A design card is how a Warehouse Builder proposes a new semantic model. The card arrives in your builder chat and asks for your decision before the model is built. Read the design, make the choices it asks for, then approve it or reject it with a note.
The card opens with a Semantic model design header carrying the model's name and description.
A status pill reads Needs your approval while the decision is open, then Approved or Rejected.
Requested by names the person who asked for the model. The header also carries a session id you can copy.
A row of chips summarises the current choices: the Owner role, the data mode (tokenized or clear values), the history option, and an object count such as n objects to build.
The card opens expanded while it waits for you, and clicks inside it do not fold it; use the header control to collapse or reopen the whole design.
The sections of the design
Sections appear in this order, and each one expands and collapses:
Goal and grain: what the model is for, the grain sentence, and a chip for its fact type: Transaction fact, Periodic snapshot, Accumulating snapshot, Event fact, Factless fact or Aggregate fact. It shows Business date and example questions; only the first five show until you click Show n more questions.
Diagram: a picture of the design with a Copy diagram source action, and the relationships listed in words below it.
Objects: the objects the model will use, grouped as Fact, Dimensions, Views the model reads and Other, each group with its count, and a badge with the number of personal data columns. Expand a row for its column table: Column, Type, Purpose and Personal data. Rows carry chips for the object's kind and status, Model reads it for a view handed over to the model, and PII n with the personal data columns named in its hover text. A size line reads "n rows · n partitions · n GB".
Your choices: the owner role, the personal data mode, the history depth and the row-level security warning, described below.
Type, plan and versions: when the design has a type column or a plan, a mapping table (Value, Meaning, Grain, Source system, Load), the plan's policy sentence (for example "Plan rows sit in the fact at the same grain as the actuals." or "The plan gets its own semantic model."), the Current version: rule and any plan limits.
Charts and cost: a table of the charts the model will serve, with Chart, Measures, By, Default range, Rows read, Partitions, Joins, Statements and Speed. The speed chips read Fast, Moderate or Slow, and when any chart is slow, a chip counts them. The note reads "Estimates from the warehouse statistics, before anything is built. Bands are advisory; only a chart over 30 seconds or a memory error blocks done."
What this model cannot serve: questions the model will not answer, grouped as Needs a warehouse change, Needs a separate model and Not possible in DataPrism today, each entry with a Recommendation:. The section is left out when there is nothing to list.
Backfill and proofs: figures for Sample load, Sample rows, Full history, Full rows and Full load time, and the table Control totals the tests must match with Total, Period, Value and Source.
Change policy: two lists, I may do without asking and I will ask first, with entries such as Add a column, Change the grain, Run a second full backfill and Change grants.
Make your choices
Owner role: pick one of the roles the requester holds. Up to four roles show as buttons; more show as a list. When the card is read-only it shows Owner role: name, and a single role adds " (your only role)". The admin option is greyed out when it cannot read every table of the design, with the hint "admin needs views over every table of this design".
Personal data in reports: choose Tokenized (recommended), where "Report viewers see tokens instead of the n personal data columns below.", or Clear values, where "Everyone who opens a report on this model sees n personal data columns in clear.". The two options appear only when the design requires this choice.
History depth: pick the history option whose figures fit your needs. Each option lists Daily history, Monthly older (Yes or No), Partitions per source, Total partitions, Disk and Backfill, plus a speed chip such as Fast charts.
You can change the choices only while the card waits for your decision. While a decision is being sent, after a decision and while the chat is answering, the card is read-only, and an option you cannot change is dimmed when it is not selected.
Row-level security
When the design reads a table with row-level security that the model will not carry over, a callout warns Row-level security will not apply. Its text names the fix, in one of two forms:
With an admin owner: "This model ignores the warehouse row-level security of table (on the object itself, or read by object): every report viewer sees every row unless row-level security is added on the model."
With a business role as owner: "object does not carry the row-level security of table, so role and every report viewer of this model will see every row. Fix: hand over table itself, or add the same rule in object's RLS tab."
The callout also carries an I understand checkbox, which appears only while the card waits for your decision. Ticking it lets you approve; changing the owner role clears the tick, so tick it again after you settle on a role. For how row-level security works on a model, see Row-level security and View as.
Approve or reject the design
Review the sections and set the choices as described above.
