AI Glossary / Data engineering and architecture / Intermediate

Single source of truth

Definition

A single source of truth is an agreed authoritative source or governed access point for a particular set of information. It tells people and systems where to obtain the accepted current value and where corrections should be made. The term describes an architectural and governance objective, not proof that the information is flawless. Its scope should be explicit: different data domains can have different authoritative sources.

What it means

If several teams maintain independent copies of a product's approved description, updates can diverge. A single-source approach identifies the authoritative record and makes other uses depend on it through controlled publication or synchronization. This does not necessarily require one physical database or forbid replicas and backups. The important question is which source has authority and how derived copies stay connected to it.

How it works

Define the information domain, its owner, and the rules for creating and correcting records. Designate the authoritative system or governed representation, then connect consumers through documented interfaces or controlled copies. Preserve version, lineage, and update information where needed. Resolve conflicts through an explicit process rather than allowing every consumer to edit its own competing master. Monitor freshness and failed synchronization. The design must also account for availability, access rights, and legitimate historical views.

Three practical examples

• Website publishing: a glossary team maintains approved definitions in one governed content source, and the website and downloadable editions are generated from that source.
• Product operations: a catalog system owns product identifiers and approved specifications, while sales channels consume controlled representations rather than inventing separate values.
• AI assistance: a company identifies the approved policy repository that an assistant should consult, with version and effective-date information supporting the answer.

Think of it this way

Think of an official noticeboard whose notices are copied to several rooms. The copies are useful, but everyone knows where an approved change originates. The analogy is limited because digital systems may have distributed storage and complex synchronization. Authority is a defined responsibility, not a property automatically created by putting a file in a central folder.

Worked example

A toy product record has an approved package weight of 2.4 kilograms in Catalog C. A sales spreadsheet still shows 2.2, and an old brochure shows 2.3. Under the agreed policy, Catalog C owns the current package-weight field. The team checks the evidence, corrects C if necessary, then updates dependent outputs from that record. It does not choose 2.4 merely because it is the largest number or average the three values. If the brochure describes a previous product revision, its 2.3 value may remain historically correct. The source-of-truth decision therefore needs scope and version context, not just one winning value.

Illustrative case study

Fictional example: Maple Guides publishes procedures on a website and in downloadable booklets. Staff had been editing both independently, causing disagreements. The team establishes one reviewed content source and records which published edition came from which version. Corrections are made there and then republished. The archived booklet remains labeled with its publication date. This invented case illustrates controlled derivation without claiming that centralization alone eliminates editorial errors.

Common misconceptions and mistakes

Do not interpret the phrase as a guarantee of objective truth. An authoritative record can still be wrong and must be correctable. Do not confuse one authoritative source with one physical copy; resilience often requires replication and backup. Avoid using the label to suppress legitimate differences in scope, time, or definition. Also distinguish a system of record from an integrated source for a particular view: the latter may assemble data whose original authority remains distributed across several domain systems.

Business use

A single-source approach can reduce avoidable disagreement when ownership and update paths are clear. Start with a high-value domain and document which fields are mastered where. Make corrections traceable and ensure consumers can identify stale versions. For AI products, use the designated source together with retrieval checks and citations; a model's memory or generated summary should not silently become a competing authority. Evaluate the process by whether users can find, verify, and correct the accepted information.

People also ask

Does it mean one database for the whole company?

No. The authority can be defined per domain or field and exposed through a governed architecture.

Can there be backups and cached copies?

Yes. They should preserve lineage and follow controlled update rules rather than become independent competing masters.

Is it identical to a single version of the truth?

Usage overlaps. A useful distinction emphasizes source authority here and consistent interpreted reporting in the latter, but organizations should define their own terminology explicitly.

Check your understanding

A cached report disagrees with the designated current source. Should someone simply edit the report's cell and leave the source unchanged?

Answer: Usually that creates another competing value. Investigate the discrepancy, correct the authoritative record or transformation as appropriate, and refresh the derived output through the agreed process.

Related terms

• System of record
• Data governance
• Single version of the truth
• Content management
• master data management
• lineage
• data stewardship

Learn next

pick one recurring piece of information and trace every place it is edited or copied. Assign an owner and an authoritative update path, then document how each consumer learns about changes and how historical versions remain distinguishable.

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, System of Record versus Source of Truth, https://www.ibm.com/think/topics/system-of-record-vs-source-of-truth — distinctions and architectural-use excerpts consulted. Microsoft, Unified Data Governance, https://www.microsoft.com/en-gb/microsoft-cloud/blog/technetuk/2020/12/10/unified-data-governance-using-azure-purview-preventing-data-lake-from-becoming-a-data-swamp/ — terminology-overlap excerpt consulted. Examples and governance workflow are original. Full editorial review remains pending.

Previous: Single program / multiple data Next: Single-subject design