Karuna Labs
A Clinical Handbook · v1.0
For pain clinicians · therapists · engineers

The Path of Purification.

Reading the Visuddhimagga as a predictive-processing model for the treatment of chronic pain & psychophysiologic disorders.

A practitioner-facing translation manual — integrating psychotherapy, cognitive neuroscience, active inference, and applied AI. For pain psychologists, physiotherapists with a pain-neuroscience orientation, psychiatrists, primary-care & rehabilitation physicians, and the engineers building the digital tools that accompany care.

VIRTUE CONCENTRATION UNDERSTANDING CESSATION
Karuna Labs  ·  A practitioner-facing translation manual
Source text: Buddhaghosa, Visuddhimagga
trans. Bhikkhu Ñāṇamoli
Buddhist Publication Society · rev. ed. 2011
Prologue

How to use this handbook.

This handbook does one thing: it takes a fifth-century systematic treatise on the architecture of mind — Buddhaghosa's Visuddhimagga, the Path of Purification — and reads it as if it were an early, remarkably precise functional model of the predictive brain.

It then uses that reading to build a concrete clinical protocol for chronic pain and related psychophysiologic disorders. It is written for clinicians: pain psychologists, physiotherapists with a pain-neuroscience orientation, psychiatrists, primary-care and rehabilitation physicians, and the engineers building the digital tools that increasingly accompany care.

No Pali, no prior contemplative training, and no commitment to any religious framework is assumed or required. The Visuddhimagga is used here strictly as a phenomenological and procedural source — a detailed, first-person map of how perception, feeling, attention, and the sense of self are constructed and can be deliberately reorganized. We translate that map into the vocabulary of predictive coding and active inference, and we anchor every clinical claim to the contemporary peer-reviewed literature.

A four-color reading

Throughout, the same phenomenon is described four ways. Watch for the colored bars on the left margin of each panel:

Deep plum
Source
The Visuddhimagga's own language and categories.
Teal
Translation
The neurocognitive / computational translation — predictive coding, active inference, interoception.
Clay
Clinical
What you actually do with a patient in the room.
Gold
Practice
Concrete practice protocols — and plum-bordered AI panels mark where computational tools fit.
Scope & safety

This model applies to primary / nociplastic / centralized pain and functional somatic disorders — pain and bodily symptoms that are real and generated by the nervous system but are not explained by ongoing tissue damage or a dangerous secondary condition. It is not a substitute for medical diagnosis.

Appropriate medical assessment to exclude structural disease, infection, malignancy, inflammatory arthropathy, fracture, and other "red-flag" conditions must precede any psychophysiologic formulation. The reattribution of symptoms to brain-based processes is a clinical conclusion reached after that work-up, not an assumption made in place of it.

Contents

A map of the argument.

Part IWhy a fifth-century manual belongs in a pain clinic
  1. 1.1The problem the model has to solve
  2. 1.2The architecture of the source text
Part VIIICautions, boundaries & honest limits
  1. Seven cautions for honest practice
IPart One

Why a fifth-century manual belongs in a pain clinic.

Two thirds of patients with chronic back pain in a rigorous randomized trial became pain-free or nearly so after a four-week psychological treatment whose entire mechanism was changing what the brain predicted about the body. That is not a result a tissue-damage model can absorb.

1.1 The problem to solve

Chronic pain is the paradigmatic disorder of prediction.

In primary or nociplastic pain — chronic back pain without a structural lesion, fibromyalgia, irritable bowel syndrome, chronic pelvic and urogenital pain, persistent post-concussive and post-viral symptoms, and much of what was once called "medically unexplained" — the felt intensity of suffering is poorly explained by tissue state and well explained by the brain's expectations.

Neurocognitive translation · the clinical anomaly

Pain intensity tracks predicted threat more reliably than it tracks nociceptive input. Expectation changes perception; uncertainty about the expectation changes how much.

This is the single empirical fact around which the whole model is organized.

