Pressure, Hysteresis, and Experience: A Friction-Theoretic Framework for Clinical Intervention

Paper 8 · Pødenphant Lund, T. (2026j) · Hypothesis & Theory · concept DOI 10.5281/zenodo.20059865

When someone relapses under stress, the usual reading is a failure of will. This paper reads it as a race resolving too fast for a slower competing route to win. The proposal is a single substrate-computational mechanism, race-resolution under arousal-modulated commit-pressure with deposited hysteresis, under which addiction, rumination, ADHD, autism-spectrum presentations, PTSD and OCD, severe phobias, and aging are not separate machineries but different parameter-configurations of the same one. It is offered as a hypothesis with twenty-eight falsifiable predictions, not as clinical advice.

DOI (concept)	10.5281/zenodo.20059865
Manuscript type	Hypothesis & Theory
Cite-letter	2026j
Target venue	Frontiers in Psychiatry (Hypothesis & Theory)
Author	Tomas Pødenphant Lund [ORCID]

The mechanism, in words

Any system that decides under bounded resources runs three steps: parallel evaluation generates competing routes, accumulation integrates evidence for each until one crosses threshold, and commit irreversibly selects the threshold-crossing route, releases computational state, and deposits a trace that biases future evaluation of similar situations. What is experienced as choice is the commit-event inside a system without access to its own resolution mechanics. Under high commit-pressure the fastest-threshold route wins whether or not it fits the situation.

Commit-pressure is a dual quantity (Paper 1): the same value that sets how fast a race resolves also sets how deeply the winning route is entrenched. High pressure resolves quickly and entrenches deeply; low pressure resolves slowly and entrenches shallowly. This is why lowering commit-pressure does two things at once. It gives alternative routes room to compete, and it leaves the substrate more responsive to subsequent intervention. Hysteresis is the accumulated record of which routes have won under which conditions, and its deposition is asymmetric: winning deposits more than losing removes. Durable clinical change is therefore slow for a structural reason. Trace is laid per race, so the relevant timescale is the number of races run, not the number of insight-events.

From this follows the paper's hardest structural claim, the additive-only intervention asymmetry: one can add to the substrate but cannot subtract from it. Once commit-pressure has crossed threshold and a route is resolving, no intervention can reach into the race and abort it. No intervention removes trace from an established route by subtraction. Trace is shallowed only by adding competing trace through experience on alternative routes, or by adding destabilisation-window interventions. Effective interventions are accordingly all additive: cofactor repletion adds substrate-availability, an SSRI dose adds synaptic transmitter, reconsolidation work adds new experience, behavioural activation adds competing routes, and environmental design adds the absence of the cue. Strategies framed as stop the urge, remove the trace, or abort the relapse are barred by the architecture; strategies framed as lower the baseline, add competing routes, prevent the cue-encounter, or add new experience are available. The paper reads this as the mechanism behind the effect-size gap between willpower-focused intervention and environmental-design-plus-competing-route intervention.

At the multi-scale level the framework imports Metabolic Control Analysis (Kacser & Burns 1973). Its summation theorem states that flux-control coefficients across a pathway sum to one: control is distributed, not localised at a single enzyme. Single-target intervention is therefore bounded by its target's control coefficient, and compound interventions across coupled scales can produce effects single-target analysis cannot predict. The paper presents this as a research-program claim about psychiatric pathways (one-carbon, monoamine biosynthesis, BH4-pteridine, glutamate-glutamine, kynurenine), supported by the structural similarity to pathways where MCA is empirically validated and by the observation that combination therapy systematically outperforms monotherapy.

Seven clinical phenomena, one machinery

Each phenomenon is treated as a parameter-configuration of the same mechanism. What differs is which routes carry deep hysteresis, what drives the arousal that pushes commit-pressure across threshold, and which cross-layer loops sustain the state.

