Free Interactive Flashcards

A merge that matches similar but not identical text, using a similarity threshold, optional transformation table and case and space options. **Why:** It helps with typos like "Infosys Ltd" vs "Infosys Limited", but check the results; a low threshold creates false matches.

All regions; RLS doesn't apply to Admins, Members or Contributors. **Why:** Those roles can edit the semantic model, so security can't be enforced on them. Give restricted users Viewer or app access.

Use Change type > Using locale, choose Date with English (United Kingdom). **Why:** Power Query reads text dates with your computer's locale by default. Setting the source locale tells it the day comes first.

Merge joins tables side by side on matching keys (adds columns); append stacks tables with the same columns (adds rows). **Why:** A simple way to remember: merge = join, append = union.

Any user with write access to an item can promote it; only reviewers the admin authorises can certify it. **Why:** Certified means the organisation vouches for it, which helps users pick the "official" semantic model.

Remove unused columns and rows, make sure steps fold, use incremental refresh, and move heavy transformations upstream; consider capacity if it still doesn't fit. **Why:** Shared capacity stops at two hours, so the fix is mostly less data and more folding, not just a bigger licence.

Give readers the Viewer role or, better, access through an app. **Why:** Members can edit and delete content and bypass row-level security. Least privilege keeps content safe.

A pointer in a lakehouse to data stored elsewhere (another OneLake location or external storage) that appears as if it were local. **Why:** Shortcuts avoid copying data; Direct Lake semantic models can use shortcut tables as sources.

Whether selecting data in one visual filters, highlights or has no effect on each other visual on the page. **Why:** Sometimes a total card should ignore clicks on a chart. Edit interactions sets this per pair of visuals.

Smaller summary tables (often Import) that Power BI uses automatically instead of a large detail table when a query only needs summary data. **Why:** Visuals at summary level hit the fast aggregation table, and only drill-to-detail queries go to the big DirectQuery table.

200. **Why:** AVERAGEX evaluates [Total Sales] for each customer through context transition (100 and 300) and averages them. Customers with no sales return blank and are ignored.

To get a quick plain-language overview of what the model contains, such as its main tables, measures and what questions it can answer. **Why:** It helps when you inherit a model or start a new report, before you dig into the model view.

Yes; M supports // for a line comment and /* ... */ for a block. **Why:** Comments in the Advanced Editor survive, so use them to explain unusual steps.

A semantic model security setting that hides tables or columns from members of a role, so they can't see or query them. **Why:** Unlike hiding a field, OLS makes visuals that use the secured column fail for those users, which protects things like salary data.

A table of categories with a Sales column calculated for each one. **Why:** ADDCOLUMNS adds computed columns to a table. The measure gets context transition per category row.

Not by itself; Power BI access is still controlled by workspace roles and permissions. **Why:** The label classifies the item and protects exported files. Who can open the report is still set by permissions.

`Sales YTD = TOTALYTD ( [Total Sales], 'Date'[Date] )` **Why:** TOTALYTD needs a proper date column, ideally from a marked date table, and adds up from the start of the year to the last date in context.

A Direct Lake semantic model on the gold Delta tables, with capacity sized for the guardrails (for example F128 or above for 3 billion rows per table). **Why:** Direct Lake avoids copying the data and picks up new versions quickly through framing, with VertiPaq query speed.

Your personal workspace. Content there can't be shared through workspace roles or apps, and it belongs to you alone. **Why:** If you leave the company, content in My workspace is hard to recover. Team content belongs in a shared workspace.

No; metadata is copied, not data. The model in Production must be refreshed to load data. **Why:** Deployment clones definitions. Refresh (and credentials) are handled in each stage's workspace.

EARLIER reaches the outer row context in nested iterations. Today, a variable that stores the outer value is clearer. **Why:** `VAR CurrentPrice = Product[Price] RETURN COUNTROWS ( FILTER ( Product, Product[Price] > CurrentPrice ) )` reads more easily than EARLIER.

Allocate shipping cost to order lines with a rule (for example by line value), or keep it at order level and show it only by order attributes. **Why:** An order-level value can't be split by product without a rule. Allocating to lines makes it additive at product grain; otherwise totals by category double count.

Microsoft suggests at least 100 observations for the selected state and at least 10 for the comparison states, and it analyses a sample of up to 10,000 data points. **Why:** Too little data gives "no influencers found"; too many categories with few rows each does too.

EVALUATE followed by a table expression; DEFINE lets you declare measures, variables, columns or functions used only in that query. **Why:** You test measure changes in DAX query view by defining a new version in DEFINE MEASURE and evaluating it before updating the model.

0. **Why:** Number.From("N/A") raises an error; try ... otherwise catches it and returns the fallback value.

Iterate a smaller table, such as VALUES ( Customer[CustomerKey] ) or the dimension you actually need, and compare with a variable. **Why:** Iterating the fact table and calling a measure per row triggers context transition millions of times. Iterating the dimension does far less work.

Only rows related to the West region, in every table filtered by Region. **Why:** The filter applies to Region and flows through relationships to fact tables, so sales of other regions disappear.

A flat table repeats descriptive text on every row, can't hold facts at different grains, and makes some DAX harder. A star schema is smaller and more flexible. **Why:** Power BI compresses repeated values but still stores more. More importantly, a second fact table (like budget) has nowhere to go in a flat design.

No. **Why:** Microsoft notes that these features can't guarantee a specific output; AI behaviour is nondeterministic. Preparation improves consistency but doesn't make it exact.

Microsoft's analytics platform that brings data engineering, data integration, warehousing, data science, real-time analytics and Power BI onto shared capacity and one data lake, OneLake. **Why:** Power BI is one Fabric workload, so Fabric features like Direct Lake and Copilot capacity affect how Power BI is deployed.

A new column computed with an M formula for each row, such as [Quantity] * [Unit Price]. **Why:** The generated step uses Table.AddColumn with an each function, which runs once per row.

Duplicate copies all the steps into an independent query. Reference creates a new query whose source is the original query's output. **Why:** Referenced queries change when the original changes, which keeps shared logic in one place.

A Fabric capacity (F SKU, or a remaining P SKU); Import and DirectQuery work with any Power BI licence. **Why:** Direct Lake also can't be created in My workspace.

Change a visual's type, fields or summarisation for their own view, without edit rights and without changing the report for others. **Why:** Readers can answer small follow-up questions themselves. Authors can turn it on for the report and choose which fields are available through perspectives.

To give one consistent calendar (year, quarter, month, fiscal periods) that all facts share and time intelligence can use. **Why:** A dedicated date table lets you control fiscal years, holidays and sorting, and avoids one hidden auto date table per date column.

Not with Pro alone; Copilot needs a capacity, such as moving the workspace to F2+ or assigning the user to a Fabric Copilot capacity. **Why:** Microsoft documents that a Pro or PPU licence alone isn't sufficient. A Fabric Copilot capacity can bill Copilot use from Pro and PPU workspaces.

A measure that can be added across some dimensions but not across time, such as a stock level or account balance. **Why:** Adding balances for every day of a month makes no sense; you want the balance at the end of the month.

1–28 February 2025. **Why:** DATEADD shifts each date back a month; March dates without a matching February date (29–31) collapse to the end of February.

60. **Why:** RUNNINGSUM adds the current row to all rows above it on the visual's axis: 10 + 20 + 30.

Owners get failure emails, and after several consecutive failures the service disables the schedule. **Why:** Leave failure notifications on and add the team to the notification list so someone acts quickly.

`Title = "Sales for " & SELECTEDVALUE ( Region[Region], "all regions" )` **Why:** VALUES returns a table, which can't be joined to text when it has more than one row. SELECTEDVALUE returns one value or the fallback.

Add a button with the Drill through action pointing to the detail page; it becomes active when a user selects a data point. **Why:** The button can show helper text such as "Select a product to see details", which teaches the interaction.

Use the semantic model's version history in the service, if it is available for that model, to restore an earlier version. **Why:** Editing models on the web is convenient; version history gives a safety net. Git integration provides a fuller history.

It lists every file in a folder and can combine files with the same structure into one table, using a sample file to define the transformation. **Why:** Monthly CSV exports dropped in one folder load automatically each refresh without editing the query.

Select the measure in Model view and choose "Create with Copilot" under the Description box, then keep or edit the result. **Why:** Descriptions appear to report authors and help Copilot. If you change the measure later, run it again; it doesn't update by itself.

Use Performance Analyzer to find the slow visuals, copy their queries into DAX query view or DAX Studio, check whether time is in the measure or the visual, then look at model size, relationships and cardinality. **Why:** The answer shows you measure before you change things: narrow down to a visual, then a query, then the model.

It answers questions and summarises content within the app's curated scope, and can return verified answers that the app's authors prepared. **Why:** App authors can turn it on or off per app (Show Copilot in app navigation). Responses respect the user's permissions within the app.

Power BI Pro (or Premium Per User) for both the person sharing and the people viewing. **Why:** On shared capacity every consumer needs Pro or PPU. Free users can only view content in workspaces on F64 or larger (or P) capacity.

1 January 2025 to 31 May 2025. **Why:** DATESYTD returns the year's dates up to the last date in context, which a CALCULATE then uses to total the year so far.

In a compressed in-memory copy inside the semantic model, loaded at refresh time. **Why:** Queries hit the VertiPaq engine instead of the source, so Import is usually the fastest mode, but the data is only as fresh as the last refresh.

Use bidirectional filtering (with security filter in both directions) only on the security table's relationship, and keep the rest of the model single direction. **Why:** Bidirectional relationships everywhere slow queries and can create ambiguity. Only the security path needs it.

Either in the source (by passing each user's identity with single sign-on) or with RLS in the semantic model. **Why:** With SSO, the source applies its own rules for each user. Without SSO, queries run with stored credentials, so you need RLS in the model.

A continuous date or number axis and a single measure, without a legend. **Why:** The forecast extends one series forward and shades a confidence interval. Multiple series in a legend aren't supported for it.

A like-for-like measure that only includes stores open in both periods, for example by filtering stores whose open date is before the start of last year's period. **Why:** New stores have no prior-year sales, so they inflate growth. A store flag or a filter on open date compares the same set of stores.

It copies the current filter context, applies context transition if there is a row context, applies its filter arguments (removing, replacing or keeping filters), then evaluates the expression. **Why:** Explaining this order shows you understand why filter arguments can override slicers and why measures behave differently inside iterators.

A table with one row per year that has sales, and a Sales column. **Why:** SUMMARIZECOLUMNS groups by Year and evaluates the measure for each; years where every measure is blank are left out.

Total sales of all regions for 2025. **Why:** ALL ( Region ) only removes filters on the Region table. The year filter on the Date table still applies.

2 March 2025 (#date(2025, 3, 2)). **Why:** 2025 isn't a leap year, so three days after 27 February is 28 February, 1 March, then 2 March.

A dimension that keeps history by adding a new row (with a new surrogate key and valid-from/to dates) when an attribute changes. **Why:** A customer who moved region last year still reports old sales under the old region, which is what auditors usually expect.

`CALCULATE ( [Total Sales], USERELATIONSHIP ( Sales[ShipDate], 'Date'[Date] ) )` **Why:** Mentioning the column isn't enough. USERELATIONSHIP activates the inactive relationship for this calculation.

Give a culture: `Table.TransformColumnTypes(Source, {{"OrderDate", type date}}, "en-IN")` (or "en-GB"). **Why:** The culture tells Power Query the text uses day-month-year order.

Rules on a semantic model that filter which rows each user can see, for example only their own region. **Why:** One model and report serve everyone, and each user sees only their permitted data.

