AI Glossary / Data engineering and architecture / Intermediate

Single version of the truth

Definition

A single version of the truth is a governed, consistently defined view of data that people agree to use for a specified reporting or decision purpose. It aims to prevent teams from producing incompatible answers because they use different definitions, transformations, or refresh points. The phrase is an organizational objective, not a formal guarantee that every conclusion is correct or that only one valid perspective can exist.

What it means

Two reports can draw from the same source and still disagree. One may count orders when placed, another when shipped; one may include cancellations, another exclude them. A single agreed reporting view makes those choices explicit. It can also preserve several legitimately different measures under clear names. Consistency means applying the stated definition reliably, not forcing distinct business questions to share one undifferentiated number.

How it works

Agree on the relevant entities, measures, inclusion rules, time periods, and calculation logic. Build governed datasets or semantic models that apply those rules and make their lineage visible. Define refresh timing and revision handling. Validate outputs against known examples, publish the definitions, and manage changes with affected users. Physical storage can be centralized or distributed. The result depends on shared interpretation and controlled transformations as much as on where the data resides.

Three practical examples

• Performance reporting: teams use an agreed definition of completed orders for a monthly operational dashboard, while retaining separately named measures for placed and shipped orders.
• Education analytics: a website distinguishes lesson views, lesson completion, and assessment success so teams do not report all three as “learning outcomes.”
• AI summaries: an assistant retrieves a governed metric with its definition and reporting period, reducing the chance that it combines incompatible totals from unrelated reports.

Think of it this way

Think of musicians agreeing on the same score and version before a performance. Agreement helps them coordinate, but it does not mean there is only one possible arrangement or that every note will be played correctly. The analogy is limited because data views involve explicit transformation rules, source quality, and timing. A shared title on two dashboards does not prove they use the same score.

Worked example

A toy month contains 120 placed orders. Ten are canceled, and fifteen of the remaining orders have not yet shipped by month-end. A report of placed orders shows 120. A report of non-canceled orders shows 110. A report of shipped orders shows 95. All three can be correct under different definitions. If management requests “shipped orders during the month,” the governed measure must apply that definition and relevant time rule consistently. Renaming all three reports “total orders” creates confusion. The solution is not averaging 120, 110, and 95; it is selecting and documenting the measure that answers the intended question.

Illustrative case study

Fictional example: Linden Services finds that regional dashboards count a resolved ticket at different stages. One uses agent closure and another uses customer confirmation. The team keeps both measures but names them distinctly and publishes a shared definition for the executive dashboard. It tests tickets that satisfy only one condition and records the reporting cutoff. This invented scenario illustrates agreement on interpretation rather than proof that a centralized dashboard always reflects the best business judgment.

Common misconceptions and mistakes

Do not assume a common database guarantees common metrics. Do not treat the agreed view as immune to correction. New evidence, late-arriving records, and changed definitions may require revisions that should remain traceable. Avoid suppressing legitimate alternative views, such as historical versus current classifications. Also do not claim that agreement makes analytical conclusions unarguable. Users can share the same data and still make different assumptions, models, or decisions.

Business use

Use the concept to reduce avoidable reconciliation work and improve the interpretability of shared reports. Begin with disputed measures and document their purpose, formula, grain, period, and exclusions. Assign owners and publish version changes. For AI reporting, provide metric definitions alongside values and require the assistant to identify the relevant period and scope. A single consistent input view helps, but it does not guarantee sound reasoning or prevent the assistant from overstating what the figures show.

People also ask

Is this the same as a single source of truth?

The phrases are often used interchangeably. A useful distinction is that source authority concerns where accepted data comes from, while a consistent version concerns how it is interpreted and presented for a purpose.

Must every report show the same number?

No. Different questions can legitimately use different, clearly defined measures.

Can the agreed version change?

Yes. Changes should be governed, explained, and versioned so users can interpret comparisons correctly.

Check your understanding

Two dashboards use identical source records but different cancellation rules. Does sharing the source establish a single consistent reporting version?

Answer: No. The transformation and inclusion rules differ. Either align the rules for the same measure or label the measures distinctly according to their purposes.

Related terms

• Single source of truth
• Data Warehouse
• Data governance
• conformed metric
• semantic model
• lineage
• metric definition

Learn next

choose one disputed dashboard number and write its full definition, including grain, date basis, exclusions, and refresh time. Recalculate a tiny known example under each team's rules. Use the differences to decide whether the measures should be aligned or kept separately named.

More in Data engineering and architecture

• Simulation decomposition - Simulation decomposition, often called SimDec, is an uncertainty and sensitivity analysis approach that explains an output distribution by separating it...
• Simultaneous and heterogeneous multithreading - Simultaneous and heterogeneous multithreading, abbreviated SHMT, is a research approach to using different kinds of processing hardware concurrently within...
• Single customer view - A single customer view is a coherent representation of a customer assembled from relevant records across systems or channels.
• Single-instance storage - Single-instance storage stores one physical instance of identical content while allowing several logical objects or references to use it.
• Single instruction, multiple threads - Single instruction, multiple threads, or SIMT, is an execution model in which a group of threads can execute a common instruction while each thread works...
• Sink (pipelines) - A sink is the destination stage or component to which a data pipeline writes its output.

Sources

Primary sources consulted: IBM, Dimensional Warehouse Model, https://www.ibm.com/docs/en/dmeu/3.0.1?topic=dmeu-dimensional-warehouse-model — excerpt on agreed conformed facts consulted. Microsoft, BI Solution Architecture, https://learn.microsoft.com/en-us/power-bi/guidance/center-of-excellence-business-intelligence-solution-architecture — source/version terminology excerpt consulted. Microsoft data-governance commentary also notes overlapping usage. Examples and governance recommendations are original; full editorial review remains pending.

Previous: Single-subject design Next: Singleton