Bureau of Data Source Inspection
Semantic Model Admissions Division
FORM PBI‑1 · REV. 2026
shouldiusethisdatasourceforpowerbi.vip
VIP CLEARANCE ON FILE

Before a data source is permitted anywhere near a semantic model, it is required to pass inspection. Most do not pass. Most were never going to pass. Read on, or scroll down to the intake form and find out for yourself.

Official Notice · Case File No. 2026‑VIP‑041

On the matter of what you are about to plug into your semantic model

This bureau was established after the eleventh consecutive incident in which a “temporary” data source outlived the analyst who built it, the manager who requested it, and in one memorable case, the department itself. We do not do this for fun. We do this because someone has to, and the DAX measures certainly are not going to defend themselves.

Excel workbooks and CSVs on a desktop, a personal OneDrive, or a mapped network drive are our most frequent offenders. They arrive under names like Q3_Final_v2_USE_THIS_ONE.xlsx, they contain at least one merged cell load-bearing enough to collapse the Power Query editor, and their refresh schedule is, in practice, whatever time the owner's laptop happens to be open and signed in. The network drive is no better: it is the same single file, now additionally dependent on a drive letter someone mapped in 2018, a VPN connection, and the continued goodwill of a file server nobody has renamed since. Columns get reordered on a whim. Nobody tells you. Your model finds out at 6 a.m.

SharePoint lists are the more respectable-looking offender, which makes them worse. Choice columns get renamed mid-quarter and every downstream measure quietly starts returning blank. Permissions inherit from a folder nobody remembers configuring. And once a list crosses the 5,000‑item view threshold, you are no longer building a report — you are negotiating with SharePoint's mercy.

None of these is a data source. All of them are a rumor with a refresh button. A semantic model built on a star schema, sound relationships, and row-level security deserves inputs that were designed on purpose, by someone, at some point, with the intent of being read by more than one person. For a full accounting of known offenders, consult the Official Ledger below. If your particular data source is not listed, that is not an endorsement — please proceed directly to the Field Intake Form.

Per Bureau policy 4.2(c): ad‑hoc use of a spreadsheet for a one‑time lookup is fine. We will not speak of it again. The moment it gets a scheduled refresh, it is our problem.
Official Ledger of Prior Offenses

Known offenders, on the record

Every entry below has been filed against a real semantic model, by a real inspector, who is still not over it. Classification reflects the Bureau's default finding; exceptions are handled case-by-case in the Field Intake Form below.

Ledger of known Power BI data source offenders
Exhibit Data source Also known as Prior offenses Classification
Exhibit A Excel Workbook or CSVRepeat offender Desktop, personal OneDrive, or a mapped network drive (the Q: drive, allegedly) 14,000+, undercounted Rejected
Exhibit B SharePoint ListRepeat offender “the master tracker”; the list Karen maintains 5,000, the view threshold, allegedly Rejected
Exhibit C DirectQuery to Production OLTP “it's basically real-time” 1, per incident report, so far Rejected
Exhibit D Access Database database.mdb, database_OLD.mdb, database_OLD_OLD.mdb unknown, last audited never Rejected
Exhibit E Power Automate Flow, Origin Unknown “don't touch it, it still runs” undetermined, terrifyingly Quarantined
Exhibit F Screen-Scraped PDF Report “OCR'd, allegedly” every 8 that used to be a B Quarantined
Exhibit G Certified Dataflow, Owned by the Data Team “the one good one” 0 Approved
*Prior-offense counts are self-reported by whichever inspector was on duty and should not be considered statistically rigorous.
Exhibit H · Filed as Supplementary Evidence

A Chronicle of the Semantic Model’s Founding

As told by the Bureau’s retained bard

Now hearken, and I shall tell of a founding not easily won.

Long before the semantic model stood, there was only rumor and wreckage: a workbook called Final_v2_USE_THIS_ONE, passed from hand to hand till none recalled who built it; a SharePoint list whose columns changed name with the seasons; a network drive that answered to the letter Q only when the VPN allowed it.

The stewards of technology had seen such ruins before — every workbook that outlived its maker, every database renamed OLD and then OLD again, every flow still running long after its maker had gone. When word came that a new model should be built upon these same foundations, the stewards convened, weighed the source against the old standard, and found it wanting.

So the stewards of technology handed down their judgment: no kingdom is built on borrowed spreadsheets, and no source shall enter the semantic model’s keep until it can answer plainly who owns it, who refreshes it, and who will remain to explain it when it breaks.

Field Intake Form · Complete Before Proceeding

Should you use this data source?

Answer honestly. We will know if you don't.