The automatic aggregation Power BI creates when you drag a numeric column into a visual, such as "Sum of Amount". **Why:** Explicit measures are better practice: they are reusable, named clearly, work with calculation groups and are easier for Copilot to use.

Investigate whether -99,999 is a placeholder for missing or test data, then replace it with null or filter it out, documenting the rule. **Why:** Placeholder values distort averages and totals. Confirm the meaning with the data owner before removing them.

It marks points outside an expected range and, when you select one, explains which fields are associated with the anomaly. **Why:** It speeds up "why did this day spike?" questions without manual slicing.

No. Until you add users or groups to the role in the service, nobody is a member of it. **Why:** Defining roles and assigning members are separate steps; also, editors of the workspace are never restricted.

Refresh breaks when the analyst leaves or changes their password. Use shared connections or service accounts, take over ownership in time, and keep content in team workspaces. **Why:** Ownership and credentials should belong to the team or a service principal, not to one person.

A simple integer key you create for a dimension instead of relying on a long text or composite business key. **Why:** Integers compress well and make relationships fast. They also let you handle slowly changing dimensions where one business key has several versions.

Select the column, Replace values, value to find null, replace with 0. **Why:** The generated M uses Table.ReplaceValue with null. Only do this when null really means zero; sometimes "unknown" is more honest.

Admin, Member, Contributor and Viewer. **Why:** Each role adds rights: Viewers read, Contributors create and edit content, Members can also share and add people, and Admins manage everything including deleting the workspace.

The Copilot pane in a report, the standalone Copilot experience (full screen from the left navigation), and Copilot in apps. **Why:** Microsoft lists the report pane as generally available, while standalone Copilot and Copilot in apps are in preview.

CALCULATE returns a scalar value; CALCULATETABLE returns a table, both evaluated in a modified filter context. **Why:** Use CALCULATETABLE when you need a filtered table, for example inside COUNTROWS or as an iterator's input.

Write a measure that returns a colour name or hex code, then use Format > Colours > fx > Field value with that measure. **Why:** Field value formatting reads the colour straight from the measure, so the rule lives in one place and can be reused.

`Category = RELATED ( Product[Category] )` **Why:** A calculated column in Sales has a row context on Sales only. RELATED follows the relationship to Product to fetch the matching row's value.

One role whose filter uses the signed-in user's identity, such as `[Email] = USERPRINCIPALNAME()`, on a table that maps users to what they may see. **Why:** Instead of one role per region, a single role reads a security table, so adding a person is a data change, not a model change.

Remove the GUID, and split the timestamp into a date column (and an hour column if needed) or drop the time part. **Why:** Both columns have nearly one unique value per row, so they barely compress. Removing columns nobody reports on is the cheapest speed-up there is.

Display folders group measures and columns in the field list; descriptions show as tooltips and help Copilot and authors understand fields. **Why:** A tidy, documented model is easier to use and gives Copilot better context, since it reads names and descriptions.

Remove top rows (3), then Use first row as headers. **Why:** Removing by position is fine if the layout is stable. If it varies, filter rows using a known header value instead.

It groups rows by one or more columns and aggregates others (sum, count, min, max, all rows). **Why:** Use it to reduce granularity before loading, for example from transactions to daily totals per store.

A newer (preview) approach where you define calendars on a table, mapping columns to categories like Year, Month and Week, and pass the calendar to functions such as TOTALYTD ( [Sales], 'Fiscal Calendar' ). **Why:** It supports any calendar (retail 4-4-5, 13-month), week-based functions like TOTALWTD, and doesn't need a gap-free date column.

The semantic model: clear table, column and measure names, descriptions and relationships, plus Prep data for AI settings. **Why:** Copilot builds from model metadata. Generic names like "Column1" or "Measure 3" give it little to work with.

Add the measure to the tooltips well, then choose Sort axis > that measure. **Why:** Visuals can only sort by fields they contain. Putting the field in Tooltips adds it without displaying it on the axis.

Apps give read-only, curated content with their own navigation and audiences, while the workspace stays a place for developers. **Why:** Developers can work on changes without users seeing half-finished reports, and access is easier to manage.

Groups combine chosen category values (like several small countries into "Other"); bins split a number or date into equal-size ranges (0–9, 10–19). **Why:** Both create a new field without writing DAX or changing the source.

The level of detail one row represents, for example one row per order line per day. **Why:** Every fact table needs a clear grain. Mixing grains in one table (daily sales and monthly targets) causes double counting.

Tenant settings, workspace roles, item and app permissions (including Build), row- and object-level security in the semantic model, and sensitivity labels for classification and export protection. **Why:** Interviewers want to see that you know which layer controls what, and that RLS only restricts viewers.

The colour if exactly one colour is in the filter context, otherwise "Several". **Why:** It is a shortcut for IF ( HASONEVALUE ( ... ), VALUES ( ... ), alternate ) and is handy in titles and dynamic labels.

`Total Sales = SUM ( Sales[Amount] )` **Why:** A base measure like this is reused by other measures, so the logic for "sales" lives in one place.

Most rows fail to match the date table, so date filters return blanks. **Why:** Relationships match exact values. Split the time into its own column, or change OrderDate to a Date type, so keys match.

Reference lines (constant, average, min, max, median, percentile), trend lines, error bars and a forecast. **Why:** These add context such as "are we above average?" directly to the chart without new measures.

No. **Why:** Anomaly detection isn't supported with legends, multiple values or secondary values, and needs a time series axis with at least four points.

Columns with many unique values: IDs, timestamps with seconds, long text and high-precision decimals. **Why:** VertiPaq compresses by column, and columns with few distinct values compress best. Removing or simplifying high-cardinality columns shrinks the model most.

Plain-language guidance for Copilot, such as business definitions ("fiscal year starts in April") or how to interpret fields. **Why:** Instructions add business context that names and descriptions alone don't carry.

No; Microsoft lists enabling Power BI Q&A on the semantic model as a requirement. **Why:** The tooling builds on the model's linguistic setup, so Q&A must be enabled even though Q&A itself is being retired.

DAX Studio runs and times DAX queries and shows model size statistics; Tabular Editor edits model metadata (measures, calculation groups) quickly. **Why:** Both are popular free or commercial external tools that launch from Power BI Desktop's External tools ribbon.

`Revenue = SUMX ( Sales, Sales[Qty] * Sales[Price] )` **Why:** The multiplication must happen per row before adding. SUMX iterates the rows and adds the products.

Every filter that reaches Sales through relationships too, such as Date, Region and Customer. **Why:** Removing filters from the fact table removes all filters on it. To ignore only products, use ALL ( Product ) or REMOVEFILTERS ( Product ).

They stop Power Query from sending data from one source to another source when combining them, for example in a merge. **Why:** Without them, values from a private file could be sent to a public web API as part of a folded query. Mismatched levels can also block or slow combinations.

Standard mode serves many users and sources and supports scheduled refresh and DirectQuery; personal mode is for one user, supports import refresh only and can't be shared. **Why:** Organisations use standard mode, managed by IT, for production reports.

A Fabric storage mode that loads columns straight from Delta tables in OneLake into memory when queries need them. **Why:** Queries run on the same VertiPaq engine as Import, but a "refresh" only updates metadata (framing), so there is no full data copy to reload.

"Fail" for 45 and "Pass" for 80. **Why:** each defines a function run for every row; [Score] reads that row's value.

`Is Loyal = IF ( COUNTROWS ( RELATEDTABLE ( Sales ) ) > 10, "Yes", "No" )` **Why:** A calculated column has a row context but no filter context, so COUNTROWS ( Sales ) counts every sale. RELATEDTABLE returns only the current customer's sales.

Filter the date column in Power Query (ideally with a folding filter) to keep only the needed period. **Why:** Fewer rows means faster refresh and a smaller model. If older years are needed occasionally, consider a separate aggregated table.

Store the measure in a variable first: `VAR AvgOrder = [Average Order] RETURN CALCULATE ( [Total Sales], Sales[Amount] > AvgOrder )` **Why:** A Boolean filter in CALCULATE can't reference a measure. The variable is evaluated first, so the filter compares the column with a fixed value.

When you need a custom range between two dates, such as from a campaign start date to the current date in context. **Why:** Unlike DATESINPERIOD, you give both ends explicitly; blank start or end means the earliest or latest date.

Suggest a sorted bar chart (or a top 5 plus "Other") and explain that people compare lengths far better than angles, especially in 3D. **Why:** Interviewers check that you can push back politely with a reason and an alternative, not just follow requests.

A customer can use several devices, so Device isn't a single attribute of each customer. **Why:** Aggregate it (for example count of devices) or pivot devices into Yes/No columns per customer, so each factor is defined at customer level.

The query keeps its last loaded data and is skipped when you refresh in Desktop. **Why:** Useful in Desktop for static tables (like a list of countries) that are slow to fetch and rarely change, so you don't wait for them on every refresh.

Blank. **Why:** KEEPFILTERS intersects the new filter with the existing one. The row says Blue and the filter says Red, so no product matches.

When CALCULATE runs inside a row context, it turns the current row's values into filters on the model. **Why:** That is why a measure used inside an iterator or calculated column returns the value "for this row's product", and why it can be slow on large tables.

For example: it doesn't support DirectQuery, live connections to Azure Analysis Services or SQL Server Analysis Services, or Publish to web. **Why:** Check the storage mode before promising key influencers on a DirectQuery report.

It removes all filters from the Product table. **Why:** It is the classic way to get a grand total in every row, for example to compute a share of total.

One row per thing you slice by (a product, a customer, a date) with a unique key and descriptive columns. **Why:** Dimensions provide the labels for slicers, rows and axes, and their filters flow into the facts through relationships.

To the most recently created Copilot capacity. **Why:** Microsoft supports only one Fabric Copilot capacity per user; with several, the newest one registers the usage.

The user isn't a member of any RLS role on the semantic model (or their group isn't added). **Why:** With RLS defined, a Viewer who isn't in any role sees no data. Add them, or their security group, to the right role.

The data source credentials on the semantic model (or the gateway connection), for example an expired password or token. **Why:** Credentials are stored in the service, separate from Desktop. Re-entering them under the semantic model settings is often the fix.

It uses your Fabric Copilot capacity if you have one; otherwise it auto-selects a Copilot-enabled workspace on F2+ capacity, which you can change. **Why:** Without an FCC or access to such a workspace, Microsoft says you can't use standalone Copilot.

Make sure the blanks are null, then use Fill down on the Region column. **Why:** Fill down copies the last non-null value into following null cells. Empty text isn't null, so replace "" with null first if needed.

No. **Why:** Microsoft says comments and text in double quotes aren't used, and the measure must be valid with no errors.

Folding is Power Query translating steps into source queries. You protect it by filtering and removing columns early, avoiding non-folding steps before them, and checking View Native Query. **Why:** Interviewers ask this to see if you understand refresh performance and incremental refresh.

A slicer is a visual on the canvas that users click; the Filters pane is a side panel that authors can lock or hide. **Why:** Slicers are visible and inviting for common choices. Pane filters suit fixed rules or less common choices.

Show budget only at levels it exists (category and month), returning blank at product level, or allocate it with an explicit rule such as last year's product share. **Why:** Showing the category budget on every product row looks like each product has the whole budget. ISINSCOPE helps detect the product level.

Sales with all Product filters removed except the filter on Category. **Why:** It gives "category total" in each product row. Filters from other tables still apply.

`Price Band = IF ( Product[Price] > 100, "Premium", "Standard" )` **Why:** It is computed once per row at refresh, so you can put Price Band on a slicer or axis.

Showing or hiding visuals, grouping them, and setting layer order and tab order. **Why:** Combined with bookmarks it lets you swap visuals in place (for example, chart vs table toggles).