• Addiction — a route with massive accumulated hysteresis. Under cue, stress, or withdrawal, arousal raises commit-pressure and the race resolves before a slower abstinence-route can win. Willpower is itself a competing route, so it loses under high pressure. Treatment order: lower basal pressure first, then accumulate alternative-route experience. GLP-1 receptor-agonism (semaglutide, Aranäs et al. 2023; exenatide, Klausen et al. 2022) is cited as a mechanism-matched instance.
• Rumination — not simply deep negative trace but a SPREAD-attractor: many interpretive routes active at once, no clear winner, commit failing to activate while hysteresis keeps depositing on non-committed routes. Treatments that feed analysis strengthen the attractor; loop-interrupting treatments (behavioural activation, ACT, MBCT, rumination-focused CBT; Watkins 2018) are framework-matched.
• ADHD — constitutional sensitivity to base-pressure across multiple substrate-axes (elevated β or chronically elevated P_commit,base), with the multi-axis biomarker pattern (ferritin, B12, folate, DHA) as anchor. Untreated substrate runs daily races that deposit compensating-route trace, which the paper proposes as part of the well-documented elevated later substance-use risk (Chang et al. 2014). Rucklidge et al. (2018) multi-nutrient trial (47% vs 28% placebo) is a compatible anchor.
• Autism-spectrum presentations — a constellation of substrate-configurational features (low β-threshold for some sensory modalities, narrow route-tuning, reduced route-shifting under arousal), with one-carbon and BH4 pathways weighted more heavily than catecholamine.
• PTSD and OCD — post-commit hysteresis dysregulation. Trauma-trace is encoded, retrievable, and silo-isolated; reconsolidation-augmented therapy works by valence-flip on existing trace, not by erasing it, which is mechanically impossible.
• Severe specific phobias — a subset (acrophobia is the cleanest case) read as predominantly-inherited substrate-configuration rather than trauma-deposited trace, on grounds of response-speed, familial clustering, and trigger-specificity.
• Aging, expertise, and dementia — substrate-decay is uniform, but trace-renewal is differential and active. Domains still engaged retain depth; unused domains decay because renewal stops. Dementia is read as failure of the renewal mechanism itself.

Two cross-cutting claims sit alongside these. Observed behaviour is a won race, not a direct read-out of substrate-configuration, which is the framework's account of masking in late-diagnosed adults and of midlife collapse after years of expensive compensation. And distribution-shape is a fourth diagnostic dimension: different phenomena are predicted to leave different distributional signatures in commit-readiness and hysteresis-depth, information that a dimensional mean alone misses.

Intervention-precision: matching to mechanism

The framework's position is that every intervention, pharmacological or behavioural or somatic or environmental or relational, works by changing a condition under which race-resolution computes, not by directly causing the outcome. An intervention matches a state when the condition it changes is the rate-limiting one for that state. The dominant clinical taxonomy organises intervention by what is added, with lifestyle and supportive measures positioned as adjuncts. The paper re-organises along the mechanism-axis into six types:

• Add-something direct — adds neurochemical activity that directly reduces friction in a specific race (acute SSRI, ketamine, benzodiazepine for acute anxiety).
• Add-something phased — offers alternative routes that compete with and dilute the dominant route through accumulated experience (behavioural activation, exposure, structured skill-acquisition).
• Reduce-friction direct — reduces the hysteresis-trace or the reactance component that sustains the pattern (CBT trace-destabilisation, reconsolidation-window protocols, stipulation-framed reframing).
• Reduce-friction baseline — reduces substrate-level friction that propagates upward (cofactor repletion, sleep adequacy, physical activity, nutritional psychiatry).
• Dilution-based — provides competing routes that dilute the dominant route in moment-by-moment resolution (social activity, hobby engagement, peer-recovery groups).
• Race-opening prevention — configures the substrate-environment-stimulus complex so the race never initiates, since intervention after threshold-crossing is mechanically too late (stimulus-control therapy, habit-formation manoeuvres, structured commitment).

Two further constructs do operational work. Trust is treated as a structurally-defined meta-variable, not a soft factor: it determines whether external framing is integrated by the substrate or rejected as pressure that triggers reactance. Low-trust intervention is read as actively harmful, because reactance hardens the current state. The paper also flags a cross-substrate caution: a trust-framing test on an instruct-model showed no modulation, so current AI-coach implementations on standard instruct-models are not framework-matched substrates for trust-mediated intervention. Outcome-cluster versus route-identity is the deeper diagnostic point: two patients with identical RDoC or HiTOP profiles may have reached them through different substrate-routes (iron-limited tyrosine hydroxylase, BH4-deficient synthesis, or a compensating behavioural-strategy substrate), and the matched intervention differs in each case. The framework places itself beneath the categorical (DSM, ICD) and dimensional (RDoC, HiTOP) levels, as a mechanistic account of how race-resolution produces the clusters those levels describe. It does not propose to replace them.