The dominant scientific framework for understanding this is predictive processing — the proposal that the brain is not a passive receiver of sensation but an active, hierarchical inference engine that is constantly generating predictions about the causes of its sensory input and updating those predictions only to the extent that incoming signals are surprising and trusted.

Pain, on this account, is not a readout of damage. It is a perceptual inference: the brain's best guess about how much protective action the body needs, assembled from prior expectation, context, and the precision (the estimated reliability) it assigns to bodily signals.

The Visuddhimagga is, among other things, an extraordinarily detailed operating manual for exactly this inference engine — written from the inside, by people whose full-time occupation was observing how experience is constructed moment to moment and learning to intervene in that construction. Where modern neuroscience describes the predictive hierarchy from the outside (lesions, oscillations, fMRI), the contemplative tradition mapped it from the inside (the sequence by which a bare sensation becomes a felt object becomes a story about a threatened self) — and, crucially, catalogued the trainable operations that loosen each stage. The two maps are of the same territory.

1.2 Architecture of the source text

A threefold training that is, read functionally, a complete clinical arc.

Buddhaghosa organized the entire treatise around a threefold training that, read functionally, is a complete arc of clinical rehabilitation:

Visuddhimagga divisionLiteral subjectFunctional / clinical reading
Part I — SīlaVirtue
Ethical restraint; "guarding the sense-doors"; sufficiency and regularity of life.
Establishing physiological safety, behavioral regularity, and a non-threatening relationship to sensation — the precondition for any learning.
Part II — SamādhiConcentration
Stabilizing attention on a chosen object; the jhānas; overcoming the five hindrances.
Training attentional precision-control: the capacity to deploy and withdraw attention from bodily signals at will.
Part III — PaññāUnderstanding
Insight into the constructed, impermanent, ownerless nature of experience; the seven purifications.
Restructuring the deep priors that generate symptoms — reattribution, model reduction, and the dissolution of the threatened-self narrative.
Clinical application · the clinical arc in one sentence

First make the system safe enough to learn (Sīla → autonomic regulation and behavioral activation); then make attention controllable (Samādhi → precision training); then change what the system believes (Paññā → reattribution and prior revision).

This is the same order, for the same reasons, that the contemplative path prescribes.

"
Modern pain neuroscience arrived, by an entirely different road, at the same destination: that suffering is amplified by prediction, and that prediction can be retrained.
— The thesis of the handbook
IIPart Two

The predictive-processing core model.

Five principles from predictive processing and active inference, each stated for clinicians and matched to its contemplative antecedent in the Visuddhimagga.

P1 Principle One

Perception is controlled hallucination, corrected by error.

The brain predicts the most likely cause of its sensory input and treats sensation primarily as feedback that confirms or corrects the prediction. What we perceive is the prediction, adjusted by prediction error. Pain is generated this way: a top-down inference about bodily threat, not a faithful transmission of nociception.

Source · Visuddhimagga I

"He stops at what is merely seen … As soon as the colour basis has been apprehended by the consciousnesses of the cognitive series … he stops; he does not fancy any aspect … beyond that. … For apart from that there is in the ultimate sense no such thing as a hand and so on."

— Visuddhimagga I, on guarding the sense-doors (citing Ud 8; with the ṭīkā commentary)

Neurocognitive translation

The cognitive series (citta-vīthi)

The text's "cognitive series" — the lawful sequence of mental events by which a bare sense-contact is received, investigated, and then elaborated into a recognized, evaluated, emotionally-charged object — is a strikingly exact description of the predictive cascade running up the cortical hierarchy.

"Stopping at the merely seen" is, in computational terms, declining to let lower-level sensory evidence be over-written by high-level interpretive priors: holding perception closer to the data and refusing the elaboration into "particulars" (here, a threatening object; clinically, a catastrophic meaning).

P2 Principle Two

Precision is the brain's gain control — and it is attention.

Predictions and prediction errors are weighted by their estimated reliability, or precision. Attention is the mechanism that sets this weighting: attending to a signal raises its precision and lets it drive learning; withdrawing attention lowers it.