If the row-level security warning appears, tick I understand.
Click the approve button, which is built from your choices: Approve · owner role · data mode · history option. To send the design back instead, click Reject with a note.
Around the buttons:
The card notes "Approving or rejecting continues this conversation as you. The model is created only when name continues it."
When the approve button cannot be used, the reason shows next to it: "None of the requester's roles can own a semantic model. Reject it with a note."; "Pick another owner role: admin needs views over every table of this design."; or 'Tick "I understand" under the row-level security warning to approve.'
Reject with a note opens an inline form: write the reason and confirm. The field shows a character count in the form n/2000. After a rejection, the card keeps the text as Rejection note: followed by what you wrote.
After a decision the card shows Approved by who · date or Rejected by who · date, where the name is the decision maker, you, or an admin.
Important
Once you approve or reject, the card locks and its choices cannot be edited. If something needs changing, reject with a note so the Warehouse Builder can propose a corrected design.
When the card could not be built
Sometimes the card cannot render the full design and shows only a summary, with this text:
"The design card could not be built for this proposal, so only its summary is shown. Approving applies the choices from the intake; a design that reads tables with row-level security cannot be approved this way: reject it with a note so the Warehouse Builder proposes it again."
In this state you can still approve, and the choices from the intake apply. If the design reads tables with row-level security, reject it with a note instead, so that the Warehouse Builder proposes the design again.
Questions, answered
What is the semantic model design card?
It is the Warehouse Builder's proposal for a new semantic model, sent into the chat for your decision. The model is not built until the design is approved, and you can reject it with a note to have it proposed again. See Warehouse Builder and Lake Builder.
What does "Needs your approval" mean?
The design is waiting for your decision. Read the sections, set the choices and act on the card. Once you decide, the pill changes to Approved or Rejected.
How do I choose the owner role?
Pick the role that will own the model from the roles the requester holds. Up to four roles appear as buttons; more appear in a list. The admin option is greyed out when it cannot read every table of the design; hover it to see "admin needs views over every table of this design".
What is the difference between "Tokenized (recommended)" and "Clear values"?
With Tokenized (recommended), report viewers see tokens instead of the personal data columns. With Clear values, everyone who opens a report on this model sees those columns in clear. Both options state the number of columns they cover. For what the data mode means on the model, see Concepts and glossary.
What do the history depth options show?
Each option lists what it covers: Daily history, Monthly older, Partitions per source, Total partitions, Disk and Backfill, plus a speed chip such as Fast charts. After you decide, only the chosen option stays enabled.
What does "Row-level security will not apply" mean?
The design reads a table with row-level security that the model will not carry over, so every report viewer of the model could see rows the table's rules would otherwise hide. The callout names the fix for each case: hand over the source table itself, add the same rule in the object's RLS tab, or add row-level security on the model. Tick I understand to approve. See Row-level security and View as.
Why can't I approve the design?
The reason appears next to the button. No role the requester holds can own a semantic model: "None of the requester's roles can own a semantic model. Reject it with a note." The admin owner cannot read every table: "Pick another owner role: admin needs views over every table of this design." The row-level security warning is not ticked: 'Tick "I understand" under the row-level security warning to approve.'
What happens after I approve or reject?
The card shows the decision line, Approved by you · date or Rejected by you · date, and notes "Approving or rejecting continues this conversation as you. The model is created only when name continues it." What follows, the access check for the owner role and the hand-off to the Model AI, is described in Access checks and Model AI handoff.
Can I change my choices after I decide?
No. You can change the choices only while the card waits for your decision. After the decision, and while the chat is answering, everything is read-only, and an unselected option is dimmed. To change something, reject with a note and the Warehouse Builder can propose a corrected design.
Why is an object row dimmed?
The object is not part of the build for your current choices. A dimmed row says "only with option" when it needs another history option, or "not built for this choice" when it is outside the chosen scope. Choose the option it needs if the object should be included.
What if the design card could not be built?
Only a summary is shown instead of the full design. You can still approve, and the choices from the intake apply; but a design that reads tables with row-level security cannot be approved this way, so reject with a note so that the Warehouse Builder proposes it again.
What period does the sample load cover?
"The sample is the last 3 closed months plus this month. Full history loads only after the model is finished." When some corrections are left out of the load, a line adds "Corrections older than n days are not loaded."