Empirical predictions and falsification

The paper states twenty-eight predictions plus five negative-prediction falsification cases, each with a falsification criterion, organised by feasibility into immediate cohort-testable, stratification-requiring, and prospective tiers. Several are testable on existing public-access cohort data (UK Biobank, MIDUS, ABCD, PGC summary statistics, Whitehall II, Rotterdam Study) without new collection. The author identifies five pivotal predictions whose combined falsification would require framework-revision:

• P7.7, UK Biobank-testable — MTHFR C677T carriers with low baseline B12 or folate or high homocysteine show elevated ten-year depression incidence relative to sufficient-baseline carriers and to non-carriers. The claim is an interaction, not a main effect of gene or cofactor alone. Falsified if the interaction term is null.
• P7.5 — cognitive-only interventions in high P_commit,base populations (acute addiction, severe PTSD, active psychosis, severe depression) fail at near-placebo rates when no concurrent pressure-reduction is provided. Falsified if effect-size is invariant to baseline pressure.
• P7.14 — ADHD-population comorbidity for non-circular conditions (restless legs, migraine, sleep fragmentation) is ferritin-stratified. Falsified if no biomarker-stratification appears.
• P7.16, MIDUS-testable — Big Five trait-vector plus biomarker-vector improves physiological stress-response prediction over trait alone. Falsified if the biomarker overlay adds no variance.
• P7.17 — ADHD severity is better explained by an ADHD-PRS × cofactor-status × Big Five three-way interaction than by any single predictor. Falsified if the three-way interaction is null.

The five negative-prediction cases (R1 to R5) state outcomes the framework itself predicts, including that cofactor-repletion in already-replete populations produces no effect. The MoodFOOD null (Bot et al. 2019, JAMA) is the named anchor for that case, read not as evidence against the intervention class but as the signature of unstratified single-axis trials. The author concedes the falsification asymmetry of a many-prediction framework directly, and addresses it by naming the pivotal subset rather than claiming it is eliminated.

Cross-substrate parallel

The paper draws a structural parallel to RLHF-trained language models, stated as structural isomorphism, not mechanism identity. Synaptic plasticity, chromatin modification, and transformer attention are different substrate-mechanisms that are not interchangeable; what is shared is the relational form under the RACE axioms. The RLHF paradox (Paper 1) is the anchor: reinforcement learning from human feedback pushes the commit-point past the thermodynamic optimum, producing rigidity, suppressed uncertainty-expression, and confidently-wrong outputs. The empirical handle is a recognition-commit slope falling from 1.726 in a base model to 0.249 in its instruct counterpart on Llama-405B. The mapping is that a high-baseline-pressure substrate resolves fast to deeply-trained routes while correctly-calibrated alternatives cannot compete, presenting as confident-and-wrong in models or stuck-and-stereotyped in clinical populations. The point is methodological: predictions verified cheaply on the LLM substrate complement slow biological verification, while the substrate-specific scope of clinical recommendations limits any direct clinical transfer.

What this paper is, and is not

What it is: an integration paper. It unifies established findings under one substrate-computational mechanism, supplies a formal extension (arousal-modulated commit-pressure), and commits to twenty-eight falsifiable predictions with explicit criteria, several of them runnable today on existing cohorts.

What it is not: it is not a diagnostic manual, not a treatment protocol, and not a source of patient-level recommendations. It does not present new data; every empirical claim rests on previously published findings re-read under the proposed integration.

Limitations the paper states plainly. The author is an independent researcher without clinical credentials, and the manuscript was drafted without a clinical co-author; clinical collaboration is invited explicitly. The genetic and cofactor interaction claims (P7.7, P7.17, P7.18 in particular) are framed as falsifiable predictions, not established mechanism. The mathematics is conceptual notation, a deliberate simplification at the framework-introduction stage, with an extension pathway named through existing biochemistry models rather than presented as a calibrated model. A prevention-treatment asymmetry follows directly from the hysteresis mechanism and bounds the paper's treatment-claims: laying trace is faster than removing it, so prevention is mechanistically privileged and treatment-effectiveness is bounded by hysteresis-depth at the time of intervention. Finally, the intervention-precision principle requires substrate-vector measurement infrastructure that does not yet exist at clinical scale, which the paper frames as a parallel research-programme rather than a present capability.

Read the paper

The full paper is on Zenodo (concept DOI 10.5281/zenodo.20059865):

Read on Zenodo → · Plain English version · Dansk version

Related on this site

• Paper 8B (Compound Race Pathology) — the somatic compound-disease companion; carries the cross-scale compound-mechanism analysis. Concept DOI 10.5281/zenodo.20059869.
• Paper 8C (A Translational Research-Program for Compound Race Pathology) — trial-design templates and the six-axis substrate-vector biomarker panel. Concept DOI 10.5281/zenodo.20059871.
• Paper 8D (Treating the Base, Not the Top) — the diagnosis-independent sub-threshold population and the base-level cofactor-support hypothesis. Concept DOI 10.5281/zenodo.20277290.
• Paper 1 (Friction Theory) — the substrate-universal framework: RACE architecture, the commit-pressure dual, hysteresis, the RLHF paradox.
• Learning — the learning-and-memory entry point for how trace is laid and renewed.