Where each mechanism comes from, how we grade the evidence, and what we do when the literature is still arguing. Nothing here is a trade secret: the method only works if it can be checked.
For a behaviour to happen, it takes capability (the person can), opportunity (the context allows it) and motivation (it is worth doing). If one fails, the behaviour does not occur — and the right intervention depends on which one failed.
That is why training someone who already knows changes nothing, and why recognition does not fix system friction. The diagnosis exists so you do not get the condition wrong.
The three conditions come from the model by Michie, van Stralen & West (2011), which brought together in a single scheme what has to be in place for a behaviour to happen. It says where behaviour fails. The why comes from other theories, tested in other literatures: psychological safety, psychological ownership, social norms, justice and equity, job demands and resources, status quo bias, habit. Each mechanism in the base comes from one of them and is placed in one of the three conditions — the placement is our classification, declared in each mechanism’s record.
The intervention follows the guide in which the same authors linked the three conditions to the ways of intervening (Michie, Atkins & West, 2014): nine intervention functions — education, training, enablement, restriction, environmental restructuring, persuasion, incentivisation, coercion and modelling — linked to the conditions by a matrix. The function says the kind of intervention; the concrete action the Engine suggests is one way of fulfilling it in your context, and the same action often fulfils more than one function.
They know they should, but lack knowledge, criteria or mental space. Six mechanisms: A1 to A6.
Here you intervene by teaching, training or providing the means. In the Engine, that means hands-on instruction in the person’s own work, putting the criteria in writing or having someone at their side in the first weeks.
The tool, the routine or the group gets in the way. Two mechanisms: B7 and B8.
Here you intervene in the environment itself: changing the system or the routine, providing the means, setting the example and, when needed, closing the old path. In the Engine, that means removing steps and bringing the tool to where the work happens, or making the new practice visible, starting with leadership.
They can and could, but choose the usual way. Seven mechanisms: C9 to C15 — the largest sector, and that is true of the problem, not an accident of the design.
This is the condition with the most paths: informing, persuading, changing what gets rewarded, setting the example, altering the environment that triggers the habit. In the Engine, that means reviewing the metric that rewards the old way, showing the real cost of the current way or creating a reminder at the point of use. Demanding and punishing also act here, but they come last: they tend to suppress the problem rather than solve it.
The three conditions say where to look. What the Engine looks for inside each one comes from theories in organisational psychology, social psychology and behavioural economics, each in its role.
Knowing how to use the tool is not enough. Capability is also missing when the rule on what to share is ambiguous, when nobody knows what is good enough to show, when the workload leaves no attention for anything new (job demands and resources) and when the old way has become a habit, triggered by the context before any decision.
The physical environment counts: every extra step in the system is a cost the old way does not have. So does the social environment: the group’s norms, both what everyone does and what everyone expects, weigh more than the written rule. Whoever follows the rule alone is left out of the conversation that matters.
Most of the theories are here. The organisation may reward one thing while hoping for another (Kerr, 1975). People who feel the work is theirs are reluctant to open it up (Pierce, Kostova & Dirks, 2001). Without psychological safety, exposing a mistake is costly, and silence becomes protection (Edmondson, 1999; Morrison & Milliken, 2000). Information can also be power, withheld on purpose (Raven, 1965, 1993; Connelly et al., 2012). Add to these the preference for the familiar, the overvaluing of one’s own work and perceived fairness between teams.
From condition to way of intervening: each of the nine functions acts on some conditions and not on others. Incentivisation and coercion, for instance, only affect motivation: no reward or punishment, on its own, creates capability or opportunity. That is why the Engine starts from the diagnosis, not from the intervention.
each with the signal that gives it away and what tells it apart from its nearest neighbour
The grade does not measure how much we like the model: it measures the state of the literature. A D stays in the base because naming the phenomenon helps the conversation — as long as you know the ground.
Effect sustained by multiple independent studies and quantitative synthesis. Direction is reliable; magnitude varies with context.
Solid base in the field or the lab, without the density of replication of an A. Use it, and observe the result.
It exists, but its size and conditions are contested. Treat it as a hypothesis to test, never as the sole basis for a decision.
Popular, yet with a thin or methodologically contested empirical base. It serves as vocabulary, not as justification.
Much of the behavioural science popularised in the 2000s did not replicate well. Rather than removing those models from the base and pretending they never existed, we mark them C or D and explain what is in dispute.
When a new meta-analysis weakens an effect, the grade drops and the change enters the changelog with date and reason. You can see what the tool used to recommend and what changed.
A mechanism graded A can have low support in your case, and a C can be shouting in the signals. The grade is about the literature; support is about your session. The two rulers appear together, never summed.
Human curation: the base does not rewrite itself from usage. Every change to the taxonomy goes through review by whoever answers for the method — and stays public.
No author quote enters the tool without human verification of the source. Popular but disputed attributions are published without attribution or simply not used.