Evidence is decision-grade only in relation to a decision.
Decision-grade evidence is sufficiently relevant, current, traceable, tested and transparent about uncertainty to support a specified decision at a proportionate level of risk.
Important qualification: “decision-grade” does not mean certain. It means the decision-maker can see the evidence’s strength, limits and effect on the available choices.
Faster access to information has not removed the need to test it.
Businesses can now discover, combine and summarise more material at greater speed. That increases the value of provenance, entity matching, recency and explicit uncertainty.
A plausible summary may contain a correct fact attached to the wrong organisation, a current publication built from old observations, an announcement mistaken for implementation or an engagement signal treated as intent. None of these failures requires fabricated information. They arise when context is lost between source and decision.
The Office for Statistics Regulation’s Code of Practice organises good evidence around Trustworthiness, Quality and Value. It stresses suitable sources and methods, transparency about uncertainty and bias, relevance and clarity and notes that excessive information can make appropriate interpretation harder.1 These principles are written for official statistics but are deliberately applicable more widely, including by commercial organisations.
Four layers prevent evidence from sounding stronger than it is.
Directly supported by an identifiable source.
The record preserves who published it, when, what period it covers and the method or definition needed to interpret it.
A reasoned explanation of what one or more facts may mean.
The inference is attributable to Frontier or the named analyst and records credible alternatives or contradiction.
A provisional input required to continue.
The assumption is explicit, bounded and linked to the evidence that would confirm, revise or remove it.
A material gap that could change the decision.
Unknown does not mean “no”. Missing evidence remains missing until it is established.
Two further labels matter. Conflicting evidence records material sources pointing in different directions. Decision implication explains which option, risk or next step the evidence affects. Neither should be hidden inside a generic confidence colour.
Quality means fitness for the decision at hand.
Data quality has several dimensions and requires trade-offs.
The Government Data Quality Framework identifies completeness, uniqueness, consistency, timeliness, validity and accuracy. It warns that perfect quality may be unattainable and that improving one dimension can affect another.2
Commercial translation: the decision owner should know which quality dimensions are material for the choice at hand.Verification and validation answer different questions.
The AQuA Book distinguishes verification (whether analysis is error-free and meets specification) from validation (whether it is appropriate and fit for the intended purpose). It calls for assurance proportionate to financial, legal, operational and reputational risk.3
Consequence: a perfectly calculated model can still be the wrong model.Outcome observation is not the same as attribution.
The 2026 Magenta Book asks whether an intervention was implemented as intended, what outcomes and costs occurred, whether there were unintended effects and what might have happened without the intervention.4
Commercial translation: pipeline movement after an activity does not by itself prove that the activity caused it.Entity fit is a first-order evidence test.
An organisation, legal entity, site, brand, buyer, property owner and service operator may be related without being interchangeable. Evidence attached at the wrong level can be accurate in isolation and unsafe for the decision.
What would change the conclusion: a retained identity relationship showing why the source record applies to the decision target.The Decision-Grade Test.
Ten questions expose whether a claim can carry the weight placed on it.
- 01Decision fit
Which precise choice could this evidence change?
- 02Provenance
Who produced it, when, by what method and from which source?
- 03Entity fit
Does it concern the correct organisation, site, market and period?
- 04Recency
Is the observation current enough for the intended use?
- 05Data quality
What is known about completeness, duplication, consistency, validity and accuracy?
- 06Corroboration
Is it independently supported, contradicted or unverified?
- 07Analytical separation
Which parts are fact, interpretation, assumption and unknown?
- 08Uncertainty
Which range, bias, limitation or alternative explanation matters?
- 09Choice comparison
What alternatives and trade-offs does the evidence inform?
- 10Action and learning
Who decides, what happens next and how will the result be reviewed?
The headline is true. The investment conclusion is not yet safe.
A board paper states that energy costs at a multi-site organisation increased by 18%.
The finance system supports the total. Before the board approves capital intervention, it must know whether the movement reflects consumption, tariff, weather, site changes, data coverage or a combination.
Source, scope and calculation are checked. Nominal cost movement is established.
The current evidence cannot isolate the problem or compare sites fairly.
The next step is proportionate, reversible and designed to create decision-grade evidence.
“The figure is accurate” and “the recommendation is supported” are different judgements. The second requires the evidence to fit the decision, not merely the sentence.