A per-user licence that unlocks Premium features (like 48 refreshes a day, XMLA endpoints, deployment pipelines) for licensed users. **Why:** Only PPU users can open content in PPU workspaces, so it suits small teams needing advanced features, not large audiences.

Check Copilot usage in the Capacity Metrics app, bill heavy users to a separate Fabric Copilot capacity, scope the Copilot tenant setting to the right groups, or scale the capacity. **Why:** Copilot shares capacity units with refreshes and queries, so its cost needs the same monitoring as any other workload.

One chart repeated in a grid, once per category value, all sharing the same axes. **Why:** Comparing the same trend across regions is easier in small identical charts than in one busy chart with many lines.

A dedicated date table with one row per day and no gaps, marked as a date table, related to the facts' date columns. **Why:** Classic functions shift and extend dates on that column. Missing days or using the fact's date column gives wrong or blank results.

It finds groups of similar points automatically and adds the result as a new field. **Why:** It is a quick way to spot segments, for example customers with similar spend and visit frequency.

12. **Why:** List.Select keeps 3, 4 and 5; List.Sum adds them.

Two DateTime parameters that incremental refresh uses to filter a table's date column, so the service refreshes only recent partitions. **Why:** You filter rows where the date is >= RangeStart and < RangeEnd. Using >= on one end and < on the other avoids counting boundary rows twice.

The service refreshes it on a schedule (or on demand) like other cloud sources, with no gateway. **Why:** Connecting to the cloud copy, not a local path, is what makes this work.

2. **Why:** By default M rounds to even (banker's rounding). Use Number.Round(2.5, 0, RoundingMode.AwayFromZero) to get 3.

When several semantic models need the same cleaned data, or when you want the transformation to run once in the service and be reused. **Why:** Duplicated queries in many files drift apart. A dataflow gives one shared, governed version.

Microsoft Learn says Q&A experiences are going away in February 2027 and recommends Copilot instead. **Why:** An earlier announcement gave December 2026, so plan to move users off Q&A visuals and dashboard Q&A before the earlier date.

Power Query that runs in the service, with its output stored for reuse (for example in a lakehouse or warehouse) by many semantic models. **Why:** Shared cleaning runs once and every report uses the same result, instead of copying the same queries into each .pbix.

How items in a workspace connect: data sources, dataflows, semantic models, reports and dashboards, and which depend on which. **Why:** Before changing a semantic model you can see which reports would be affected.

A line chart (or area chart), with dates on the X axis. **Why:** Lines make rises, falls and seasonality easy to see. Bar charts work better for comparing categories.

It stops report consumers from removing or changing that level; they can still explore paths inside it and add levels after it. **Why:** Authors can fix the first steps of an analysis while leaving room to explore. Unlocked levels can't come before locked ones.

It is a built-in statistical analysis on a selected data point that ranks contributing factors, without generative AI or Copilot capacity. **Why:** It works in many reports where Copilot isn't available, but only answers that one kind of question.

31. **Why:** List.Dates returns 31 dates, one day apart, starting 1 January; List.Count counts them.

Report authors should slice by descriptive columns, not by surrogate keys. **Why:** Hidden columns still work in relationships and DAX. Hiding them keeps the field list clean and prevents filtering on the fact table's key.

Rename the column in the sample file transform (or standardise column names in the combine function) before appending. **Why:** Append matches columns by name, so different names create separate columns filled with nulls.

Viewing and interacting with content in the workspace, without editing it. **Why:** Viewer is for consumers. Row-level security applies to Viewers, unlike the other three roles.

It swaps rows and columns, useful for reports where headers run down the first column. **Why:** After transposing you usually promote headers. Transpose ignores column names, so do it before naming columns.

It makes you the owner of the semantic model's settings, such as credentials and refresh schedule. **Why:** Workspace members can take over when the original owner is unavailable, then re-enter credentials so refresh works again.

There would be two active filter paths between the same tables, so Power BI couldn't tell which one a filter should follow. **Why:** Keep one path active, make the other inactive and use USERELATIONSHIP in measures, or reshape the model so paths don't loop.

Group membership is managed in Microsoft Entra ID, so joiners and leavers get the right access without editing the semantic model. **Why:** It scales and keeps access in line with HR or IT processes rather than manual updates by report authors.

How long the DAX query, the visual display and other work took, and it can copy the DAX query the visual sent. **Why:** It tells you whether the slowness is in the model (DAX query time) or the visual itself, and gives you the query to test elsewhere.

Blank. **Why:** With more than one value in context and no alternate result given, SELECTEDVALUE returns blank.

Data can be prepared once in OneLake (lakehouse or warehouse) and served by Direct Lake models, Copilot runs on Fabric capacity, and items share Git integration and deployment pipelines. **Why:** The answer shows you understand storage choices, capacity and lifecycle, not just Power BI Desktop.

Both return unique values visible in the current context, but VALUES also includes the blank row that appears when there are unmatched keys; DISTINCT does not. **Why:** The difference only shows up with referential integrity problems, but it explains a surprise "(Blank)" count.

Transform data > Data source settings (to change source or edit permissions), or edit the Source step or a parameter. **Why:** Data source settings holds credentials and privacy levels for each source. Parameters make server and database names easy to switch.

null. **Why:** In M, arithmetic with null returns null. Use a check like if [x] = null then 0 else [x], or the ?? operator, to supply a default.

400. **Why:** SUM without CALCULATE gets no context transition, so for each customer row it returns total sales of all customers (400), and the average of 400 and 400 is 400.

No; the reader selects the visual's refresh button to get an updated summary. **Why:** Microsoft documents that the narrative doesn't update automatically when filters change.

Separate workspaces per stage connected by a deployment pipeline, deployment rules for data sources, Git integration for source control, and an app published from Production. **Why:** The answer shows you separate building from consuming and can roll back changes.

A UserRegion table (Email, Region) related to Region, and one role filtering UserRegion[Email] = USERPRINCIPALNAME(). **Why:** The relationship from the security table to Region must pass the filter to Region (often bidirectional with "apply security filter in both directions"), so each manager sees exactly their regions.

A calculated column is computed for each row at refresh and stored; a measure is computed at query time for whatever filters the visual applies. **Why:** Columns cost memory but can slice and filter. Measures cost nothing until used and react to slicers, which is why most numbers should be measures.

The number of all products, because the Sales filter doesn't flow to Product. **Why:** Use CALCULATE ( COUNTROWS ( Product ), CROSSFILTER ( Sales[ProductKey], Product[ProductKey], BOTH ) ) to count only products sold online.

"Mar 2025" (as text). **Why:** FORMAT returns text using the format string; MMM is the short month name. Because it is text, it sorts alphabetically unless you sort by another column.

No. Power Query compares case-sensitively, but the model engine is case-insensitive, so it sees "abc" and "ABC" as the same value. **Why:** Standardise the case (for example Uppercase) before removing duplicates on key columns.

Relate it to the dimensions it shares at its own grain: Region to the Region dimension and Month to a month-level column or month table. **Why:** Power BI relationships use one column each side. Joining targets at their own grain (not to the daily fact) avoids double counting.

When you need a value per row to slice, filter or group by, such as an age band or a "Large order" flag. **Why:** Calculated columns are computed at refresh and stored. Measures are computed at query time and can't be used as slicers or row labels.

It tells standalone Copilot the model is ready, removing the extra friction treatment on its answers; reports using the model count as approved too. **Why:** The setting was previously called "prepped for AI". It is set on the semantic model in the service; reports and apps can't be marked directly.

It applies the values of one table or column as filters on other columns, as if a relationship existed. **Why:** It is useful for virtual relationships, for example filtering a Budget table by values chosen in an unrelated slicer.

When there is no relationship to follow, for example looking up an exchange rate by currency and date from an unrelated table. **Why:** RELATED is faster and simpler when a relationship exists; LOOKUPVALUE searches by the columns you give it.

4. **Why:** DISTINCTCOUNT counts blank as a value. DISTINCTCOUNTNOBLANK would return 3.

Creating a calculation group turns on "discourage implicit measures", so automatic summaries of columns stop. **Why:** Calculation items only apply to explicit measures, so the model requires them. Create explicit measures for the numbers people need.

A page setting that re-queries visuals on a fixed interval or when a change is detected. It works with DirectQuery (and some other non-Import) sources. **Why:** Import data only changes at refresh, so re-querying it on a timer would show nothing new. Admins can set a minimum interval.

The report canvas: the visuals you include. **Why:** That is why clear visual titles and axis names matter. Visuals on hidden pages aren't included unless a supported bookmark makes them visible.

A feature that lets report readers choose which measures or columns a visual shows, through a slicer. **Why:** One chart can switch between Sales, Profit and Units, or between Region and Product, instead of building one chart per option.

A let expression: named steps separated by commas, then in followed by the step whose result the query returns. **Why:** Each step usually refers to the previous one by name. The result is whatever comes after in, normally the last step.

Use a Top N filter on the Product field in the visual's filters, ranked by a measure such as Sales. **Why:** Top N filters work per visual and update as other slicers change.

Split column by delimiter (comma), then Trim both new columns. **Why:** Splitting leaves a leading space on the second part, so Trim keeps values consistent for slicers and relationships.

When attribute-value pairs (Attribute = "Colour", Value = "Red") need to become separate columns. **Why:** Pivot turns one column's values into column headers and fills them from another column, optionally aggregating.

`Avg Price = DIVIDE ( [Total Sales], [Units] )` **Why:** IFERROR hides problems and can slow queries. DIVIDE handles the only expected error here, a zero or blank denominator.

What Copilot used from the semantic model to reach an answer. **Why:** It helps authors check whether Copilot picked the right fields and whether Prep data for AI settings were applied.

Edit access to the report or workspace, or build access to the semantic model. **Why:** It's a visual, so it needs the same rights as creating any visual. Readers without edit rights can use the Copilot pane to get a summary instead.

Build one staging query for the shared cleaning, turn off its load, and Reference it from the three queries. **Why:** Shared steps live in one place. Note that each referenced query may still read the source separately at refresh.

Filters on this visual, filters on this page and filters on all pages. **Why:** Each applies to a different scope. Report-level filters are useful for rules like "exclude test stores" everywhere.

The data preparation engine (opened with Transform data) that connects to sources and shapes data before it loads into the model. **Why:** Each change you make is recorded as an applied step in the M language, and the steps re-run on every refresh.

When you right-click a data point that rose, Power BI looks across other fields to find which ones contributed most to the change. **Why:** It speeds up the first "why did this go up?" question, though you should still check the result makes business sense.

After removing duplicates, Add Column > Index column (from 1), then merge it back to the fact on the business key to replace it. **Why:** Integer keys are compact. Keep the business key in the dimension for lookups and drillthrough.

Reference measures without a table name ([Total Sales]) and columns always with one (Sales[Amount]). **Why:** The convention tells readers instantly whether something is a measure (with implicit CALCULATE) or a column, which changes how it behaves.

Not on its own; you need a bridge table (CustomerAccount) between them. **Why:** The pairs live in the bridge. Customer filters the bridge, and the bridge filters Account, usually with one relationship set to both directions or CROSSFILTER in the measure.

Identify the business process and grain, separate facts from dimensions, build a date table, set relationships one-to-many single direction, then write base measures. **Why:** Interviewers look for a repeatable method: grain first, star schema next, then DAX. It shows you won't start with visuals and fix the model later.

No; Copilot speeds up building and exploring, but it depends on a well-designed semantic model, and people still check its output and own the business logic. **Why:** Microsoft itself says unprepared models give generic or inaccurate answers, which is why modelling and Prep data for AI matter more, not less.

A field parameter that lists the three measures, or a disconnected table with a SWITCH measure on SELECTEDVALUE. **Why:** Field parameters are the newer, no-DAX option; the SWITCH pattern still helps when you need custom logic per choice.

When values are spread across columns, like Jan, Feb, Mar as separate columns, and you need one Month column and one Value column. **Why:** Models and DAX work with tall tables. Unpivoted data lets one date relationship and one measure handle every month.

A paid Fabric capacity (F2 or higher) or Power BI Premium (P1 or higher). **Why:** Microsoft states that trial capacities and free SKUs aren't supported, and that a Pro or PPU licence alone isn't enough; Copilot needs organisational capacity (or a Fabric Copilot capacity for billing).

When you switch topics, so earlier conversation doesn't confuse new answers. **Why:** Clearing removes previous context. It doesn't refresh data.

"Users can use Copilot and other features powered by Azure OpenAI". **Why:** Microsoft says it is on by default, but admins can turn it off or scope it to groups. If Copilot is missing, this is one of the first things to check.

Reduce the number of visuals on the page, for example with drillthrough pages, tooltips or bookmarks. **Why:** Every visual sends its own queries and renders separately. Many small fast visuals still add up, so fewer visuals per page loads faster.

Each category's share of total sales across all products. **Why:** The denominator ignores product filters, so it always shows the grand total; the numerator keeps the row's category.

Variables store a value once, make formulas easier to read and debug, and avoid computing the same expression twice. **Why:** A variable is evaluated in the context where it is defined, which also makes behaviour clearer when you change filters later.

Refresh fails unless an on-premises data gateway can reach that path; the service can't see your desktop. **Why:** Store the file in OneDrive or SharePoint and connect to it there, so the service refreshes without a gateway.

Return blank for dates after the last sales date, for example `IF ( MIN ( 'Date'[Date] ) <= CALCULATE ( MAX ( Sales[OrderDate] ), ALL ( Sales ) ), [Sales YTD] )`. **Why:** YTD repeats the running total on future dates. Comparing with the last date that has data hides those points.

Generative AI built into Power BI that answers questions about data, summarises reports, builds report pages and helps write DAX. **Why:** It uses the semantic model when a question is about your data, and the language model's general knowledge otherwise, so model quality shapes the answers.

Q&A is being retired (February 2027 on Microsoft Learn), so the report would need rework soon. **Why:** Use Copilot experiences, verified answers and well-designed visuals for natural-language questions instead.

Look at column quality and distribution (on the entire dataset), confirm data types, check keys for duplicates and blanks, then look for odd values like "N/A" or dates in 1900. **Why:** Catching quality problems before modelling saves you from explaining wrong totals later.

Use "View as" with a role (and optionally another user) in Desktop, or "Test as role" on the semantic model's Security page in the service. **Why:** Testing catches roles that show too much or nothing, before real users see the report.

Copilot can answer from cache when the same prompt is asked on an unchanged model (same schema, data and instructions) within a rolling 24 hours. **Why:** Microsoft documents this caching; clearing the chat doesn't change it. Change the prompt or refresh the model if you expect a different result.

The link between an item in one stage and the same item in the next stage, used to decide what a deployment overwrites. **Why:** Unpaired items with the same name are not overwritten; deploying creates a duplicate instead, which surprises people.

Use variables to avoid repeated work, filter columns rather than whole tables, and iterate small dimension tables instead of the fact; also check model design (cardinality, relationships). **Why:** Most slow DAX comes from iterating big tables with context transition or from complex FILTER arguments.

The key influencers visual. **Why:** Microsoft notes the narrative visual doesn't support all visual types yet, giving key influencers as an example.

Put the content in a workspace on a Fabric F64 or larger capacity (or remaining P SKU), and give readers the Viewer role or app access. **Why:** Microsoft's migration guide states that on F64 and larger, Free users with the Viewer role can view content; on F2 to F32 every viewer needs Pro or PPU.

"POWER". **Why:** LEFT takes the first five characters, "power", and UPPER converts them to capitals.

A model with fact tables in the middle, holding events and numbers, linked to dimension tables that describe them (date, product, customer). **Why:** Power BI's engine, DAX and filters are designed around this shape, so star schemas give simpler DAX and faster queries.

In a row context on the one side, the related rows from the many side, for example all Sales rows for the current customer. **Why:** COUNTROWS ( RELATEDTABLE ( Sales ) ) in a Customer calculated column gives each customer's number of orders.

A view where you write and run DAX queries (EVALUATE ...) against the model and see the result grid. **Why:** You can paste a slow visual's query from Performance Analyzer, test changes to measures, and define measures before adding them to the model.

The relationship between two measures across many items, for example discount vs margin for each product. **Why:** Each dot is an item, placed by two values (and sized by a third). Clusters and outliers stand out at a glance.

`Red Only = CALCULATE ( [Total Sales], KEEPFILTERS ( Product[Colour] = "Red" ) )` **Why:** The plain filter overrides the row's colour. KEEPFILTERS intersects it with the row, so non-red rows become blank.

Use the SharePoint folder connector with the site URL, filter to the right folder and file type, then combine files. **Why:** A local folder path needs a gateway and breaks when the author's laptop is off. SharePoint is a cloud source, so the service refreshes it directly.

When Power Query turns your steps into a single query the source runs, such as one SQL statement, instead of processing data itself. **Why:** Folding moves the work to the database, transfers fewer rows and makes refreshes much faster. It is also required for incremental refresh to be efficient.

"Apple, Pear, Plum". **Why:** CONCATENATEX joins the values of an expression over a table with the separator, here sorted ascending by name.

DIVIDE handles division by zero or blank, returning blank (or an alternate value you give) instead of an error or infinity. **Why:** Ratios such as margin % often hit zero denominators in some filter contexts. DIVIDE keeps visuals clean.

42. **Why:** (x) => x * 2 defines a function; calling Double(21) runs it with x = 21.

Reusable functions you define with the FUNCTION keyword and call from measures, calculated columns, visual calculations or other functions. **Why:** Microsoft documents them as generally available in Power BI Desktop and the service from the June 2026 release (compatibility level 1702 or higher). They keep business rules in one place.

It runs a logistic regression (with ML.NET) and keeps factors that pass a Wald test at a p-value of 0.05. **Why:** Numeric metrics use linear regression instead, and Top segments uses a decision tree, all with the same 0.05 threshold.

The set of filters active when a measure is evaluated: from slicers, visual rows and columns, page and report filters, and CALCULATE. **Why:** The same measure returns a different number in each cell of a matrix because each cell has a different filter context.

A format string written as a DAX expression, so one measure can show, for example, "₹1.2M" or "₹950K" depending on its value. **Why:** The value stays numeric (charts and sorting still work) while the display changes.

Use folders in the workspace to group reports, models and other items. **Why:** Folders keep a busy workspace navigable. Deployment pipelines copy the folder structure to later stages.

No; Microsoft says it must return a str (string) to be added to a report. **Why:** The button shows the returned message to the user; the actual changes happen in the data source.

Power BI translates the DAX query into the source's query language (for example SQL) and sends it to the source at that moment. **Why:** Data stays in the source and is current, but every visual depends on the source's speed and capacity.

In workspace settings, choose the license mode (for example Fabric capacity) and select the capacity, if you are a workspace admin with capacity assignment rights. **Why:** The workspace's capacity decides which features (Direct Lake, Copilot, 48 refreshes) and which viewers are supported.

0. **Why:** The third argument is the alternate result used when the denominator is zero or blank.

Referring to other rows in an ordered set: OFFSET for a relative row (previous month), INDEX for a position (first or last), WINDOW for a range (moving totals). **Why:** They make "compare with previous row" logic much simpler than older patterns with EARLIER or FILTER.

Expand adds the nested table's rows and columns to the parent (often adding rows); Aggregate summarises the nested values into one value per parent row. **Why:** After a merge, choose Aggregate when you need a count or sum per row instead of extra rows.

There are far fewer administrators, so the visual doesn't have enough data to be confident the pattern isn't chance. **Why:** The regression accounts for the number of data points; small groups need a bigger effect to count.

It stacks the rows of two or more tables with the same columns into one table. **Why:** Use it for monthly files or regional tables with identical structure. Columns are matched by name, and missing columns are filled with null.

Creating new reports on the model, using Analyze in Excel, and querying it with tools like DAX query view or a composite model. **Why:** Read permission only lets users view reports; Build lets them reuse the model for their own analysis.

Use // or -- for a line comment and /* ... */ for a block. **Why:** Short comments that explain why a filter is removed or a rule exists save the next person (often you) a lot of time.

40. **Why:** SUMX computes 2 × 5 = 10 and 3 × 10 = 30, then adds them. Multiplying SUM(Qty) by SUM(Price) would give 75, which is wrong.

A measure-based approach, for example a disconnected Segment table and a measure that counts customers whose sales fall in each band. **Why:** A calculated column is fixed at refresh and can't react to slicers. Dynamic segmentation needs the band to be evaluated in the current filter context.

Connecting a workspace to a branch in Azure DevOps or GitHub, so items are saved to and updated from Git. **Why:** Teams get version history, code review and branching for semantic models and reports, often saved as Power BI projects.

It returns a table grouped by the columns you list, with the measures or expressions you add as new columns. **Why:** It is the function Power BI visuals use in their queries, and it is the usual choice for summary tables in DAX queries.

Yes, the PDF connector detects tables and pages. Watch for tables split across pages and layouts changing between files. **Why:** PDF extraction depends on the layout. Test a few months of files and add steps that check expected columns exist.

Add alt text to visuals, set a logical tab order, and don't rely on colour alone (use labels or markers too) with enough contrast. **Why:** Screen readers read alt text and follow tab order; colour-blind users need a second cue besides colour.

It is computed at refresh, so it ignores slicers and goes stale until the next refresh. **Why:** Anything that should react to report filters or the current date belongs in a measure.

Control who can create workspaces, certify official semantic models, and review usage metrics or the activity log to archive unused content. **Why:** Governance is about making the right content easy to find and the wrong content easy to retire.

`YoY % = DIVIDE ( [Total Sales] - [Sales PY], [Sales PY] )` **Why:** Build it from base measures. DIVIDE returns blank when there is no prior year, instead of an error.

Background colour, font colour, data bars, icons and web URLs, based on values or rules. **Why:** It draws the eye to what needs attention, such as red for negative growth, without adding more visuals.

ALLSELECTED ( Product ). **Why:** ALL ( Product ) keeps the slicer on Region but clears any product slicer; ALLSELECTED clears only the visual's product rows and respects every outer selection.

A one-column table with 0, 10, 20 … 100 (11 rows), in a column named Value. **Why:** Numeric range parameters use it behind the scenes to build the slider values.

A gateway cluster: install another gateway on a second machine and add it to the same cluster. **Why:** Clusters give high availability and can balance load across members.

`Dates = CALENDAR ( DATE ( 2024, 1, 1 ), DATE ( 2025, 12, 31 ) )` **Why:** The end date was 31 January 2025, so most of 2025 is missing. Date tables should cover whole years for every date in the facts.

Make both columns the same data type, ideally whole number. **Why:** Relationships match values exactly, and mismatched types either fail or cause silent misses. Integer keys are also faster.

Copilot proposes possible pages based on the model, which you can choose from and then refine. **Why:** It is a starting point for a report, not a finished design. Authors still check the visuals and numbers.

Compare with a number: `CALCULATE ( [Total Sales], 'Date'[Year] = 2025 )` **Why:** DAX won't compare a whole number column with text. Remove the quotes so both sides are numbers.

Groups of users inside one app, each seeing a different set of the app's reports and dashboards. **Why:** Executives and analysts can get tailored views from the same workspace without building two workspaces.

Settings on a target stage that change data source connections or parameter values when content is deployed there. **Why:** Test can point at the test database and Production at the live one, without editing the .pbix between stages.

Change a relationship's filter direction (Both, OneWay or None) for that calculation only. **Why:** You keep the model single-direction and make one measure filter in both directions, avoiding the risks of a bidirectional model.

Connecting tools such as Tabular Editor, SQL Server Management Studio or DAX Studio to semantic models in capacity or PPU workspaces, to read or write model metadata and run refreshes. **Why:** It brings enterprise lifecycle tools to Power BI, such as partition refreshes and scripted deployments.

`Share = DIVIDE ( [Sales], COLLAPSE ( [Sales], ROWS ) )` **Why:** Dividing a value by itself is always 1. COLLAPSE evaluates Sales one level up the rows axis, giving the category total as the denominator.

So Power BI does not add them up when you drag them into a visual. **Why:** Numbers that are really labels or keys should not be summed. Turning off default summarization prevents silly totals like "Sum of Year".

Answer questions about the data, Analyze report visuals, and Create new report pages. **Why:** Authors pick skills to simulate what end users see in each place, for example only "Answer questions" for standalone Copilot.

Add the report as a tab in a Teams channel (or share the app in Teams). **Why:** Teams embedding respects the same sign-in, permissions and row-level security as the Power BI service.

Let the security filter reach Region: set that relationship's cross-filter direction to Both and tick "Apply security filter in both directions". **Why:** With single direction, filters flow from Region to UserRegion, not the other way, so filtering UserRegion doesn't limit Region or the facts.

The column with 1 million distinct values. **Why:** VertiPaq stores a dictionary of distinct values and compresses repeats. Cardinality, not row count, is usually what hurts.

Direct Lake on OneLake reads Delta tables from one or more Fabric sources and never falls back to DirectQuery; Direct Lake on SQL uses one source's SQL analytics endpoint and can fall back to DirectQuery. **Why:** Fallback happens, for example, with SQL views or SQL-based security. The Direct Lake behavior property controls whether fallback is allowed.

Put the key numbers top-left, keep one question per page, align visuals to a grid, cut duplicate charts and move detail to drillthrough or tooltips. **Why:** People scan from the top left. A clear hierarchy of information beats extra visuals.

Use the date table: `TOTALYTD ( [Total Sales], 'Calendar'[Date] )`, with Calendar related to Sales and marked as a date table. **Why:** Time intelligence should run on a complete date column from the date table, so filters from the calendar's slicers and axes apply.

The RangeStart/RangeEnd filter doesn't fold to the source, so Power Query downloads everything and filters locally. **Why:** Incremental refresh depends on query folding. Check View Native Query on the filter step and move non-folding steps after it.

Add a comma at the end of the Promoted line. **Why:** Every step in a let block except the last is followed by a comma.

No; only Admins can delete a workspace. **Why:** Members can add people, publish and update apps, and edit content, but deleting the workspace is reserved for Admins.

No; clicking a tile opens the report (or link) it came from. **Why:** Dashboards are monitoring surfaces. Interactive filtering happens in reports, or in a pinned live report page.

`CALCULATE ( [Total Sales], REMOVEFILTERS ( 'Date' ) )` **Why:** Removing filters only from the Date table keeps every other filter, such as region, in place.

Connect with Web or the JSON connector, convert the list or record to a table, then expand the record columns. **Why:** JSON arrives as nested lists and records. Expanding (the double-arrow button) flattens them into columns.

It multiplies the totals instead of each row, so the result is too big whenever there is more than one row. **Why:** Row-by-row maths needs an iterator: SUMX ( Sales, Sales[Qty] * Sales[Price] ).

Two hours on shared capacity and five hours on Premium/Fabric capacity (refreshes through the XMLA endpoint can run longer). **Why:** Long refreshes are a signal to reduce data, fix folding or use incremental refresh.

The current state of a page: filters, slicers, visual visibility, sort order and drill state (you choose which of data, display and current page to include). **Why:** Buttons linked to bookmarks let you build toggles, guided stories and "reset filters" buttons.

Two date dimension tables, for example Order Date and Ship Date, each with an active relationship to Sales. **Why:** An inactive relationship only works inside a measure. Separate tables give each role its own active filter path, so both slicers can apply together.

On dashboard tiles of cards, KPIs and gauges in the Power BI service. **Why:** Alerts notify you when a value crosses a threshold. They are set on dashboards, not directly on report visuals.

A table shows flat rows and columns. A matrix groups rows and columns into levels with subtotals, and can be expanded and collapsed. **Why:** Use a matrix for pivot-style analysis (Region by Month), and a table for detail lists.

Up to 10,000 characters, across Copilot surfaces including the standalone experience, the report pane, apps, mobile and Embedded. **Why:** Longer prompts let you give detailed instructions and context in a single message.

[Sales] - 0 returns 0; [Sales] * 1 returns blank. **Why:** In subtraction and addition, blank is treated as 0. In multiplication, blank stays blank.

Standalone Copilot can find and answer questions about any report or semantic model you can access; the pane only covers the open report. **Why:** It is a preview, full-screen experience and needs its own tenant setting plus a Copilot-supported capacity for billing.

Use a page navigator, or buttons with a Page navigation action. **Why:** The page navigator creates a button for each page and updates when pages are added, renamed or hidden.

It opens Excel connected live to a Power BI semantic model, so users build PivotTables with the model's measures. **Why:** Excel users keep their habits while numbers come from the governed model. It needs Build permission.

You can include Copilot-generated summaries when you set up subscriptions to a report. **Why:** Recipients get a short written summary along with the snapshot, which helps busy readers.

Merge the two tables into one Customer dimension. **Why:** A one-to-one relationship adds a join without adding meaning. One table is simpler for authors and for the engine.

Summarise the report or a topic and answer questions about the report's data. **Why:** It is scoped to the open report. To search across reports and models you have access to, use the standalone Copilot.

A separate portrait layout of a page where you place and resize visuals for phones. **Why:** The Power BI mobile app shows the mobile layout when it exists, so phone users don't have to pinch and zoom a desktop page.

Not to get data; queries go to the source live. Power BI may still refresh tile caches on dashboards on a schedule. **Why:** DirectQuery reads current data on demand, so you manage source performance rather than refresh windows.

Key influencers, decomposition tree and the Q&A visual (smart narrative/narrative is a fourth). **Why:** These visuals use machine learning or natural language to explain data, without writing code.

`(_) => _ * 2` **Why:** each creates a one-parameter function whose parameter is named _. In table steps, [Column] inside each means _[Column].

A way to group workspaces by business area (like Finance or Sales) for governance and discovery. **Why:** Domain admins can manage settings for their area, and users can filter the catalog by domain.

App audiences for which reports each department sees, and row-level security for which rows they see. **Why:** Audiences control item visibility; only RLS filters data inside a report.

It creates (or replaces) a semantic model and a report with the same name in that workspace. **Why:** Republishing overwrites both. Report-only changes still republish the model unless the report uses a live connection to a separate model.

Summarising data to the level reports need (for example daily per store instead of per transaction) means far fewer rows to scan. **Why:** If nobody looks at individual receipts, storing them only makes the model bigger and slower.

When an Import model in a capacity workspace must grow beyond the default size limit (10 GB compressed for upload). **Why:** It lets the model grow up to the capacity's memory limit and improves XMLA write performance.

To create a column from if-then-else rules, for example "If Amount > 1000 then Large else Small". **Why:** It writes an M if expression through a simple form. For complex rules, edit the formula directly.

Merge the columns into one text column with a separator (for example Store|Date) or build an integer key. **Why:** Relationships use one column. A separator avoids clashes like 1 + 23 vs 12 + 3.

A deployment pipeline (Development, Test, Production), with only a few people allowed to deploy to Production and changes tested in Test first. **Why:** Developers work in Development, testers check Test, and Production changes only through controlled deployments that can be compared first.

The lookup table had duplicate keys, so some rows matched more than once. **Why:** A merge returns every match. Remove duplicates in the lookup (or fix the key) before merging.

One standard deviation of tenure (in its unit, here months). **Why:** For continuous factors the visual reports the effect of increasing the factor by its standard deviation.

Lead with three to five key numbers and their trend, add one or two charts that explain them, and move detail to drillthrough pages or tooltips. **Why:** Executives want the answer first and the evidence on request. Mentioning layout, hierarchy and drillthrough shows practical design sense.

In the Microsoft Fabric Capacity Metrics app, filtered to the "Copilot in Fabric" operation. **Why:** Copilot usage counts as background capacity operations measured in capacity units, so it shares resources with refreshes and queries.

Open the report's usage metrics report in the service, or use the admin activity log for tenant-wide analysis. **Why:** Usage data tells you which reports matter and which can be retired.

Microsoft Purview Information Protection labels (such as Confidential) applied to Power BI items. **Why:** Labels show users how sensitive content is, and they can stay on data exported to Excel, PowerPoint or PDF, carrying protection outside Power BI.

120. **Why:** Square brackets with name = value pairs make a record; [Sales] after it reads that field.

`each [Price] - ([Discount] ?? 0)` **Why:** Arithmetic with null returns null in M. The ?? operator substitutes 0 when Discount is null.

Certify the official one, retire or consolidate the others, and point reports to the certified model. **Why:** Endorsement badges appear in the OneLake catalog and Get data lists, guiding users to the right model.

Sharing gives individuals access to one item; an app packages several items with navigation and audiences, and changes reach users only when you update it. **Why:** Apps scale better for large audiences and protect users from changes in progress.

The Analytics pane forecast on a line chart with a date axis; explain it is a statistical projection with a confidence band, not a business plan. **Why:** It extends past patterns and seasonality. It doesn't know about price changes or new stores, so treat it as one input.

A descriptive key, such as an order number, that stays in the fact table because it has no other attributes worth a separate table. **Why:** Creating a dimension with only the key would just duplicate the column. It can stay in the fact and still be used in tables or drillthrough.

For example: files (Excel, CSV, PDF, folders), databases (SQL Server, Oracle, PostgreSQL), cloud services (SharePoint, Dataverse, Salesforce) and Fabric items (lakehouses, warehouses, semantic models). **Why:** There are hundreds of connectors. Choosing the right one matters because some support DirectQuery and query folding, and others don't.

Absolute picks the highest or lowest value; Relative picks the value that stands out most compared with the others in its column. **Why:** Absolute can choose a field with a single value (trivially the highest). Relative finds surprising splits, like a value four times the expected share.

Default colours, fonts, and formatting for visuals across the report. Themes can be saved and imported as JSON files. **Why:** Themes keep branding consistent and save formatting each visual by hand.

Headcount at the last month in context, for example CALCULATE ( SUM ( HR[Active] ), LASTDATE ( 'Date'[Date] ) ) (or LASTNONBLANK if month-ends can be missing). **Why:** Headcount is semi-additive: it adds across departments but not across time.

Microsoft is retiring them and no longer sells new ones; customers migrate to Fabric (F) capacity SKUs when their P subscription term ends. **Why:** Pro and Premium Per User licences aren't affected. Plan the F SKU move before the P subscription ends to avoid throttling after the grace period.

11. **Why:** taxRate is optional with a default of 0.1, so AddTax(10) computes 10 × 1.1.

120. **Why:** COUNTROWS counts rows regardless of values. To count only sales, filter first, for example CALCULATE ( COUNTROWS ( Sales ), Sales[Amount] > 0 ).

