Evidence and policy begin their relationship at a particular moment. An institution faces a problem, tests an intervention, observes a result and decides that the evidence is strong enough to justify action. What happens next, however, is asymmetric. The policy enters the institutional world, where it can acquire legislation, budgets, staff, infrastructure, procedures and expectations that help it persist. The evidence remains connected to the conditions under which it was generated. Years later, the institution may still be implementing essentially the same policy even though the world that made the original experiment informative has changed substantially.
This does not mean that the experiment was wrong. Nor does it mean that an old policy is necessarily obsolete. The problem is more precise: policy and evidence can run on different clocks. A policy can become increasingly embedded while the evidential relationship that once justified it becomes progressively less secure. Institutional persistence and evidential validity are therefore different properties, and an institution that confuses them can mistake durability for continuing justification.
Policies have many mechanisms that support continuity. They become part of legal frameworks, annual budgets, organisational structures and professional routines. Citizens and other organisations adapt to them. Information systems are built around them. Staff acquire specialised knowledge for administering them. Political commitments accumulate, and complementary programmes emerge on the assumption that the original intervention will remain. Each development can make the policy more difficult to change even though none necessarily provides new evidence that its underlying causal assumptions remain valid.
Evidence has no equivalent guarantee of persistence. Its authority depends on whether the conditions connecting the original finding to present action continue to hold. Populations change, technologies develop, behaviour adapts, institutions reorganise and other policies alter the environment in which the intervention operates. The original study may remain methodologically sound while becoming increasingly distant from the system to which its conclusions are now being applied.
This creates evidence–policy drift. At the beginning, the policy and its evidential foundation may fit closely. The intervention tested is similar to the intervention implemented, the target population resembles the experimental population and the surrounding institutional conditions remain comparable. Over time, however, those elements can move apart. The policy may be modified incrementally, its users may change, implementation may scale or the wider system may evolve. None of these changes needs to be dramatic. Drift can accumulate through many small departures until the current policy is being justified by evidence generated for a configuration that no longer quite exists.
The danger is especially easy to overlook because administrative stability is visible while evidential drift often is not. An institution can see that a programme continues to operate, that staff know how to deliver it and that established processes remain functional. Those observations create a sense of continuity. Yet the evidential question is different. It asks whether the causal relationship that supported the original intervention continues to apply under present conditions. A smoothly operating programme can therefore be administratively stable while becoming epistemically less secure.
Scale can intensify this divergence. An intervention tested in a limited setting may later become a national or system-wide programme. The original experiment may have involved unusually capable implementers, motivated participants or close monitoring. Once scaled, the policy interacts with a much more heterogeneous population and with organisational environments that were absent from the original study. The policy persists because the institution has successfully expanded it, but expansion itself may have changed some of the conditions that made the evidence persuasive.
The same problem appears when the intervention evolves. Policies rarely remain completely static. Eligibility criteria change, technologies are introduced, administrative procedures are simplified, responsibilities move between organisations and new objectives are added. Each modification may seem small enough not to require reconsidering the original evidence. After years of incremental adaptation, however, the institution may be operating something meaningfully different from what was initially tested while continuing to cite the same evidential foundation.
Evidence–policy drift should not be confused with forgotten provisionality. A policy can drift away from evidence even when the original evidence was strong enough to justify adoption. The institution may have been entirely reasonable at the beginning. The problem emerges because the policy persists while the evidential conditions change. Forgotten provisionality describes a different situation in which uncertainty was never fully resolved but disappeared from institutional awareness as implementation became routine. One begins with justified confidence that may decay; the other begins with conditional confidence that becomes falsely settled.
Recognising evidence–policy drift also avoids a common mistake in debates about legacy policy. The relevant question is not whether a policy is old. Age can increase the opportunity for drift, but it does not prove that drift has occurred. Some policies rest on mechanisms that remain stable for long periods, while others operate in rapidly changing environments. A decades-old intervention may still be well supported; a much newer one may already be poorly aligned with the evidence that justified it. What matters is the changing relationship between the policy, the mechanism and the context.
This is why institutional entrenchment can be epistemically misleading. The more deeply a policy is embedded, the easier it becomes to interpret its persistence as evidence of its validity. Yet entrenchment often tells us more about institutional dependency than about causal effectiveness. A programme may survive because budgets, organisations and legal obligations have formed around it. These are powerful reasons for continuity, but they are not themselves evidence that the original intervention remains the best response to the problem.
The asymmetry can create a cumulative governance problem. As institutions add new policies without systematically revisiting the evidential foundations of existing ones, they accumulate layers of intervention generated under different historical conditions. Each layer can continue to shape present behaviour while carrying assumptions from the environment in which it originated. The institution may therefore become increasingly dependent on inherited evidential claims whose current applicability varies considerably.
A mature response is not periodic scepticism towards everything that already exists. Constantly reopening settled decisions would impose its own costs, making stable administration difficult and consuming analytical capacity that could be better used elsewhere. The objective is to identify where evidence–policy alignment is most vulnerable. Policies operating in rapidly changing environments, interventions whose implementation has changed substantially, programmes expanded far beyond their original population and decisions based on context-sensitive mechanisms deserve more attention than policies supported by highly stable relationships.
This requires institutions to maintain a visible connection between policies and the evidence that supports them. Decision-makers should be able to identify not only which study originally justified an intervention but what features of the study made its conclusions transferable. If those conditions change, the institution can recognise that the evidential relationship needs attention before policy failure becomes obvious.
The distinction is particularly important when new technology transforms an existing policy environment. A programme developed when administrative information was scarce may operate differently once large datasets become available. A regulatory intervention designed before digital platforms reorganised a market may continue long after the structure of that market has changed. A public service evaluated before widespread automation may behave differently once human and machine decisions become intertwined. The policy can remain recognisably the same in administrative terms while the system in which it operates becomes fundamentally different.
Evidence–policy drift can also occur in the opposite direction: changing conditions may make an intervention more effective or relevant than the original evidence suggested. Revalidation is therefore not merely a search for reasons to abandon old policies. Its purpose is to restore alignment between confidence and current evidence. Sometimes that process will weaken the case for an intervention; sometimes it will strengthen it; sometimes it will reveal that the policy should be modified rather than removed.
What matters is that persistence should not silently substitute for justification. Institutions need continuity, but continuity becomes cognitively dangerous when it prevents them from noticing that the reasons for continuing have changed. A policy may still deserve to exist, but the case for its existence should eventually depend on more than the fact that an experiment once supported it and the institution subsequently built itself around the result.
A policy can outlive the experiment that justified it because institutional action and evidential authority operate on different temporal trajectories. Policies acquire structures that help them persist, while evidence remains dependent on the conditions that made its original inference valid. When those trajectories diverge, administrative stability can coexist with declining evidential fit. An intelligent institution therefore does not assume that a durable policy carries a durable justification. It preserves the relationship between action and evidence strongly enough to notice when the policy has remained in place but the world that once made its justification persuasive has moved on.