Consistency and Its Discontents
"Be consistent" seems like uncontroversial design advice. Nielsen's usability heuristics include "consistency and standards" - users "should not have to wonder whether different words, situations, or actions mean the same thing" (Nielsen, 2003). Every design system espouses consistency. Every style guide enforces it. But the literature reveals that consistency operates at multiple layers, that arguments against it are as substantial as arguments for, and that the critical question - consistency with whose worldview? - is often left unasked.
Defining Consistency
Across the literature, "consistency" is treated as the degree to which elements of a design display predictable sameness that supports transfer of knowledge, coordination, and trust. But the focal point shifts with disciplinary lens.
In human-computer interaction, the emphasis falls on visual, semantic, and behavioural uniformity that enhances usability. Jordan (2002) calls consistency "a property of interface design that is recognised as being central to product usability", and Burns and Hajdukiewicz (2017) frame it as enabling transfer: "the way the system looks and works should be consistent at all times". In service design, the emphasis moves to cross-channel coherence; Polaine, Løvlie and Reason (2013) describe services that "speak with one voice" across touchpoints, and Downe (2020) makes this explicit: "The service should look and feel like one service throughout, regardless of the channel it is delivered through". In policy design, consistency refers to the absence of contradiction among goals and instruments, and Howlett and Mukherjee (2018) distinguish consistency (non-contradiction) from coherence (positive synergy) and congruence (fit with context) - related but distinct qualities. In critical design, consistency raises questions of power; D'Ignazio and Klein (2020) ask whose worldview is being enforced, and Resnick (2019) warns that apparently "consistent" visions often naturalise dominant perspectives and erase marginalised identities.
A Hierarchy of Consistency
Synthesising the layered models across these literatures produces eight inter-dependent tiers. Violations higher in the stack carry deeper ramifications than deviations at lower levels.
Tier 1: Logical/Data Consistency
The foundation. Same data, same rules, same referential integrity across systems. Glenn (2009) describes cross-consistency assessment in morphological analysis - every combination of parameters must be logically non-contradictory. In machine learning, cycle-consistency ensures that transformations are reversible (Zhu et al., 2017).
Tier 2: Semantic/Terminological Consistency
The same concepts named the same way. Morville and Rosenfeld (2006) emphasise developing "consistent labelling systems, not labels" - the framework for naming matters more than individual decisions. Jackson (2021) shows how inconsistent metaphors confuse users even when individual controls are well-labelled.
Tier 3: Visual/Perceptual Consistency
Colours, typography, iconography, and spacing that signal belonging. Frost (2016) shows how design systems "promote consistency and cohesion, speed up productivity, establish a shared vocabulary". Head (2016) extends this to animation - consistent easing and timing create visual coherence.
Tier 4: Behavioural/Interactional Consistency
Controls that work the same way across contexts. Jordan (2002) describes this as "similar tasks done in similar ways". Wroblewski (2008) shows how consistent form patterns reduce errors, and Johnsen and Porathe (2021), studying ship bridges, found that consistent design helps operators "predict how interfaces will look and function".
Tier 5: Structural/Architectural Consistency
Navigation patterns, information architecture, and page layouts that create spatial predictability. Burns and Hajdukiewicz (2017) discuss ecological interfaces where structure maps to domain constraints. Beyer and Holtzblatt (2016) emphasise consistent grids and navigation structures.
Tier 6: Systemic/Cross-Channel Consistency
What Resmini and Rosati (2007) call "external consistency" - "maintaining the same logic along different media, environments and times". Risdon and Quattlebaum (2018) describe touchpoints that "should be orchestrated as one with the same underlying intent". Holliday (2022) shows how design systems enable cross-team consistency.
Tier 7: Strategic/Vision Consistency
Alignment between design decisions and organisational purpose. Goodwin (2011) argues that internal consistency "makes all the parts of the design seem to belong together" and supports elegance. Stanford (2022) extends this to organisations - leaders must "behave consistently in similar situations".
Tier 8: Philosophical/Ethical Consistency
Value coherence. Resnick (2019) asks whether design practice aligns with declared values. Cottam (2021) examines consistency between stated principles and material outcomes. D'Ignazio and Klein (2020) interrogate whose values are being made consistent.
The Temporal Lens
Cutting across all tiers is temporal consistency - stability and evolution through time. Overkamp (2019) warns that assuming "temporal consistency" blinds designers to needed change. Lundberg (2008) shows how measurement instruments must maintain consistency for longitudinal validity. The rigid present becomes the inconsistent legacy.