In chronic pain, precision on interoceptive and nociceptive channels becomes pathologically high and inflexible — the system over-trusts bodily threat signals and cannot turn the gain down. This is mechanistically what hypervigilance is.

Source · Visuddhimagga III

"Concentration is … the centring of consciousness and consciousness-concomitants evenly and rightly on a single object … the state in virtue of which consciousness … remains evenly and rightly on a single object, undistracted and unscattered."

— Visuddhimagga III, on the nature of concentration (samādhi)

Neurocognitive translation

Concentration = trainable precision control

The entire technology of samādhi — selecting a single object, returning to it, stabilizing it — is, in predictive-processing terms, training voluntary control over precision allocation.

The clinical target is not relaxation; it is restoring the ability to flexibly raise and lower the gain on bodily signals at will, which is precisely the capacity chronic pain erodes.

P3 Principle Three

The body is inferred, not read.

Feelings and bodily sensations are not direct reports from the viscera; they are predictions about the body's internal state, issued to regulate it (allostasis) and only corrected by interoceptive prediction error. Emotion, on this view, is the brain's categorization of predicted bodily change.

This is why unprocessed emotion and bodily symptoms are not separable problems: both are interoceptive inferences drawing on the same machinery, and an avoided emotion can be re-inferred as a bodily symptom.

Source · Visuddhimagga XIV

"Feeling is the chief here … [it] is to be regarded as the bare experiencing of the taste of an object … painful, pleasant, and neither-painful-nor-pleasant."

— Visuddhimagga XIV, on the feeling aggregate (vedanā)

Neurocognitive translation

Vedanā as the affective tag on every inference

The text isolates vedanā — the immediate pleasant / unpleasant / neutral valence stamped on every moment of experience — as a distinct, observable component that precedes and seeds reaction.

This is the affective-valence dimension of interoceptive inference — the hedonic prior that, in chronic pain, has collapsed toward "unpleasant-and-dangerous" as a near-permanent setting.

P4 Principle Four

Action is inference too — the second dart.

The brain can minimize prediction error in two ways: change the prediction to fit the world (perceptual inference), or change the world to fit the prediction (active inference).

Guarding, bracing, avoiding, scanning the body, seeking reassurance — these are active-inference behaviors that temporarily reduce the error but confirm the catastrophic prior, raising its precision and deepening the disorder.

The Buddhist tradition's "second dart" — the suffering we add to unavoidable pain through our reaction to it — is exactly this self-confirming loop.

Neurocognitive translation

The two-darts loop, computationally

First dart: the prediction-error signal we experience as raw unpleasant sensation.

Second dart: the cascade of aversion, fear, guarding, and meaning-making that the system mounts in response.

Because guarding and avoidance are themselves actions taken to fulfill the prediction "this is dangerous," they provide the brain with evidence for that very prediction. The loop is closed and self-reinforcing — and it is the loop, not the first dart, that maintains chronic suffering.

P5 Principle Five

Deep priors can be revised — and the self is one of them.

The highest levels of the hierarchy hold the most abstract, slowest-changing priors — including the model of a continuous, bounded self who owns the body and is threatened by its sensations.

Deconstructive contemplative practice progressively reduces this temporally-deep, self-referential processing, and current predictive-processing accounts of meditation describe its endpoint as Bayesian model reduction: the pruning of over-elaborated, unnecessary priors so that experience is met with less anticipatory overlay.

Clinically, we do not need the full deconstruction; we need a targeted version of the same operation — demoting the prior "my body is damaged and dangerous" and dissolving the rigid identification with the pain.

Source · Visuddhimagga XIX

"… mentality-materiality … is void of self or of anything belonging to self … there is no doer of a deed or one who reaps the deed's result; it is bare phenomena that roll on."

— Visuddhimagga XIX, paraphrasing the purification by overcoming doubt

Neurocognitive translation

Model reduction and the predicted self

The insight stages of the Visuddhimagga systematically dismantle the prior of an owned, solid self standing behind experience. In predictive terms this is the relinquishing of the deepest, most abstract self-model — the prior that turns "there is an unpleasant sensation" into "I am in danger and it is happening to me."