`[RepEmail] = USERPRINCIPALNAME ()` **Why:** USERPRINCIPALNAME returns the signed-in user's UPN, usually their email. USERNAME can return DOMAIN\user in Power BI Desktop, which doesn't match an email column.

Use the Power BI add-in for PowerPoint (or Export > PowerPoint > Embed live data) to insert the report page. **Why:** The slide shows current data and can be filtered during the presentation, respecting the viewer's permissions.

It emails users a snapshot of a report page (or attachment) on a schedule, with a link back to the report. **Why:** Busy readers get the numbers without logging in. Subscribing other people needs a Pro or PPU licence and at least the Contributor role in the workspace.

A setting that automatically writes the data of Import tables to Delta tables in OneLake. **Why:** Other Fabric tools (SQL, notebooks, shortcuts) can then use the same data without a separate export.

The service needs to know the data source before running the query; a fully computed URL hides it. **Why:** Use Web.Contents with a fixed base URL and the RelativePath and Query options, so the base address is static.

A Microsoft-managed gateway that connects to data sources inside an Azure virtual network without you installing gateway software. **Why:** It suits Azure sources locked down with private endpoints. Microsoft lists it for P, F and A4-or-higher capacities (F8 or above recommended).

You type the result you want for a few rows and Power Query works out the transformation (like extracting a first name or a year). **Why:** It writes the M for you. Check the formula it generated before relying on it.

Publish one shared semantic model and build the five reports with live connections to it. **Why:** One model means one refresh, one set of measures and one place to fix things.

Write a query like (x as number) as number => x * 2, or right-click a parameterised query and choose Create function. **Why:** Functions let you apply the same logic to many files or rows, which is how Combine files works behind the scenes.

Bidirectional filters can create ambiguous paths, slow queries and surprising results. **Why:** Use single direction by default. When one visual needs a reverse filter, use CROSSFILTER inside a measure, or a measure-based filter on the slicer, instead of changing the model.

HASONEVALUE is true when the filter context leaves one value; ISINSCOPE is true when the column is a grouping level of the current visual row or column. **Why:** ISINSCOPE is the reliable way to tell detail rows from subtotals in a matrix, because a slicer can also leave one value.

No; each level shows the top 10 values by default. **Why:** Microsoft documents a top n of 10 per level, a maximum of 50 levels and 5,000 data points on the tree at once.

Steps after a custom native query generally don't fold (unless you enable folding with Value.NativeQuery and EnableFolding), so Power Query processes them locally. **Why:** Do the shaping in the SQL itself, or connect to a table or view and let Power Query generate folded SQL.

TOTALMTD and TOTALQTD (or CALCULATE with DATESMTD or DATESQTD). **Why:** They work like TOTALYTD but reset at the start of each month or quarter.

Merge chains of lookup tables (Product → Subcategory → Category) into one product dimension. **Why:** Each extra hop is another relationship to filter through and another table for report authors to understand. One wider dimension is usually faster and simpler.

Up to 8 on shared capacity (Pro), and up to 48 on Premium, Premium Per User or Fabric capacity. **Why:** For more frequent data, consider DirectQuery, Direct Lake, or refreshing through the XMLA endpoint or API on capacity.

Visuals then show every combination, such as every product for every day, even with no data, which bloats tables and slows reports. **Why:** Visuals hide rows where all measures are blank. Forcing zeros removes that, so only return 0 where it matters, such as a card.

A date column with unique values, no blanks and no missing dates between the first and last date. **Why:** Classic time intelligence relies on a complete, contiguous date column. Covering whole years is the usual practice, and marking the table tells Power BI which column to use.

`CALCULATE ( [Total Sales], DATESINPERIOD ( 'Date'[Date], MAX ( 'Date'[Date] ), -12, MONTH ) )` **Why:** DATESINPERIOD returns the dates from the last visible date back 12 months, so each month shows the trailing year.

Hiding only removes it from the field list; anyone with build access can still query it. OLS makes the column or table invisible and unqueryable for the roles you choose. **Why:** Hiding is for tidiness. Sensitive fields such as salary need OLS (or not being in the model at all).

Group with SUMMARIZE (or VALUES/SUMMARIZECOLUMNS) and add the calculation with ADDCOLUMNS, or use SUMMARIZECOLUMNS directly. **Why:** Calculating new columns inside SUMMARIZE can produce unexpected results because of how it builds context. The ADDCOLUMNS pattern is predictable.

`Filtered = Table.SelectRows(Source, each [Qty] > 0)` **Why:** M is case-sensitive: function names, step names and column names must match exactly.

Import for the best speed when data fits and refresh latency is acceptable; DirectQuery when data must be current or is too large to copy; Direct Lake when the data is already in OneLake Delta tables on Fabric capacity. **Why:** A good answer ties the choice to freshness, size, source and licensing, not to habit.

Rename measures in plain business language (Total Sales, Gross Margin %) and add descriptions. **Why:** Copilot and report authors rely on names and descriptions to choose the right measure; cryptic names lead to wrong answers.

Share the report directly with them, or include it in an app audience they belong to. **Why:** Item-level access avoids giving workspace rights to people who only need one report.

Framing: updating the model's references to the latest version of the Delta tables, which copies metadata only and usually takes seconds. **Why:** There is no full data copy like an Import refresh. Data preparation happens upstream in OneLake.

No. The image in the email shows data based on the subscription owner's permissions. **Why:** Microsoft warns about this for RLS models. When managers click through, the report opens with their own permissions, so keep confidential data out of the snapshot or let each manager subscribe themselves.

A policy that partitions an Import table by date, keeps older partitions, and refreshes only recent ones (for example the last 10 days of 5 years). **Why:** Refreshes become faster and lighter on the source. It needs RangeStart and RangeEnd parameters filtering a date column in Power Query.

A calculated table of values, a slicer, and a measure that returns the selected value. **Why:** You can use the selected value in other measures, for example "Sales if price rises by X%", and let users move the slider.

A Contributor can't add people to the workspace or share items onward, and can update the app only if the workspace allows it. **Why:** Contributor suits report developers who build content but shouldn't control access.

Rename the column that now exists: `Table.RenameColumns(Source, {{"Customer Name", "Customer"}})` **Why:** Steps that name columns fail when the column disappears. (MissingField.Ignore would only skip the missing "Cust Name" and leave "Customer Name" unrenamed, so later steps that use Customer would still fail.)

You add a field to a target page's Drillthrough well; users right-click a data point elsewhere and jump to that page filtered to the item. **Why:** It keeps summary pages clean while giving detailed pages for one customer or product on demand.

Reference the extract, keep the product columns, remove duplicates on the product key, and load it as its own table. **Why:** Then remove the descriptive product columns from the fact table and keep only the key, which gives a star schema.

No. **Why:** Microsoft lists recursion (and mutual recursion), overloading and explicit return types as unsupported for UDFs.

Import with a nightly scheduled refresh. **Why:** Daily latency is fine, the volume fits comfortably in memory, and Import gives the fastest visuals. Direct Lake needs the data in OneLake as Delta tables, which this source is not.

One dimension used for several meanings, like a Date table that relates to Order Date, Ship Date and Due Date. **Why:** You either keep one active and others inactive (and use USERELATIONSHIP), or load copies of the table (Order Date, Ship Date) so each can filter on its own.

Some Sales rows have a ProductKey that doesn't exist in the Product table, so Power BI groups them under a blank member. **Why:** This is a data quality signal. Fix the missing products at the source or in Power Query, or map unknown keys to an "Unknown" row.

Showing one or a few headline numbers, such as Total Sales or Margin %. **Why:** A key number at the top of the page answers the first question quickly. The newer card visual can show several values with reference labels.

A small report page set to Tooltip use, shown when users hover over a data point, filtered to that point. **Why:** It gives rich context (a mini chart, extra numbers) on hover without adding visuals to the main page.

Adding an index often breaks folding, so the filter runs locally after downloading all rows. **Why:** Put filters and column removal first while they still fold, then do steps that don't fold.

The tables, relationships, measures and settings that reports query. It used to be called a dataset. **Why:** Reports never hold data themselves. They send queries to a semantic model, so one well-built model can serve many reports.

For example MEDIAN, PERCENTILE.INC and STDEV.S (also AVERAGE, MIN, MAX and their X versions). **Why:** These let you describe a distribution, for example median order value, which is less sensitive to a few huge orders than the average.

The step lists every column by name, so a renamed or removed column makes refresh fail. **Why:** Delete auto-generated type steps and set types once, deliberately, after the shape is final.

It silently drops rows whenever any column has an error, so totals shrink without anyone noticing. **Why:** Fix errors at their cause, or replace errors in specific columns, and keep a check (for example a row count) to notice changes.

Trim (and standardise case of) the Code column in both queries before the merge. **Why:** Merges compare values exactly; "P001 " and "P001" don't match. Clean keys first.

No. **Why:** Direct Lake supports only cloud connections and can't operate through any gateway.

Yes. **Why:** The IN operator with a table constructor is a valid column predicate and is clearer than chaining OR conditions.

Both West and East. **Why:** Role permissions are combined (a union), so membership in more roles shows more rows, never fewer.

For each customer in context, get their first purchase date across all time, and count customers whose first date falls in the visible period. **Why:** For example: COUNTROWS ( FILTER ( VALUES ( Sales[CustomerKey] ), CALCULATE ( MIN ( Sales[OrderDate] ), ALL ( 'Date' ) ) >= MIN ( 'Date'[Date] ) ) ). Removing date filters is the key step.

It creates a hidden date table for every date column. It is off in most production models because it inflates model size and gives no shared calendar. **Why:** Several dates mean several hidden tables. A single date table you build is smaller and works across all facts.

Each product key appears once in Product and can appear many times in Sales. **Why:** This is the normal dimension-to-fact relationship. Filters flow from the one side (Product) to the many side (Sales).

Almost always when you compare more than a few categories. **Why:** People judge lengths more accurately than angles. A pie with eight slices is hard to read; a sorted bar chart is not.

No; Copilot can't transform or explain across multiple queries in a single input. **Why:** Microsoft lists working on one query at a time as a current limitation.

A Power BI template: the queries, model, measures and report layout without the data. **Why:** Opening a template asks for parameter values (if any) and then loads fresh data, so it is a clean way to reuse a design.

One row per event or measurement (a sale, a shipment, a reading), with numeric values and keys to the dimensions. **Why:** Facts are the tall tables you aggregate. Keeping descriptive text out of them keeps them narrow and compresses better.

BLANK() + 5 = 5, and BLANK() * 5 is blank. **Why:** In addition and subtraction blank acts like zero; in multiplication and division it stays blank. This affects measures that combine missing values.

Blank, because there is no stock row on the last calendar day. **Why:** Use LASTNONBLANK ( 'Date'[Date], CALCULATE ( SUM ( Stock[Qty] ) ) ) to take the last date that actually has data.

Keep Budget as its own fact table, relate it to Date at month level (for example to a month key or the first day of month), and compare measures by month. **Why:** Facts with different grains shouldn't be merged. Compare them through shared dimensions, and only show budget at levels it really exists.

The sales of red products. **Why:** The filter argument replaces the existing filter on Product[Colour], so every row shows the same red total. Use KEEPFILTERS to intersect instead.

Give the YoY % item a format string expression, for example "0.0%". **Why:** Calculation items can override the format string of the measure they modify, so the same column shows money and percentages correctly.

Decide who gets Copilot through tenant settings and groups, choose the capacity (or a Fabric Copilot capacity) that pays for it, monitor usage in the Capacity Metrics app, and approve only prepared models for Copilot. **Why:** Copilot consumes capacity and its answers depend on model quality, so access, cost and readiness all need owners.