Arguments For Consistency
The case for consistency is well-established across several dimensions. Consistent interfaces reduce mental load: users learn patterns once and apply them across contexts, and Stanton et al. (2017) document how re-use of familiar patterns lowers error rates. Johnsen and Porathe (2021) found that consistent ship bridge interfaces help operators "predict how interfaces will look and function due to familiarity and experience". In high-stakes domains, consistency prevents catastrophic slips; Branaghan et al. (2021) document how standard colour codes and consistent pedal placement reduce harm in medical devices and vehicles.
Uniform visual and interaction languages also create credibility. Frost (2016) argues that design systems "lead to cohesive, consistent experiences" that build trust, and for public services this matters considerably - the NHS brand evokes trust and reassurance. At an organisational level, design systems reduce coordination costs; Holliday (2022) shows how GOV.UK's approach lets "different teams learn from the research and experience of others", and Howlett and Mukherjee (2018) argue that policy packages exhibiting internal consistency outperform ad-hoc assemblages. Even in machine learning, models whose outputs remain stable under perturbation are harder to attack (Tsiligkaridis and Tsiligkaridis, 2021), and consistency enables transfer learning and composability (Zhu et al., 2017).
Arguments Against Consistency
But the literature also documents substantial risks. Excessive regularity can ossify; Hey et al. (2007) warn that strict consistency may "have a limiting effect on the development of new ways of thinking about design", and Meadows (2008) notes that rigid systems can "trap organisations in patterns that ignore emerging context". Global rules may disregard local needs - Morville and Rosenfeld (2006) caution that "one size does not fit all", and Tsekleves and Cooper (2017) show how colour codes consistent in Europe fail in Nigeria, where red evokes different meanings.
The argument becomes more difficult around power. "Consistent" systems can naturalise the worldview of their creators; D'Ignazio and Klein (2020) document how data systems built by homogeneous teams "fail to detect harm to others precisely because the designers seek general rules that ignore situated difference". Arista et al. (2021) analyse airport body scanners where the "consistency" of the binary gender template merely reproduces cis-normativity, and Resnick (2019) warns that apparently "consistent" visions often erase people of colour and queer identities. The critical question is: consistency with what - and with whose worldview?
Wicked problems rarely permit single consistent views. Howlett and Rayner (2013) show how policy mixes may "degenerate" when rigid adherence prevents necessary adaptation, and real-world systems often require "patching, layering or drift" rather than perfect coherence. Sometimes breaking patterns creates value: Kolko (2010) shows that "breaking the consistency" at critical moments creates memorability, though rule-breaking must be strategic and scarce; games deliberately violate expectations to create surprise (Sylvester, 2013); and DiSalvo (2015) describes robots that purposely embed "irrationality" to provoke richer affect. Blackwell (2007) shows that consistent metaphors can mislead novices, and that an "inconsistent" metaphor may provoke creative exploration that a consistent one forecloses. Goodwin (2011) gives a mundane example: dual stair switches violate the usual up=ON convention, but users accept the inconsistency because the benefit outweighs the rule.
Consistent, Not Uniform
Government design guidance navigates this tension with a useful formulation: "be consistent, not uniform" (GDS Design Principles). Use the same patterns wherever possible, but "when this isn't possible we should make sure our approach is consistent" with underlying principles even if surface details differ. Singla (2019) elaborates: teams should maintain consistency so users can transfer learning, but with flexibility to improve "when better ways of doing things are found". Local adaptation is legitimate when grounded in evidence; the pattern is the default, exceptions require justification. This framing acknowledges that consistency serves users by reducing cognitive load while leaving room for innovation, contextual fit, and evolution. It is consistency of intent and principle rather than rigid uniformity of implementation.
Toward Purposeful Consistency
The literature suggests a balanced stance. Shared language (Tier 2) stabilises meaning before visual styling (Tier 3), and Jackson (2021) shows that conceptual consistency matters more than surface uniformity, so prioritising semantics first creates a firmer foundation than beginning with colours and typography. Pattern libraries should allow justified deviations; Singla (2019) argues that evidence-based exceptions are legitimate, and the pattern should function as a sensible default rather than an absolute rule. Deviations should be evaluated against user value, breaking rules only when the benefit exceeds cognitive cost - Kolko (2010) insists that rule-breaking must be strategic and scarce. Consistency is experienced rather than declared, and Jordan (2002) reminds us that consistency judgements must come from users, not just internal standards; a pattern that looks consistent on paper may feel inconsistent in use.
Temporal drift also demands attention. What is consistent today may need evolution tomorrow, and Overkamp (2019) shows how rigid systems become inconsistent legacies; planning graceful transitions is part of the work. Most fundamentally, the question of whose worldview is being enforced requires participatory methods to surface hidden biases. D'Ignazio and Klein (2020) and Resnick (2019) show that "consistent" often means "designed for people like us", and inclusive design practices help identify when consistency becomes exclusion.
