Behavioural Friction Theory

Paper 0 · Pødenphant Lund (2026a) · Read on Zenodo

You are in a meeting and you want to say something, but you hold back. In that moment your nervous system pays a price. That price is what BFT calls friction, and the same quantity turns up in everything from burnout to learning to a mouse that dares not leave the wall. Behavioural science has more than 20 research traditions that have each measured that quantity on their own, without talking to one another. This paper places them on a single map (Porges, Bandura, Sweller, Kahneman, Edmondson, each in their own cell) and shows what follows from that.

What it is about

Behavioural Friction Theory (BFT) is the original version of the framework. It is about biological systems: humans, animals, organisms that can die, move, and burn energy. It is the version that started it all, and the version that connects to clinical and educational practice.

The motivation is simple. Behavioural science has produced an enormous body of well-established findings over the last century: nudge theory, polyvagal theory, self-determination theory, cognitive load theory, psychological safety, expectancy-value, fight-or-flight, learned helplessness, the list is long. But those traditions do not talk to one another. They use different vocabularies, they answer different questions, they generate different interventions. There is no common currency.

BFT proposes that friction can be that currency. The proposal is programmatic rather than a delivered measurement: friction works as a scientific currency to the degree it is independently operationalised at the field level, and that is the task the paper's propositions are built to carry.

What friction means here

Friction is the cost the nervous system assigns to a potential action in a given situation. Not a metaphor. In a language model it has already been measured directly, and measured independently of the behaviour it predicts. For humans there is not yet a validated instrument that measures friction field by field, and building one is the first item on the paper's empirical programme. The task in the biological substrate is instrumentation: getting access to the competition a language model already shows outright.

When you hold back from saying something in a meeting, there is friction. When a mouse hesitates to step out from the safety of the wall into open space, there is friction. When a language model "struggles" with a question, there is friction too. In the model's case you can even see it directly in its output (that is the bridge to Paper 1 and Paper 3).

What "measurable" actually means. When you hesitate before saying something in a meeting, several things change at once, and they are all observable. Your heart rate climbs. The muscles around your jaw and shoulders tighten. The electrical conductance of your skin shifts. Your brain runs a parallel competition between candidate utterances and "keep quiet." Had you said it automatically, none of that would be present.

For a language model, friction shows up as longer per-token computation time, lower confidence in next-token predictions, and (most directly) more candidate tokens within the probability margin of the final choice. That last number is CR (Competing Routes), and it can be read straight off the model's output at no extra inference cost.

The bodily markers are, so far, adjacent measures. Capability reads most cleanly as the calibration error between estimated capacity and actual task demand. Effort reads reasonably as pupillometric and cardiac load. Markers specific to Safety and to Meaning still have to be found, and that is the first job of the measurement programme.

Different substrates. The same underlying quantity: how much computational cost the system pays to commit to one route rather than its alternatives.

What makes this usable as a common currency is that friction turns up in every behavioural decision, across every tradition that has tried to study behaviour. Loss aversion, willpower depletion, decision paralysis, status quo bias, freeze-flight-fight, motivation, demotivation, learned helplessness, burnout. They all describe states of the same underlying quantity, seen from different angles. BFT names the quantity and gives it a structure.

The four fields

BFT organises behavioural friction into four computational fields. They are not arbitrary slices. They answer to four failure modes any system that selects actions has to avoid: acting when action is unsafe, acting without directional value, acting beyond current capacity, and acting at disproportionate energetic cost. The paper's primary domain is the biological substrate, organisms that can die, move, and burn energy, and the evolutionary argument is the route to the structure there:

Safety: am I safe? Does this action threaten my existence? This field handles threat detection, risk avoidance, the "freeze-flight-fight" repertoire. High Safety friction shows up as caution, vigilance, withdrawal.
Meaning: am I moving in the right direction? Is what I am doing connected to something larger? This field handles goal coherence, values, social belonging. High Meaning friction shows up as drift, alienation, "going through the motions".
Capability: can I even carry this out? Do I have the skill, the energy, the conditions? This field handles self-efficacy and competence appraisal. High Capability friction shows up as hesitation, "not feeling ready", self-doubt.
Effort: is the cost worth it? Are there easier paths? This field handles the cost-benefit weighing of action under metabolic constraint. High Effort friction shows up as procrastination, energy conservation, "I'll do it later".

Whether the same four-field structure appears outside biology is carried forward as an open companion research direction, and that direction does not bear the behavioural and clinical claims made here. What has already been shown is narrower and sharper: a field's gating function can be installed by training. Companion work fine-tunes a model on an invented threat domain with no pre-trained valence and obtains a Safety gate that transfers to entities of that domain the model has never seen, and that opens its race at comprehension of the input rather than at the response. The necessity claim is kept to the structure, the four fields in that order; the contents are whatever the substrate's exposure installs. The split is written so it can be falsified: the function should be installable on machinery that lacks it, while the involuntary timing of the oldest fields should not transfer without the prepared priors.

The five layers

The four fields play out across five regulatory layers. Each layer is faster than the one below it, but more fragile under pressure:

Biological: the deepest, slowest layer. Cell biology, the autonomic nervous system, the immune system. Stable but slow to change.
Emotional: faster than biology, slower than cognition. Affect, mood, somatic markers in Damasio's sense.
Inner: the self-experience layer. Identity, narrative, self-talk, "who I am".
Cognitive: conscious reasoning. Fast, flexible, but easily overloaded under load.
External: the social and cultural layer. Norms, expectations, language, institutions.

One implication follows directly: biology always pays the final price. Chronic high pressure that is never resolved in the upper layers does not disappear, and over time it is absorbed by the biological layer. That is burnout. That is the body taking over what the upper layers could not dissolve. The layers form a downward cascade for unresolved load, and biology is where it lands.

The five regulatory layers form a hierarchy of speed versus robustness. Higher layers (External, Cognitive) handle load quickly but are fragile under sustained pressure; lower layers (Emotional, Biological) are slower but robust. Unresolved friction does not disappear. It sinks downward. Burnout is the bottom of that cascade.

The mechanism: RACE

How does friction arise mechanistically? Through what BFT calls the RACE architecture:

Several action candidates run in parallel as competing routes
Each accumulates evidence at a rate set by the nervous system's current state
The first to cross a threshold wins and becomes the executed action
The rest are suppressed, but at a cost (suppression takes work; unresolved alternatives leave traces; over time the suppression itself becomes the friction)

A small example. When you have to decide whether to say something in a meeting, several action candidates race in parallel inside your nervous system. The race happens; you experience only the winner. The slower routes have to be suppressed, and that suppression is what we call friction.

Friction is the cost of running this race: the unavoidable price of having to choose. Irreversible computation carries an irreducible energy price (Landauer), and the energy a substrate spends running its races is the energy side of the Effort field. It is not a flaw in the architecture. It is the architecture working correctly. Without it, no decisions could be made at all.

The friction matrix: where the existing theories live

The integrative claim becomes concrete when you place the major behavioural-science research traditions on the four-field × five-layer grid. The result is a striking pattern: each tradition has worked in one specific cell or two, often with no contact with traditions working in adjacent cells. Theories that have been treated as competing turn out to describe different cells of the same matrix.

	Safety	Meaning	Capability	Effort
Biological	Porges (polyvagal theory); Sapolsky (stress physiology)	Damasio (somatic markers)	—	—
Emotional	LeDoux (threat processing); Panksepp (affective neuroscience)	Russell (core affect)	ADHD / arousal research	Kahneman (System 1 under load)
Inner	Bowlby & Ainsworth (attachment); Masten (resilience)	Schwartz (values); Rogers (self-concept)	Bandura (self-efficacy); Vygotsky (ZPD)	Duckworth (grit)
Cognitive	Slovic (risk perception)	Weick (sensemaking)	Norman (mental models); Lave & Wenger (situated learning)	Sweller (cognitive load theory)
External	Edmondson (psychological safety); Cialdini (social norms)	Tajfel & Turner (social identity); Deci & Ryan (relatedness)	Deci & Ryan (competence); Bakker & Demerouti (JD-R)	Skinner (reinforcement); Thaler & Sunstein (nudge); Fogg (Behaviour Model)

Three observations follow from this mapping:

Each tradition has its cell. Polyvagal theory describes Safety friction at the biological layer. Cognitive load theory describes Effort friction at the cognitive layer. Psychological safety describes Safety friction at the external layer. They do not compete; they describe different cells of the same architecture.
The emptiness is informative, and most substantial. Some empty cells are research gaps, but others are predictions from the framework itself. The two empty cells at the bottom are empty for different reasons. The Effort field has almost no weight in the deep layers, and Effort is primarily a cognitive and external phenomenon, which is exactly why process design (the External × Effort cell) is so effective at addressing it. The Capability column at depth is thinly populated for a different reason: research attention has gone elsewhere. The biological consequences of load are well documented in stress physiology but rarely connected to the process-design research working in the External × Effort cell. The framework predicts which cells should be thinly populated and why: the deeper layers carry more regulatory weight for the earlier fields (Safety, Meaning), and almost none for the later fields (Effort).
Adjacent cells get conflated. Self-Determination Theory's three needs do not decompose symmetrically. Relatedness maps onto Safety friction at the External and Emotional layers, and competence maps onto Capability friction at the Inner and External layers. Those two land cleanly in specific field and layer combinations, which is why factor analyses keep finding them as separate dimensions. Autonomy does not. Autonomy is read here as the state in which the inner and emotional layers are not carrying prohibitive friction, so the upper layers can still compete in the race, a property of the whole landscape rather than of any one cell.

The empty cells are productive: each is a prediction that something specific should be found there. The non-empty cells re-place existing research as descriptions of parts of one regulatory architecture, rather than competing accounts.

30 testable claims

The full paper develops 30 formal claims, each a specific empirical commitment the framework takes on. Each comes with: a minimal test design, the existing empirical support, the falsification criteria, and the connection to other behavioural-science traditions. They are written as a programme to be pre-registered, and two of them are genuine reframes of the field's basic constructs and stand as hypotheses within that programme. A few examples:

Upper-layer interventions work only when biological baseline pressure is low enough for the upper layers to remain viable competitors in the race. Above that level, from chronic stress, sleep deprivation, or sustained threat, friction reduction at the biological layer is a precondition, and behavioural, pharmacological, and physiological interventions become structurally distinct entry points into the same architecture (P17)
The greatest marginal gain comes from reducing friction in the field currently carrying the most of it, provided the other fields are already below threshold. Which field is the bottleneck shifts from situation to situation with baseline pressure, and that gives a different reading of decision fatigue: what looks like a resource depleting is a change in which constraint is binding (P4)
"Motivation" loses its independent predictive power once friction across all four fields is measured and controlled for. Motivation is a valid phenomenological description of a low-net-friction state. This is a testable reduction, and it fails if motivation measures retain independent variance (P18)
The nature-versus-nurture debate dissolves when it is reframed in friction terms: when friction profiles are matched, intervention effectiveness is predicted to be independent of whether the profile has a constitutional or experience-based origin. The fight is about origin; the lever is about current state (P20)
Cues and nudges convert existing low-friction states into action, but they do not create them. With Safety or Meaning above threshold, cue effects approach zero, because the attentional gate the cue has to pass opens only once the higher priority is resolved (P5)

Why it matters in practice

If friction is a common currency, the implications fan out:

For clinical practice: depression, anxiety, burnout, and addiction can be read as patterns of field-and-layer dysfunction rather than as separate diseases in separate mental modules. Two patients with the same diagnosis can carry different friction profiles and respond differently to the same intervention. That diagnostic categories aggregate over several mechanisms is not new here; it is the central commitment of the RDoC programme and of computational psychiatry. What the framework adds is a candidate structure for the axes the profile varies over: which field is binding, which layer carries the limiting friction, and which history of short-range compensation lies behind it (P21). The field-level instrument that prediction requires does not yet exist.
For educational practice: a student who is not learning is not necessarily unmotivated. They may be experiencing high friction in Capability ("I can't do this") or in Meaning ("this isn't connected to anything I care about") or in Safety ("getting it wrong in front of others is dangerous"). Each calls for a different intervention. Treating all three as the same thing ("lack of motivation") is why so many educational interventions fail to transfer.
For organisational practice: psychological safety as a concept is a single field of friction (Safety in social contexts). It does not replace the others, and it subsumes neither motivation nor autonomy nor competence. BFT shows where it sits in the larger landscape and predicts when interventions that target it will help, and when they will miss the actual constraint.

How BFT relates to Friction Theory (FT)

BFT was the original version, biology-specific. Paper 1 (Friction Theory) built the instrument: a way to read friction directly out of a substrate’s output. The relation is written as BFT ⊂ FT, where FT names the general route-competition frame and the biological case is one instance of it. The nesting stands as a proposal.

This paper owns the content of the four fields, their computational forms, and their native always-on organisation in the biological substrate. The claim that friction and the four-field structure generalise beyond biology is carried forward as an open companion research direction, framed as a resource-rational task analysis rather than a settled law, and it does not bear the behavioural and clinical claims made here.

BFT adds no new mechanism. Its components are inherited and cited at the point of use: accumulation-to-threshold and race-to-commit in Ratcliff and in Usher and McClelland, control cost and cost-benefit arbitration in Expected Value of Control, hierarchical control in Powers and in Rao and Ballard, ordered appraisal in Scherer, and the limiting-factor structure of the four-field gate in Einhorn and Tversky. Where BFT coincides with one of these, that formalism has priority. Against resource-rational analysis, BFT stands as a specific answer to the question that framework leaves open, which statistics a bounded agent monitors: BFT proposes four particular ones, with a non-monotone Capability term.