Managing Credentials, Privacy Levels, and Data Source Settings
A sales report refreshes successfully, its charts contain plausible numbers, and nobody sees an error. Is that enough evidence that its credentials, data source settings, and privacy levels are correct? Not necessarily. A successful refresh tells you that one attempt worked under one set of conditions. It doesn’t prove that the right identity was used, that each source was classified appropriately, or that combining those sources respects the organization’s data-handling requirements.
That distinction matters in the PL-300 scope. The course blueprint explicitly includes changing data source settings, including credentials and privacy levels. This isn’t just a request to remember where a dialog box appears. You need to understand what each setting controls, choose it from the business requirement, and confirm the resulting behavior.
Imagine that Harbor Retail has a Power BI sales solution. Power Query connects to an internal sales database, a restricted workbook containing customer exceptions, and a public source of currency rates. The analyst needs valid access to each source, and Power Query must combine some of their data during refresh. Before changing anything, the analyst separates three related questions.
How does Power BI locate and connect to a source? Which identity or authentication information allows access? How should Power Query isolate that source when combining its data with another source?
Those questions correspond to data source settings, credentials, and privacy levels. They’re related because all three can affect whether data acquisition succeeds, but they aren’t interchangeable. Credentials establish access to a source.
A privacy level describes how Power Query should treat the source when data from multiple sources might be combined. Data source settings are where source-related connection and access choices are managed. The transformations that clean, filter, join, or reshape the data belong to Power Query logic, often expressed in the M formula language.
Suppose an analyst removes columns and filters the sales table to the current year. Those are transformations. Editing the filter won’t repair invalid credentials. Replacing credentials won’t automatically correct an inappropriate privacy classification.
Changing a report visual won’t solve either problem because the visual is downstream from data acquisition. The first troubleshooting decision is therefore not which button looks relevant. It’s which layer owns the failed behavior.
Privacy levels matter when Power Query combines data from multiple sources. Their purpose is to isolate sources and reduce unintended data transfer. Power Query recognizes the classifications Private, Organizational, Public, and None. Private represents a source requiring strong isolation.
Organizational represents a source intended for use within a trusted organizational boundary. Public represents information that may be shared broadly. None means that no privacy classification has been designated. None shouldn’t be casually interpreted as another name for Public or as evidence that the data is safe to expose.
Return to Harbor Retail. The internal sales database is governed company data, while the currency-rate source is genuinely public. Classifying the sales source as Organizational and the rate source as Public expresses a meaningful boundary.
Power Query can evaluate that classification while processing the query that combines sales amounts with exchange rates. The analyst still needs valid credentials for the sales database. Public data and organizational data having compatible privacy treatment doesn’t grant database access by itself.
Now consider the customer-exceptions workbook. It contains restricted information and has been classified as Private. A query attempts to use values from that workbook while retrieving or combining data from the organizational sales source.
Power Query may block the combination because privacy settings can prevent movement between sources when the requested operation would violate their isolation. That failure is not merely an inconvenience between the analyst and a finished report. It’s a protection signal indicating that the planned combination deserves scrutiny.
The quickest-looking response might be to weaken the privacy setting or bypass isolation so the refresh completes. That confuses the disappearance of an error with proof of a correct solution. A classification should represent the source’s actual sensitivity and sharing boundary. If the workbook really requires private isolation, labeling it Public to make a query work would misrepresent the requirement. Unsafe combination options such as indiscriminate Fast Combine usage shouldn’t replace an informed design decision.
A better response begins with the intended information flow. Does the sales calculation truly need restricted customer details, or only an approved customer category? Could the required, nonrestricted field be supplied through a governed organizational source instead? Should the combination be redesigned so private values aren’t transferred into another source’s processing context?
The exact architecture depends on Harbor Retail’s requirements and governance. The durable rule is that the requirement determines the classification and architecture. The desired absence of an error does not.
This also explains why there’s no universally correct privacy level for every Power BI solution. Private isn’t automatically the best answer simply because it sounds safest. Public isn’t correct simply because it permits broad combination.