"Mumbai, Pune, Delhi". **Why:** Text.Combine joins a list of text values with the separator you give.

To sign in and have write access to a workspace on a paid Fabric or Premium capacity (or a Fabric Copilot capacity assigned to you). **Why:** Desktop runs Copilot through a capacity for billing, even though the report isn't published yet.

The Date table runs to 31 March, so last year's YTD covers dates past 15 March. Limit both periods to dates up to the last date with data, for example with a "date has passed" flag in the Date table. **Why:** Comparing a partial period with a fuller one makes growth look worse. Like-for-like dates fix it.

A .pbix is one binary file holding the report, model and data. A .pbip saves the report and model as folders of text files (including TMDL for the model). **Why:** Text files can be diffed and stored in Git, which is why teams that use source control save as .pbip.

When people need pixel-perfect, printable output that can run to many pages, like invoices or long operational lists. **Why:** Power BI reports are interactive and fit a screen. Paginated reports (built in Power BI Report Builder) are designed to print or export every row.

No; you change it with prompts. **Why:** To edit freely, switch to the Custom (smart narrative) mode and paste or write the text there.

It lets the Power BI service reach data sources inside a private network, such as an on-premises SQL Server or a file share. **Why:** The service runs in Microsoft's cloud and can't connect to private networks directly. The gateway makes outbound connections to relay the data.

Canvas size (such as 16:9 or custom) and canvas background or wallpaper. **Why:** A consistent page size and background across pages makes the report feel designed rather than assembled.

It applies one slicer's selection to other pages, and can show or hide that slicer on each page. **Why:** Users pick a region once and see it everywhere, instead of re-selecting on every page.

It tells Power BI what the values mean, so maps place them correctly and URLs or images display properly. **Why:** "Paris" could be a city in France or Texas. Categories, ideally with country columns too, help visuals interpret values.

Spot problems such as unclear structure or inconsistent naming, recommend best practices, and edit measures, tables, columns and relationships. **Why:** It needs write access to the model in a capacity workspace and works in Editing mode of the model view in the service.

Not with a single-direction one-to-many relationship. **Why:** Filters flow from the one side to the many side. Slicing by attributes in a dimension table is the cleaner design anyway.

A DirectQuery partition for the newest period on top of the imported partitions, making it a hybrid table. **Why:** Users see current data for today while history stays fast in Import. It needs Premium, Premium Per User or Fabric capacity.

Copilot checks the syntax and retries once automatically; if it is still invalid, it returns the query with a note saying it contains errors. **Why:** You can rephrase the request or fix the DAX yourself. Always run and read generated queries.

No; you have to update the app. **Why:** The app shows the version published last time. Updating the app pushes your changes to its users.

Check the numbers against the report or a trusted measure, and look at how Copilot arrived at the answer. **Why:** Copilot can be wrong, especially on ambiguous models. Treat it as a fast first draft, not an audited source.

No. **Why:** Both are cloud sources the service can reach directly. A gateway is needed for sources in a private network (including Azure sources behind a private network, which may need a VNet data gateway).

No. **Why:** Copilot in Fabric and Power BI only runs on paid SKUs. Trial capacities are explicitly excluded.

`Title = "Sales for " & SELECTEDVALUE ( Region[Region], "all regions" )` **Why:** Read the column the slicer filters. The fallback text also covers the case where several regions (or none) are selected.

`CALCULATE ( [Total Sales], Sales[Region] = "West" )` **Why:** The FILTER version iterates the whole fact table and puts all its columns into the filter. A column predicate filters only one column and is faster and clearer.

10. **Why:** PREVIOUS returns the value from the previous row on the axis, so February shows 20 - 10. January has no previous row.

The semantic model's refresh history in the service, which shows status, duration and error details for each run. **Why:** The error message usually points to credentials, gateway availability, a changed column or a timeout.

1. **Why:** Only the product priced 10 is cheaper than 20. The variable captures the current row's price before FILTER starts its own row context.

A one-column table named Date with 365 rows, one for each day of 2025. **Why:** CALENDAR returns every date between the start and end dates, inclusive, in a column called Date.

When you test several conditions in order, such as margin bands. **Why:** SWITCH ( TRUE (), [Margin] < 0, "Loss", [Margin] < 0.1, "Low", "Healthy" ) reads top to bottom and is easier to maintain.

Write a DAX query from a plain-language request, modify a selected query, explain a query, and explain DAX functions or topics. **Why:** You can run the suggested query and keep or discard it. The shortcut to open Copilot there is Ctrl+I.

A report has one or more interactive pages built on one semantic model. A dashboard is a single service-only canvas of pinned tiles, which can come from many reports. **Why:** Dashboards are a summary for monitoring and alerts; reports are for exploring. Dashboards are created in the service, not in Desktop.

false. **Why:** Text comparison in M is case-sensitive by default. Use Comparer.OrdinalIgnoreCase or convert both to the same case to compare loosely.

It creates a public link or embed code that anyone on the internet can view without signing in. Never use it for confidential or internal data. **Why:** There is no authentication or row-level security. Admins can disable it in tenant settings.

It lets you choose which tables, columns and measures Copilot should focus on, so it ignores helper or confusing fields. **Why:** A smaller, relevant schema reduces ambiguity, for example between three different "Amount" columns.

You pin visuals (or a whole live report page) from reports, or add tiles such as text, images or web content. **Why:** A dashboard can collect key visuals from several reports and semantic models on one screen.

`each "Order " & Text.From([OrderID])` **Why:** The & operator joins text. Convert the number to text first with Text.From.

Build permission on the semantic model (directly or through the app), without a workspace role. **Why:** They can connect to the model from Desktop or the service and create reports elsewhere, while the model stays protected.

A list: {"2025", "03", "07"}. **Why:** Text.Split returns a list of text pieces. Use List functions (for example {0} to take the first item) to get one part.

10. **Why:** The query returns whatever step is named after in. Step c is computed only if something needs it; here nothing does.

The preview shows the data as it was after that step. **Why:** It is the quickest way to find where a problem started. New steps are inserted after the selected step, which can break later steps that expect the old shape.

"Unknown". **Why:** ?? is M's coalesce operator: it returns the left value unless it is null, otherwise the right value.

No; classic functions have no week-level variants. Calendar-based time intelligence adds TOTALWTD. **Why:** Week-based calendars were a long-standing gap. With a calendar that defines week categories, TOTALWTD ( [Sales], 'Calendar' ) works.

Options that look across the remaining fields and drill into the one giving the highest (or lowest) value of the measure. **Why:** AI levels show a light bulb and recalculate when you select a different node or cross-filter the tree.

A visual an author marks as the answer to certain questions, with trigger phrases, so Copilot returns that visual when users ask those questions. **Why:** It gives consistent, approved answers to common questions like "sales this quarter". Verified answers are saved on the semantic model.

A dimension shared by several fact tables, such as one Date or Product table used by Sales, Returns and Budget. **Why:** One slicer on the shared dimension filters every fact consistently, so numbers line up across visuals.

Use the Copilot report pane in Power BI Desktop, choosing skills with the skill picker, and close and reopen the pane after each change. **Why:** Changes save on the model; reopening the pane makes Copilot pick them up. "How Copilot arrived at this" and Download diagnostics help troubleshoot.

Organisation-wide switches that admins set, such as allowing Publish to web, export, Copilot and who can create workspaces. **Why:** Many "why can't I do X?" questions are answered by a tenant setting, often scoped to security groups.

Smaller deviations are flagged as anomalies; lowering it makes the algorithm more selective. **Why:** Tune sensitivity so users see the anomalies that matter, not every wiggle.

The model's metadata (tables, columns, measures and the first 200 characters of descriptions), and it may use a reduced view of very large models. **Why:** It can't read your table data (apart from min/max samples in Import mode), so descriptive names and descriptions matter.

Pass the year-end date to the function, for example TOTALYTD([Sales], 'Date'[Date], "3-31"). **Why:** TOTALYTD and DATESYTD accept a year_end_date argument. A fiscal year column in the date table also helps with labels and slicers.

Left outer, Right outer, Full outer, Inner, Left anti and Right anti. **Why:** Anti joins return only rows with no match, which is a quick way to find orphan keys or missing products.

The product's rank by sales among all products (1 = highest). **Why:** RANKX evaluates the measure for every row of the table in its first argument and finds where the current value fits.

It writes a text summary of the whole report, chosen pages or chosen visuals, which you can steer with suggested or custom prompts. **Why:** It gives readers a written takeaway with footnotes that point back to the visuals it used.

Create or edit report pages, summarise the semantic model, add a narrative visual, write DAX queries and suggest measure descriptions. **Why:** Some run in the Copilot pane and others inline, such as in DAX query view or the model view's description box.

Hide keys and technical columns, group measures into display folders, rename fields in business terms and add descriptions. **Why:** A smaller, well-named field list speeds up authoring and reduces mistakes like slicing by the wrong column.

A set of tools (in preview) for model owners to improve Copilot answers: an AI data schema, verified answers and AI instructions. **Why:** Copilot can misread an unprepared model. These tools tell it which fields matter, what terms mean and which visuals answer common questions.

Classic time intelligence expects a complete date column; the fact table has gaps on days with no sales. **Why:** Use the date table's column ('Date'[Date]) with a relationship to Sales, so shifts land on real calendar dates.

COUNT counts non-blank values in a column (numbers, dates or text), COUNTA counts non-blank values of any type, COUNTROWS counts rows of a table, and DISTINCTCOUNT counts unique values. **Why:** For "number of orders" use COUNTROWS on the fact; for "number of customers who bought" use DISTINCTCOUNT on the customer key.

Generate transformation steps or new queries from a plain-language request, summarise a query and its steps, and undo the last applied step. **Why:** Microsoft documents these features with limits: Copilot works on one query at a time and can be inaccurate when the data it needs isn't in the sample.

Trim (remove leading and trailing spaces), Clean (remove non-printing characters) and Capitalize Each Word or Uppercase. **Why:** "Delhi", "delhi " and "DELHI" look different to a relationship or slicer. Standardise them before loading.

Find the cause by clicking the error, fix the values before the type step, or use Replace errors or Remove errors as a last resort. **Why:** Removing error rows can silently drop real data. Understanding the cause (often text like "N/A" in a number column) is better.

To show how a starting value moves to an ending value through positive and negative steps, like a profit bridge. **Why:** Each bar starts where the last one ended, so viewers see which items added or removed the most.

The formula language behind Power Query. Every applied step is an M expression, visible in the Advanced Editor. **Why:** Most work is done through the ribbon, but reading M helps you debug, reuse and fine-tune queries.

About 5.8%. **Why:** The multiplier compares the influencer's rate with the average of the other values: 14.93% ÷ 2.57 ≈ 5.8%.

Relate both to shared dimensions (Date, Product, Customer) and compare them with measures. **Why:** Fact-to-fact relationships are many-to-many and confusing. Shared (conformed) dimensions let one slicer filter both facts correctly.

Desktop is the free Windows app where you connect, shape, model and design reports. The service (app.powerbi.com) is where you publish, share, refresh and govern them. **Why:** Most authoring starts in Desktop, while collaboration, scheduled refresh, apps and security settings live in the service.

The single, tenant-wide data lake in Fabric where items such as lakehouses and warehouses store their data, in Delta format for tables. **Why:** Because everything sits in one lake, Direct Lake models and other engines can read the same tables without copying them.

Whatever measure the visual is currently evaluating. **Why:** A YTD item written as CALCULATE ( SELECTEDMEASURE (), DATESYTD ( 'Date'[Date] ) ) works for Sales, Cost or Units.