Better evidence makes judgement inspectable.
Review focuses on the material gap, trade-off and authority required.
Teams know when a rapid check is enough and when independent challenge is needed.
Unknowns are converted into tests, owners and review points.
The record keeps the source, date, identity and limitations alongside the conclusion.
Evidence strength and action state must remain separate
A fact can be strongly evidenced but commercially irrelevant. A weak early indicator can justify low-cost investigation without supporting external action. Combining both into one red–amber–green score hides the reasoning a decision-maker needs.
Build one evidence record the decision-maker can challenge.
- Decision question and accountable owner.
- Exact claim the evidence is being used to support.
- Source, publisher, method and direct link.
- Observation, publication, detection and review dates.
- Correct organisation, site, period and geographic scope.
- Verified facts separated from interpretation and assumptions.
- Contradictory evidence and alternative explanation.
- Material unknowns and how they could change the choice.
- Assurance completed and assurance not completed.
- Decision, next action, owner and review trigger.
Language that protects understanding
| Evidence state | Use language such as | Avoid |
|---|---|---|
| Established fact | “The source records…” | “It is obvious that…” |
| Supported interpretation | “Taken together, this suggests…” | “This proves…” |
| Working assumption | “For this assessment, we assume…” | Leaving the input unstated |
| Unknown | “The available evidence does not establish…” | “No evidence means no activity” |
| Scenario | “If X occurs, Y could follow…” | Presenting a conditional future as a forecast |
More assurance is not automatically better.
Evidence can be incomplete and still sufficient for a reversible step. Conversely, a large dataset may remain unsuitable where it uses the wrong entity, period or outcome definition. The purpose and consequence determine the standard.
This paper does not provide fixed weights for the ten tests. A safety-critical property decision, a marketing experiment and a routine account-priority review have different failure costs and assurance needs.
What would change the conclusion? If a material limitation is resolved, a conflicting source appears or the decision itself changes, the evidence must be reassessed. “Decision-grade” is a versioned judgement, not a permanent property of the data.
Evidence and supporting sources used in this paper.
- Office for Statistics Regulation, Code of Practice for Statistics, edition 3.0, published 3 November 2025. Accessed 18 October 2026.
- Government Data Quality Hub, The Government Data Quality Framework, published 3 December 2020. Accessed 18 October 2026.
- Government Analysis Function, The AQuA Book, published 30 July 2025. Accessed 18 October 2026.
- HM Treasury and Evaluation Task Force, Magenta Book: Central Government guidance on evaluation, updated 15 May 2026. Accessed 18 October 2026.
- HM Treasury, The Green Book 2026, updated 5 February 2026. Used for proportionality, objectives, options and uncertainty.
- Frontier Insight Group synthesis. The Decision-Grade Test and evidence labels are original Frontier frameworks grounded in the cited quality and assurance disciplines.
Method, accountability and reuse
Editorial method. This is an original Frontier Insight Group analytical paper based on research completed by 18 October 2026. It synthesises primary and authoritative public sources, regulatory material, disclosed industry research and, where stated, peer-reviewed or emerging research. The scope and material limitations of each source are recorded in the paper. Public-sector or professional frameworks are translated only where the boundary is made clear; they are not presented as proof of guaranteed outcomes.
Human responsibility. Research systems supported source discovery, comparison and quality checking. Editorial judgement, verification and publication approval remained under human direction.
Corrections. Material corrections or revisions will be dated and described in the version record. Evidence published after the cut-off belongs in a later edition.
Frontier Insight Group (2026). Decision-Grade Evidence: What a business needs to know before a claim is safe to act on. Frontier Paper 03, version 1.0. Available at: https://frontierinsightgroup.co.uk/research/decision-grade-evidence/
Copyright and reuse© 2026 Frontier Insight Group. All original editorial analysis, frameworks, diagrams and presentation are protected. Short quotations with full attribution are permitted where lawful. Reproduction, adaptation or commercial redistribution requires permission. Third-party material remains the property of the relevant owner. Public-sector information is attributed to its source and used under the Open Government Licence v3.0 where applicable.
Professional boundaryThis publication is for general business understanding. It is not legal, financial, regulatory, procurement, investment or building-safety advice. Decisions should be tested against the organisation’s facts, duties, professional advice and decision rights.