07 · Governance & Data Source Certification

Once multiple teams publish workbooks against Orders-like data, governance answers: which data source is the trustworthy one, who can change it, and how do consumers know. This module works through certification and governance using the published data source from Level 2 Module 9/10.

1. The problem certification solves

1. Suppose three analysts each publish their own extract of Orders to Tableau Server: Analyst A's copy has the full 8 rows (6880 total sales), Analyst B's is missing the two Office Supplies rows (accidentally filtered, totaling 6880 − 60 − 120 = 6700), and Analyst C's has a stale extract from before order 1008 was added (6880 − 1100 = 5780).
2. Without governance, a viewer searching Server for "Orders" finds three data sources with three different Sales totals and no way to tell which is authoritative — a governance failure that produces silently wrong downstream reports.

2. Certifying a data source

1. A site admin or designated certifier marks one published data source (Analyst A's complete, correct 6880-total version) as Certified — it then shows a badge in the Server/Cloud catalog, and Tableau's connection dialog recommends it above uncertified sources with a similar name.
2. Certification doesn't technically prevent someone from still using Analyst B's or C's uncertified copies — it's a discoverability and trust signal, not an access control. Pair certification with actually removing or archiving the incorrect duplicates once discovered, since a badge alone doesn't fix the 6700 and 5780 totals still sitting on Server.

3. Data quality warnings

1. Independent of certification, any data source can carry a data quality warning — e.g. "Warning: this extract's last successful refresh was 2024-01-10; the source table has since added rows through 1008" — shown to every viewer who opens a workbook built on it.
2. Verify the warning's value: if Analyst C's stale source (5780 total, missing order 1008's 1100 in Sales) is flagged with a warning, a viewer who sees "6880" on a certified dashboard and "5780" with a warning banner on Analyst C's has enough information to know which number to trust and why they differ (5780 + 1100 = 6880 — exactly the missing order).

4. Managed metadata: lineage and impact analysis

1. Tableau's Data Catalog / lineage view traces a field like Sales backward to its source column and forward to every workbook/view that uses it — for the certified Orders source, lineage would show every downstream dashboard built on it (e.g. the Level 2 Module 9/10 sales dashboard, Level 3 Module 3's RLS dashboard).
2. Impact analysis answers "if I change or deprecate this data source, what breaks?" — before archiving Analyst B's incorrect 6700-total source, lineage confirms whether any published dashboard currently depends on it, so removing it doesn't silently break a report someone still relies on.

5. Governance roles and workflow

1. Typical governance workflow: an analyst builds and validates a data source (hand-checking totals as done throughout this course); a designated certifier (often a BI team lead) reviews it against known- correct totals (6880 grand total, 2210/3750/920 by Region); once confirmed, the certifier applies the Certified badge; any subsequent change to the certified source ideally goes through the same review before republishing, so the badge continues to mean something.
2. This module's dataset makes the review step concrete: a reviewer checking a to-be-certified Orders source can independently re-derive 6880 (East 2210 + West 3750 + Central 920) from the raw 8 rows before approving — exactly the kind of hand-verification this course has used throughout, applied as an actual governance gate.

How It Actually Works

1. Certification is stored as metadata on the published data source's content record on Server/Cloud — technically a flag plus an optional certification note, set via the content management page or the REST API's certification/certificationNote fields on a datasource resource. Setting it does not touch the .hyper file or trigger any recompute — it only changes what the Server's search index and catalog surface to viewers (badge, ranked higher in "recommended" results), so two data sources with the same name but different Sales totals (6880 vs 6700) remain equally queryable by connection string; certification changes discovery, not access.
2. A data quality warning is a separate metadata object attached to the content item, with a type (warning, deprecated, stale data, etc.) and free-text message; Tableau resolves and displays it at view render time by checking every data source feeding the workbook being opened, which is why the warning banner appears on any dashboard downstream of a flagged source — including ones the flagging admin never directly touched.
3. Lineage and impact analysis are built from Tableau's Metadata API (a GraphQL endpoint over the Data Catalog): every published workbook, data source, field, and database table is a node, and every "this field is used in this worksheet" or "this data source feeds this workbook" relationship is a stored edge. A lineage query for the Sales field walks these edges outward (upstream to the source column, downstream to every consuming sheet); impact analysis before deprecating Analyst B's 6700-total source is the same graph traversal run in reverse — "which workbook nodes have an edge into this data source node" — computed from the graph, not by re-scanning every workbook's XML on demand.
4. Certification, warnings, and lineage all read from metadata the certifier or Server admin must explicitly set or that Tableau infers from publish-time parsing — none of them re-validate the data itself against ground truth. That's exactly why Section 5's workflow puts a human hand-recomputation step (independently re-deriving 6880 from the 8 raw rows) before applying the badge: the certification mechanism has no way to know 6700 is wrong on its own, it can only advertise which source a human has vouched for.

Cheat sheet

Concept	What it does	What it does NOT do
Certification	Marks the recommended, trusted source	Doesn't block use of other sources
Data quality warning	Flags known issues (staleness, etc.)	Doesn't fix the underlying data
Lineage	Shows source → field → downstream use	Doesn't prevent breaking changes alone
Impact analysis	Shows what breaks before a change	Doesn't make the change safe automatically

Exercise

A fourth analyst's data source shows a Sales grand total of 6940 — 60 more than the certified 6880. Using the certified source's per-region breakdown (East 2210, West 3750, Central 920) as ground truth, describe the governance steps you'd take before either certifying or rejecting the new source (re-derive its per-region totals from its own row-level data, locate which row(s) account for the extra 60, and determine whether it reflects a legitimately new order or a data error) — do not assume either number is correct without recomputing from raw rows.