A table of the five products with the highest sales (more if there are ties at fifth place). **Why:** TOPN returns a table, so use it in CALCULATE or iterators, for example to compute "Sales of the top 5 products".

A shared container where a team keeps related items such as semantic models, reports, dashboards and dataflows. **Why:** Workspaces are where collaboration, roles, capacity assignment and app publishing are managed.

A Fabric capacity (F2 or P1 at least) that an admin designates to collect and bill the Copilot usage of assigned users, wherever their content lives. **Why:** It lets people use Copilot in Power BI Desktop and in Pro or PPU workspaces, and brings Copilot costs into one capacity for monitoring.

It joins two queries on matching columns, like a SQL join, and adds the matched rows as a nested table column you can expand. **Why:** Use it to bring a lookup value into a table, or to find rows that do or don't match.

Rarely: for example on a bridge table in a many-to-many pattern, or for a security table in dynamic RLS. Otherwise prefer single direction and CROSSFILTER in specific measures. **Why:** Interviewers want to hear that you understand ambiguity and performance risks and treat "Both" as a deliberate exception.

Set the title text with fx to a measure, for example "Sales for " & SELECTEDVALUE(Region[Region], "all regions"). **Why:** Titles bound to measures tell readers exactly what they are looking at after they filter.

Inside CALCULATE, it activates an inactive relationship for that calculation only. **Why:** `CALCULATE ( [Total Sales], USERELATIONSHIP ( Sales[ShipDate], 'Date'[Date] ) )` gives sales by ship date while the order date relationship stays active.

For example Data Factory, Data Engineering, Data Warehouse, Data Science, Real-Time Intelligence and Databases. **Why:** All of them run on the same Fabric capacity and store data in OneLake, which is what makes Fabric a single platform.

It evaluates an expression in a filter context changed by its filter arguments. **Why:** Almost every "sales for X", "ignoring Y" or "compared with Z" measure is a CALCULATE with filters that add, replace or remove context.

Decomposition tree for exploring how a measure breaks down across chosen dimensions step by step; key influencers for ranking which factors statistically drive an outcome. **Why:** One answers "where is it?" by drilling; the other answers "what drives it?" with regression.

Wrap it: `IF ( ISINSCOPE ( Product[Product] ), [My Measure] )` **Why:** ISINSCOPE is true only on rows grouped by Product, so subtotals and totals return blank.

Clean up names and descriptions, prepare the main models for AI, test common questions, then mark ready models Approved for Copilot. **Why:** Readers judge Copilot by its first answers. Preparation before rollout avoids generic or wrong results that hurt trust.

`Rank = RANKX ( ALL ( Product[Product] ), [Total Sales] )` **Why:** Each row filters Product to one item. ALL on the product column ranks against every product.

A security table mapping each rep's email to their territories, a relationship into the model, one dynamic role using USERPRINCIPALNAME(), testing with View as, and reps given Viewer or app access. **Why:** Dynamic RLS scales without a role per rep, and the answer shows you know RLS doesn't apply to editors.

Fixed decimal (Currency) stores four decimal places exactly, which avoids floating-point rounding surprises in money totals. **Why:** Decimal number is a floating-point type. For money, Fixed decimal gives exact sums and often compresses better.

No; classic data alerts are set on dashboard tiles showing a single number (card, KPI, gauge). **Why:** For report visuals, pin a card to a dashboard and set the alert there, or use other tools such as Fabric Activator.

Name measures in business language without table prefixes, keep them in a dedicated measure table or display folders, and add descriptions. **Why:** Report authors and Copilot both read measure names and descriptions, so clear naming reduces mistakes.

Breaking a measure down across dimensions in any order, drilling level by level, for ad hoc root-cause analysis. **Why:** Users choose the next dimension themselves, or let AI splits pick it.

When many measures need the same variants (YTD, PY, YoY). The trade-off is that implicit measures stop working and the logic is less visible to new authors. **Why:** Calculation groups cut measure count dramatically, but the team must understand SELECTEDMEASURE and precedence.

In the report's settings, allow only summarised data export (or none), and check the related tenant settings with the admin. **Why:** Export options exist at report and tenant level; the stricter one wins.

No; keep the customers and leave the email blank (or "Not provided"). **Why:** Removing rows would drop their sales from the model. Blank attributes are normal; missing keys are the real problem.

Workspace roles, apps, direct sharing of an item (or a link), and embedding (Teams, SharePoint, a secure portal or your own app). **Why:** Each suits a different audience. Apps are usually best for large read-only audiences; workspace roles for the team building content.

Is the workspace on paid F2+/P1+ capacity (or is the user on a Fabric Copilot capacity)? Is the Copilot tenant setting on for them? Is the capacity in a supported region, and is cross-geo processing allowed if needed? **Why:** Missing Copilot is nearly always capacity, tenant settings or region. Microsoft also notes new capacity can take up to 24 hours to be recognised.

1,000. **Why:** ALL ( Region ) removes the row's region filter, so every row shows the grand total.

It removes filters coming from inside the visual (rows, columns) but keeps filters from outside it, such as slicers. **Why:** Use it for "share of the visible total" when a slicer has narrowed the data, so percentages still add up to 100% on screen.

No. Refresh fails and the model can't be queried until the Delta tables are brought within the limits. **Why:** Only Direct Lake on SQL falls back (if allowed). Microsoft's guardrail table gives, for example, 300 million rows per table up to F32 and 1.5 billion on F64.

You change one value to point all queries at test or production, and you can edit parameter values in the service's semantic model settings. **Why:** Parameters also drive deployment pipeline rules and incremental refresh (RangeStart and RangeEnd).

A CALCULATE modifier that clears filters from a table or columns. It works like ALL used as a filter argument, with a clearer name. **Why:** Many teams prefer REMOVEFILTERS because readers see the intent immediately. Unlike ALL, it cannot be used as a table expression.

Select the ID columns and choose Unpivot other columns. **Why:** "Unpivot other columns" names only the columns to keep, so new month columns are unpivoted automatically.

A measure reference is always wrapped in an implicit CALCULATE, so it triggers context transition; the bare SUM does not. **Why:** This is the core reason "a measure is not the same as its formula" inside iterators and calculated columns.

They apply the same logic (YTD, PY, YoY %) to any measure, so you don't write one variant of every measure. **Why:** Ten base measures × five time variants would mean fifty measures. A calculation group with five items does it once.

A report that has no model of its own and queries a semantic model already published in the service. **Why:** Many reports can share one certified model. You can add report-level measures, but not tables or Power Query steps, unless you turn it into a composite model.

Two rows: East 15 and West 7. **Why:** Table.Group passes each group's rows to the function; [Sales] is that group's list of sales, which List.Sum adds.

The factors that most affect a metric (on the Key influencers tab) and combinations of factors, called segments (on the Top segments tab). **Why:** It answers "what drives this?", for example what makes customers give low ratings.

The "current row" that exists while DAX iterates a table, as in a calculated column or inside SUMX. **Why:** Row context lets you write Sales[Qty] * Sales[Price] for one row. It doesn't filter other tables by itself.

For tenants or capacities outside the US or EU data boundary, Copilot is disabled by default. The admin must enable the setting that allows data sent to Azure OpenAI to be processed outside the tenant's geographic region. **Why:** It is a compliance choice for the organisation, so Microsoft leaves it off until an admin opts in.

False. **Why:** Visual calculations are stored on the visual that defines them. If several visuals need the logic, write a measure (or a DAX user-defined function) instead.

With a Year slicer on 2025 and a product slicer on three products, ALL ( Product ) in a denominator gives every product's 2025 sales; ALLSELECTED ( Product ) gives the total of the three selected products. **Why:** The difference shows when users filter products. ALLSELECTED keeps percentages adding to 100% of what they selected.

It wastes capacity and source resources and can interfere with other refreshes, with no fresher data. **Why:** Match the schedule to how often the source changes; one refresh after the nightly load is enough.

Returns [Sales], or 0 when [Sales] is blank. **Why:** COALESCE returns the first argument that isn't blank. It is shorter than IF ( ISBLANK ( ... ) ).

Profile it, find the cause with the data owner, fix what you can upstream, apply clear documented rules in Power Query, and keep checks (like row counts or an error table) so new problems show up. **Why:** Interviewers look for a process, not just "remove errors", because silent fixes hide problems.

They help on charts with few points where exact values matter; they hurt on dense charts where labels overlap. **Why:** Too many labels turn a chart into a table of numbers. Use tooltips for exact values on busy charts.

Power BI pauses it and emails the semantic model owner. **Why:** After two months without visits to any report or dashboard on the model, the service treats it as inactive. A subscription on the report keeps the refresh from being paused.

In the report settings (export data options) and the tenant settings the admin manages. **Why:** Export of underlying data can leak detail, so many organisations allow summarised exports only or none.

The same value as [Total Sales] in the current context; ALL has no effect. **Why:** The variable was already evaluated before CALCULATE runs. CALCULATE can't change a value that is already computed; put the measure itself inside CALCULATE.

`Sales PY = CALCULATE ( [Total Sales], SAMEPERIODLASTYEAR ( 'Date'[Date] ) )` **Why:** SAMEPERIODLASTYEAR shifts the dates in context back one year, so it works at day, month, quarter or year level.

"POWER". **Why:** Text.Start takes the first five characters and Text.Upper makes them capitals.

For your organisation, users sign in with their own Power BI accounts (user owns data). For customers, your app authenticates with a service principal and users don't need Power BI licences (app owns data). **Why:** The second needs embedding capacity, such as a Fabric F SKU, and your app must enforce security itself (for example by passing RLS identities).

It sets the level at which a measure or summarised field is analysed, such as per house ID, without treating that field as an influencer. **Why:** Measures are otherwise analysed at the level of the Explain by fields, which can be too summarised to find patterns.

It evaluates the expression at the parent level of the axis, so this returns each row's share of its parent. **Why:** In a matrix of Category > Product, each product is divided by its category total, giving "percent of parent" in one line.

Invite them as Microsoft Entra B2B guests and share through an app or item with row-level security for their company, if your tenant settings allow guest access. **Why:** Guests sign in with their own accounts and see only what RLS allows. Publish to web would expose the data publicly.

`[Region] = "West"` **Why:** The trailing space makes the text different. Better still, use a security table so region names aren't typed into roles.

Yes; by default profiling only looks at the first 1,000 rows. **Why:** Switch the status bar setting to "Column profiling based on entire data set" to check every row, which is slower.

The query stays in Power Query for others to reference, but its result is not loaded into the model. **Why:** Staging and helper queries shouldn't appear as tables in the model; disabling load keeps the model clean and smaller.

People in your organisation, people with existing access, or specific people; you can also choose whether they can share or build on the data. **Why:** Organisation-wide links are convenient but broad. Specific people is the controlled option for sensitive reports.

CALENDAR takes the start and end dates you give it. CALENDARAUTO scans the model's date columns and builds whole years around them. **Why:** CALENDARAUTO is convenient but can pick up odd dates such as birth dates or 1900 placeholders. CALENDAR keeps the range under your control.

It lets users without access find the model in the catalog and request access. **Why:** It helps people reuse an existing certified model instead of building another copy.

A plain title such as "Sales by category", and measures named the way business users talk. **Why:** Default titles come from field names. Clear names make reports easier to read and also help Copilot understand the model.

Right-click the step: if View Native Query is available it folds; the step folding indicators also show this. **Why:** Steps after the first non-folding step usually run locally, so keep foldable steps (filters, column selection) early.

Templates in Power BI Desktop that write DAX for common calculations, such as running totals or year-over-year change, from fields you pick. **Why:** They are a quick start and a way to learn patterns. Always read the generated DAX and rename the measure to something meaningful.