Organizational isn’t a convenient default for every business-owned file. Each classification expresses a different boundary. Choosing one without understanding the source can either allow an unwanted transfer or prevent a legitimate operation.
Credentials require the same requirement-driven thinking. An authentication problem and a privacy problem can both appear during data acquisition, yet their causes differ. If Harbor Retail’s sales database rejects the supplied identity, the analyst should investigate the credential and access state for that source. Changing the workbook’s privacy level doesn’t make the rejected identity valid. If every source accepts its credentials but Power Query blocks their combination, repeatedly re-entering passwords doesn’t address the isolation decision.
The observable symptom helps identify which question to ask, but it isn’t a substitute for controlled testing. The analyst first reproduces the failure under known conditions and records what was expected. Perhaps the expected result is a completed refresh containing sales converted with public exchange rates. The actual result is a credentials prompt for the sales database.
That points toward source access. In another test, each source can be accessed separately, but the combined query reports a privacy-related failure. That points toward source classification or the architecture of the combination.
Once the likely layer is known, the analyst changes the setting that directly governs it. There’s no reason to modify a DAX measure when Power Query can’t access a source. DAX calculations operate on data available to the semantic model. They don’t establish source credentials or redefine Power Query’s privacy boundaries. Likewise, changing a chart interaction might alter what a report consumer sees, but it won’t repair a failed refresh that prevented the required rows from reaching the model.
The separation between settings and transformations is useful during maintenance too. Suppose the location or access requirements of Harbor Retail’s sales source change, while the business shaping rules remain valid. The analyst can review the source settings and credentials without pretending that the filter, rename, and type-conversion steps are the cause. Conversely, if authentication succeeds but a transformation produces incorrect rows, replacing credentials would be irrelevant. Clear ownership narrows the investigation and reduces unrelated changes.
After correcting credentials, the analyst doesn’t stop when a connection test or settings dialog accepts them. The query must actually retrieve the expected data under the identity that will perform the relevant operation. The analyst refreshes the affected query and compares the outcome with the requirement. Did access succeed?
Were the expected rows available? Did another source fail? A saved credential is configuration evidence. Successful, expected data acquisition is behavioral evidence.
Privacy validation also needs more than a saved selection. For the sales and currency-rate query, the analyst refreshes the combined result and checks whether the expected converted sales records appear. The analyst then tests the restricted workbook path separately. If that combination is supposed to remain isolated, a protective block may be the correct outcome. If an approved redesign supplies a permitted organizational field instead, the revised query should succeed without falsely reclassifying the private source.
Notice that success and failure aren’t inherently good or bad. Their meaning depends on the requirement. A refresh failure caused by invalid credentials is undesirable when the authorized process should have access.
A blocked combination can be desirable when it prevents private information from crossing a boundary. A completed refresh can be undesirable if it was achieved by assigning a source a classification that doesn’t match its sensitivity. Validation compares observed behavior with the intended rule, not with a general preference for green status indicators.
You also need to test the environment and condition in which the solution will actually be used. A query working during one analyst’s normal case doesn’t prove that every required refresh or access condition is valid. The relevant identity may differ, a credential may no longer authorize the source, or an alternate query path may combine sources that the first test never touched. The appropriate evidence is the query output, refresh and access state, and resulting consumer experience for the condition under examination.
Suppose Harbor Retail’s main sales refresh works because it uses only the organizational database and the public currency source. A less common monthly process also brings in the private exceptions workbook. Testing only the daily path would miss the privacy conflict. The analyst deliberately reproduces the monthly condition, notes the expected result, and observes where processing stops. That turns a vague complaint that the report sometimes fails into a specific finding about one source combination.
The correction must then preserve the normal case as well as the edge case. If the approved solution moves a nonrestricted exception category into a governed organizational source, the analyst retests the monthly process. The necessary category should appear, the private details should remain outside the combination, and the daily sales path should still refresh as expected. Fixing the monthly failure while breaking ordinary sales refresh would not satisfy the business requirement.