Implications for Design Systems Work
For those building design systems in constrained environments, this analysis has several consequences. Consistency is necessary but not sufficient: it supports usability, trust, and scalability, but it can also ossify, exclude, and create contextual blindness, so the goal is purposeful consistency rather than maximum uniformity. The hierarchy matters because semantic and behavioural consistency (Tiers 2-4) have the most direct usability impact, strategic and ethical consistency (Tiers 7-8) determine whether the whole system serves its stated purpose, and visual consistency (Tier 3) is often over-emphasised relative to deeper layers.
The question "consistent with what?" is sharper than it first appears, particularly in constrained environments where multiple design systems coexist. A platform built on a vendor's application builder inherits one set of defaults; the organisation it serves - in our case the NHS - has another. Pursuing consistency with the vendor's patterns means inconsistency with NHS brand expectations; pursuing consistency with NHS standards means fighting the platform's defaults on every decision. The strategic level - which design system we are trying to be consistent with - matters more than visual uniformity, and failing to answer it explicitly means answering it implicitly by deferring to whatever is easiest, which is usually the vendor.
Critical reflection is required throughout: whose standards are being made consistent, and whose needs are being marginalised? These questions cannot be answered by the design system itself - they require ongoing interrogation. And documentation should acknowledge limits, because principles that claim consistency as an unqualified good are hiding complexity. Honest documentation acknowledges when consistency constrains, when exceptions are legitimate, and whose perspective the "consistent" approach embodies. The design principles we have been developing attempt to hold this tension, though whether they succeed in practice remains to be seen.
References
Arista, N., et al. (2021). Designing a human future with machines.
Beyer, H. and Holtzblatt, K. (2016). Contextual Design. Morgan Kaufmann.
Blackwell, A. (2007). The reification of metaphor as a design tool.
Branaghan, R., et al. (2021). Humanizing Healthcare. Springer.
Burns, C.M. and Hajdukiewicz, J. (2017). Ecological Interface Design. CRC Press.
Cottam, H. (2021). Radical Help. Virago.
D'Ignazio, C. and Klein, L.F. (2020). Data Feminism. MIT Press.
DiSalvo, C. (2015). Adversarial Design. MIT Press.
Downe, L. (2020). Good Services. BIS Publishers.
Frost, B. (2016). Atomic Design. Brad Frost.
Glenn, J. (2009). Futures Research Methodology v.3.
Goodwin, K. (2011). Designing for the Digital Age. Wiley.
Head, V. (2016). Designing Interface Animation. Rosenfeld Media.
Hey, J., et al. (2007). Design methods and representation. Design Studies.
Holliday, B. (2022). Multiplied. Public Digital.
Howlett, M.P. and Mukherjee, I. (2018). Routledge Handbook of Policy Design. Routledge.
Howlett, M.P. and Rayner, I. (2013). Policy design and non-design.
Jackson, D. (2021). The Essence of Software. Princeton.
Johnsen, S. and Porathe, T. (2021). Sensemaking in Safety Critical Situations. Routledge.
Jordan, P.W. (2002). Designing Pleasurable Products. Taylor and Francis.
Kolko, J. (2010). Exposing the Magic of Design. Oxford.
Lundberg, J. (2008). Social status - a state of mind?
Meadows, D. (2008). Thinking in Systems. Chelsea Green.
Morville, P. and Rosenfeld, L. (2006). Information Architecture for the World Wide Web. O'Reilly.
Nielsen, J. (2003). Enhancing the explanatory power of usability heuristics.
Overkamp, T. (2019). How Service Ideas Are Implemented. Linköping University.
Polaine, A., Løvlie, L. and Reason, B. (2013). Service Design. Rosenfeld Media.
Resmini, A. and Rosati, L. (2007). Pervasive Information Architecture. Morgan Kaufmann.
Resnick, E. (2019). The Social Design Reader. Bloomsbury.
Risdon, C. and Quattlebaum, P. (2018). Orchestrating Experiences. Rosenfeld Media.
Singla, P. (2019). Government as a Platform. Sage.
Stanford, N. (2022). Designing Organisations. Economist Books.
Stanton, N.A., et al. (2017). Human Factors Methods. CRC Press.
Sylvester, T. (2013). Designing Games. O'Reilly.
Tsekleves, E. and Cooper, R. (2017). Design for Health. Routledge.
Tsiligkaridis, T. and Tsiligkaridis, A. (2021). Diverse Gaussian noise consistency regularisation.
Wroblewski, L. (2008). Web Form Design. Rosenfeld Media.
Zhu, J.-Y., et al. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks.