens as soon as the run starts, and drafts arrive in it batch by batch. Each row shows the column with its business name, technical name, and definition. The Confidence column next to the column name shows how sure the AI is of the row’s proposal (the attribute it matched, or the attribute it drafted) as a whole percentage; hover the percentage to read the AI’s rationale. A column that matched an existing attribute shows that attribute’s name with an Existing attribute tag and keeps the attribute’s own definition; when the attribute has no definition yet, the row offers a drafted one for you to review. Give each row one of three outcomes with the icon buttons in its Actions column (hover one to read its name):

Accept (the check mark) creates a new attribute of the table’s entity from the row, or maps the column to the existing attribute it matched.
Save (the disk icon) keeps the business name and definition on the column only, without an attribute.
Reject (the cross) discards the draft and saves nothing.

Before you decide, you can edit any drafted field, have the AI edit or regenerate it, ask for a fresh suggestion for the row with Suggest (the sparkle icon), or choose an attribute yourself with Map (the link icon). Every row gets its own decision (there is no bulk accept), and nothing you decide is written until you choose Apply N decisions. The summary beside that button counts new attributes, mappings to existing attributes, column-only saves, rejections, and rows still undecided. Accepting a column also copies the attribute’s quality expectations onto the column as a starting point. The one exception to waiting for Apply N decisions is the Auto-accept acceptance setting, which applies drafts at or above your organization’s confidence floor for you when the run finishes (see How the acceptance setting changes column review below).

Whatever you apply exactly as drafted keeps the AI badge with the model’s confidence; a value you edited first, or a mapping you chose yourself with Map, is yours. See How accepted AI values stay marked.

Close the dialog and the panel reports what the review decided. If decisions are staged but not yet applied, closing first asks “Leave without applying?”; Discard decisions drops them but keeps the drafts, so you can review again. A read-only grid lists every column with its status (Mapped, Described, Needs a decision, Not started, or System) and its result, and the All, Needs a decision, Done, and System filters narrow it. System columns stay out of All, and a caption counts them. While drafts are waiting, a callout above the grid says how many and offers Review N columns to reopen the dialog. A table with no columns at all reads “No column suggestions yet.”

When the table’s mapped entity belongs to an entity group (see Sub-entity grouping), a column suggestion is not limited to that one entity: it can point to any entity in the group, and accepting it creates or maps the attribute on that entity. The column cards under Review each name that entity with an owner chip, for example on Order Line; see How the acceptance setting changes column review.

How the acceptance setting changes column review

Map and match opens the Review enriched columns dialog under every acceptance setting. What changes is the panel behind it and what happens when the run finishes.

Under Review each, the panel shows a card walk in place of the read-only grid, one card for each suggestion still waiting, including any you left undecided in the dialog. Each card names the source column and the proposed attribute, with the match’s confidence percentage and the AI’s rationale, and offers the decision that fits it: Map to followed by the attribute’s name, Create attribute, or Mark skipped when the AI found no confident match, alongside Reject and Skip. On a new-attribute card, Refine lets you adjust the business name, data type, and definition before you create it. Each decision is applied as you make it, and a Captured so far panel lists what you have decided. Press A to accept, R to reject, E to refine, and the right arrow key to skip to the next card.

Under Allow bulk and Auto-accept, the panel shows the read-only grid described above. Under Auto-accept, one thing changes: when the run finishes, every draft still waiting that scores at or above your organization’s confidence floor is applied for you. A matched column is mapped to its attribute, and a drafted attribute is created and mapped. The drafts below the floor stay waiting for your decision in the dialog.

Column quick actions

The Columns tab carries two quick actions for a single column, reached from the row’s actions menu without opening the Map & Match column panel. Both need the table to already be mapped to a business entity: a pending suggestion does not count, so until the table’s mapping is accepted, both actions stay disabled with the reason “Map the table to an entity first.”

Suggest runs matching for just that one column:

Open the column’s row menu and choose Suggest.
The column’s Business attribute cell reads “Matching…” while the single-column run is live.
When it finishes, the column carries a Suggested: chip naming the attribute. Open it to read the rationale, then choose Accept or Reject. The same suggestion also waits in the Map & Match review. Once accepted, the column’s Business attribute cell shows the attribute with a small sparkle icon beside it until someone maps the column to a different attribute.

Promote creates attributes from columns, with drafted definitions you review first:

Open the column’s row menu and choose Promote. The Promote columns dialog opens with that column ticked; tick any others you want to promote with it. When the table’s mapped entity belongs to an entity group, pick the Owner entity that should receive the new attributes.
Choose Next. Each column gets an Attribute name from your organization’s naming convention. A column that already has a definition starts with it, marked From source metadata; for the rest, the AI drafts the Definition, and it drafts the AI context for every column. When the owner entity has no definition or AI context yet, the dialog drafts those too. Edit anything before you continue.
Choose Apply. The attributes are created on the owner entity. A definition you kept exactly as the AI drafted it is recorded as the AI’s: the attribute’s editor shows the AI badge beside Definition, and the attributes grid shows a small sparkle icon in its Definition column. The badge has no percentage, because this dialog’s drafts carry no confidence score. A definition you edited is yours. The owner entity’s drafted definition follows the same rule.

How accepted AI values stay marked

When you accept an AI proposal in Map & Match, the value you accepted keeps a record that the AI proposed it, together with the model’s confidence and rationale, until a person changes that value. Anyone reading the catalog later can tell an AI-proposed mapping or definition from one a person wrote.

These values stay marked:

Mappings. A table mapped to the entity the AI suggested, and a column mapped to the attribute the AI suggested, including a new entity or attribute the suggestion created. When accepting a table’s mapping also maps the other layers of its family, each of those layers carries the same mark: the tables grid’s Business entity column shows it (turn the column on in the column chooser), while this workbench lists those layers as Mapped via family.
Drafted names and definitions. A new entity’s or attribute’s name and definition, and a definition you accept or save in the Review enriched columns dialog, when you keep them exactly as drafted.
Definitions from Promote columns that you kept as drafted. That dialog gives no confidence score, so their badge shows no percentage.

The AI badge comes in two sizes. In detail and editor panels it sits beside the field’s label (for example, beside Name or Definition in an entity’s panel) and reads AI beside a sparkle icon, followed by the confidence as a whole percentage, with the Low confidence icon below 50%. In grids, including this workbench’s Entity column and the Business attribute column on a table’s Columns tab, it is a small sparkle icon beside the value that turns amber when the confidence is low. Hover either one to read “Accepted from an AI proposal”, “Confidence N%” when there is a score, the AI’s rationale, and “It stays marked until someone edits it.” The full badge in a panel also shows this when it has keyboard focus.

The mark stays until the value it describes changes:

Choosing a different target before you accept, with Map to an existing entity in Guided Review or Map in the Review enriched columns dialog, makes the mapping yours from the start. Editing a drafted name or definition before you accept or apply makes that value yours.
Pointing a table at a different entity, or a column at a different attribute, makes the mapping yours. Re-pointing a table at the entity it already maps to keeps the mark.
Editing a marked name or definition, in its editor or in a grid cell, clears the mark on that field only: renaming an entity leaves its definition’s mark in place. Saving without changing a value keeps its mark.
An import that changes a marked value removes the mark from that value. A value the import leaves unchanged keeps it.

Check source readiness

Before you build loads on a newly landed source, run Check whether source tables are ready to load to answer one question per table: is this source ready to load? In the Ask AI panel, while you have the Map & Match workbench open, choose Check whether source tables are ready to load and pick the tables under Tables to assess. The same skill is also offered in the schema explorer and on a project or workspace canvas. Readiness evidence is gathered for exactly the tables you pick, together with the Business Modeling and Data Quality & Profiling foundations the skill cites, and the skill returns a report: it proposes nothing, applies nothing, and leaves your mappings untouched.

For each table, the report covers:

Business keys and their mappings. Every declared business key, whether it is mapped to a business attribute, and whether the key column is nullable. These decide the verdict: a table with no declared key, an unmapped key, or a nullable key column is not ready, because a key that can be null cannot identify a row reliably.
Unmapped columns. How many of the table’s columns have no attribute mapping yet.
Classification coverage. How many columns carry a classification from your organization’s schemes; see Data governance overview for the schemes it reports against.

Unmapped columns and thin classification coverage never block a load on their own: the report counts them as readiness debt, with the exact numbers per table. Each table gets a verdict, ready or not ready, plus the evidence that decided it and what would have to change to reach ready. The report closes with an overall summary that orders your tables from closest to ready to furthest away.

Acceptance settings

Your organization’s AI recommendation workflow setting, found at Settings → Artificial Intelligence, governs how the workbench handles suggestions across your workspace. There are three levels, shown here with the short name the rest of this page uses:

Review each suggestion, or Review each: table suggestions go through Guided Review one at a time before they can be accepted, and Accept selected and Accept all at or above N% are not available. Column suggestions are decided row by row, in the Review enriched columns dialog or on column cards. This setting suits teams that want a person’s decision on every mapping.
Allow bulk review (default), or Allow bulk: the bulk grid, Accept all at or above N%, and the Guided review toggle are all available. You decide when to accept in bulk and when to step through suggestions individually.
Auto-accept high-confidence matches (Not recommended), or Auto-accept: matches and drafted attributes at or above a confidence floor you configure are applied automatically as soon as a run finishes, whether it matched tables or matched and drafted columns. Anything below the floor stays pending for you to review. The confidence floor control is only shown when this level is selected: enter the floor in the Confidence floor (%) field as a whole percentage from 0 to 100 (for example, 80), because a decimal such as 0.8 is refused with “Enter the confidence floor as a whole percentage, for example 90”. The floor does not move the Low confidence mark on review screens, which stays at 50% whatever the floor is.

When Auto-accept runs, the workbench shows a calm banner, “Auto-accepted N suggestions at or above the confidence floor.” Click Review applied on that banner to jump directly to the Mapped filter and inspect what was applied. Each mapping applied this way keeps the suggestion’s confidence and rationale, so the small sparkle icon beside the entity name shows both on hover.

Only organization administrators see the Artificial Intelligence section of Settings, so only they can change the AI recommendation workflow setting. If you cannot find the setting, ask an organization administrator.

Everything is branch-aware

Column matching runs on a feature branch, like any catalog change. Suggestions land as drafts, accepting them is an edit you review, and nothing reaches main until you commit. Viewers see the workbench read-only: the Suggest and accept controls need the editor role. On the main branch the controls explain why column drafts can’t be accepted there.

See Data governance overview for what the mappings feed, and Table details for the derived badges that result.