The same discipline prevents an analyst from making several changes at once. If credentials, privacy classification, query logic, and report filters are all modified together, a later successful result won’t reveal which change repaired the problem. Worse, one change may hide a second defect. A controlled investigation captures the original behavior, changes the setting or architecture most directly connected to it, and repeats the same test. Afterward, it checks a nearby condition for side effects.
Consider a second symptom. The report refresh finishes, but the model contains no converted sales for one region. It would be tempting to conclude that the currency source’s credentials failed. Yet the public source might have loaded correctly.
The missing records could instead come from a transformation that filtered the region, a mismatch in the query combination, or another downstream modeling decision. Credential and privacy settings are candidates only when the evidence connects the symptom to source access or source isolation. Familiarity with a setting doesn’t make it the cause of every data problem.
That’s why layer identification comes before correction. Power Query owns acquisition and refresh-time transformation. The semantic model organizes the data delivered to it. DAX calculations evaluate within that model.
Report interactions determine how consumers explore the result. Service access and refresh conditions govern what happens in the operating environment. This episode’s settings belong with source acquisition and combination, even though their consequences can eventually appear as an empty model, an incorrect calculation, or an incomplete report.
When a downstream symptom appears, trace it backward without assuming the first visible layer is responsible. If a visual is empty, determine whether the expected rows reached the query output. If they didn’t, inspect the relevant source access, privacy behavior, and transformation path. If the rows did arrive correctly, credentials are less likely to explain the empty visual, and the investigation can move into the model or report. This avoids changing source security settings to compensate for a downstream design issue.
Secure handling extends beyond choosing a label. Connections should use appropriate protection, parameters should be handled carefully, and identities should receive deliberate access treatment. Those practices support the same principle as privacy classification. A solution should obtain the required data through an approved path without exposing more information or access than the requirement justifies. Convenience isn’t a reliable security boundary.
For an exam-style requirement, listen for what must change and what evidence would demonstrate it. If a prompt says an authorized refresh can no longer reach a source, credentials and access state deserve attention. If it describes combining a restricted source with another source, privacy isolation is central. If it says the source moved or its connection information changed, review data source settings. If the complaint is merely that a chart total is wrong, don’t leap to credentials without evidence that data acquisition failed.
Then ask what the nearest incorrect choice would actually do. Re-entering credentials may restore authentication, but it won’t authorize unsafe information flow between sources. Changing a privacy level may alter whether Power Query permits a combination, but it won’t correct an expired or unauthorized identity. Editing M transformations may change which rows and columns are processed, but that doesn’t redefine who may access the source. Adjusting DAX may change a model calculation, but it won’t make a blocked Power Query combination complete.
Finally, define acceptance evidence before declaring the work finished. For Harbor Retail, that evidence could include successful access to the organizational sales source using the intended authorized identity, successful combination with genuinely public currency data, continued isolation of restricted private details, and correct query output for both daily and monthly conditions. The analyst also checks that the downstream semantic model and report receive the expected data. Those observations tie configuration to business behavior.
The original report that simply refreshed without an error offered only a narrow fact. A defensible result is richer. You can state which source settings are in effect, why each source has its classification, which identity is expected to access it, how Power Query treats the combination, and what test confirms the outcome. That explanation shows that the configuration works for the intended reason rather than by accident.
Credentials answer whether an identity can access a source. Privacy levels answer how Power Query should isolate that source when data is combined. Data source settings manage those source-related choices, while Power Query transformations shape the data after access and isolation requirements are considered. Keeping those responsibilities separate makes troubleshooting much more precise.
When a refresh fails, identify whether the evidence points to authentication, source classification, or query logic instead of changing whichever setting is most familiar. Then test the behavior that the requirement actually depends on. The right result might be a successful refresh, or it might be a protective block that prevents an unsafe combination. What matters is that the observed result matches the intended access and information boundary.
Explain why credentials, source settings, and privacy levels are separate decisions. What refresh or combination behavior would you test after changing one of them? For the exact feature names and practical validation steps, continue with the companion episode reference. This lesson was produced by Bare Metal Cyber dot com.