For pain, demoting that prior is therapeutic in itself: it removes the high-precision threat context that amplified the signal.

IIIPart Three

The master crosswalk.

A term-by-term dictionary translating the Visuddhimagga's categories into computational and clinical language. Everything that follows is built on it.

3.1 Core construct crosswalk

From the Pāli — to the algorithm — to the clinic.

Visuddhimagga constructPredictive-processing / AI translationClinical target in chronic pain
Citta-vīthicognitive series
The hierarchical predictive cascade from sensory evidence to elaborated, categorized percept.
Interrupting automatic elaboration of sensation into catastrophe.
"Stopping at the merely seen"Ud 8
Holding perception near the sensory data; not over-writing it with high-level interpretive priors.
Sensory de-fusion: experiencing a sensation as sensation, stripped of meaning.
Samādhiconcentration
Voluntary control of precision / attentional gain on a chosen channel.
Restoring flexible up- and down-regulation of attention to the body.
The five hindrancesnīvaraṇa
Characteristic failures of precision-weighting that derail stable inference.
The specific maintaining processes to detect and defuse (see 3.2).
Vedanāfeeling-tone
The affective-valence dimension of interoceptive inference (the hedonic prior).
Decoupling raw valence from the reactive cascade it seeds.
The two darts & papañcaproliferation
Active inference: error-minimizing actions and conceptual elaboration that confirm the threat prior.
Identifying and dropping guarding, avoidance, and catastrophic narrative.
Yoniso manasikārawise attention
Adaptive reallocation of precision; choosing which errors to trust.
Trained meta-awareness of where attention is being captured.
The vipallāsasperceptual distortions
Maladaptive high-precision priors overriding evidence (e.g. "permanent," "mine," "dangerous").
The specific false beliefs reattribution must revise.
Anicca / anattāimpermanence / not-self
Empirical observation that percepts are constructed, transient, and ownerless; demotion of the self-prior.
Dissolving the fixed, threatened-self context that amplifies pain.
The seven purificationssatta-visuddhi
A staged curriculum of progressive prior revision and model reduction.
The treatment protocol's stage structure (Part V).
3.2 The five hindrances

Recognizable failure modes of precision allocation.

Buddhaghosa devotes extended analysis to the five hindrances (nīvaraṇa) — the predictable ways the mind loses its footing during attentional training. Read computationally, each is a recognizable failure mode of precision allocation, and each maps onto a maintaining process clinicians already track in chronic pain. The text also supplies antidotes; these become micro-interventions.

Hindrance Precision-weighting failure Face in the pain clinic Source-derived antidote
Sense-desirekāmacchanda
Precision captured by appetitive predictions; gain pulled toward craved relief.
Reassurance-seeking, medication focus, "fix-it" urgency.
Attend to the unsatisfactoriness of the craved object; redirect to the chosen anchor.
Ill-willvyāpāda
Excess precision on threat / aversive channel; defensive over-weighting.
Pain-related fear, frustration, self-directed anger, bracing.
Cultivated goodwill (mettā) toward the sensation and the body; soften resistance.
Sloth-and-torporthīna-middha
Collapsed precision; gain too low to sustain inference; disengagement.
Fatigue, fog, behavioral shutdown, "giving up."
Energize attention: posture, light, brief movement, clearer object.
Restlessness-and-worryuddhacca-kukkucca
Unstable, oscillating precision; runaway prediction error; rumination.
Anxiety, racing "what-ifs," inability to settle, catastrophizing.
Stabilize on a single simple object (e.g. breath length); structured grounding.
Doubtvicikicchā
Failure to commit to a generative model; corrosive meta-uncertainty.
"Nothing will work," mistrust of the brain-based explanation, drop-out risk.
Provisional commitment to the method; gather one's own evidence experientially.
Clinical application · using the hindrance map in session

When a practice stalls, name the hindrance with the patient rather than treating the stall as failure. "That sounds like restlessness-worry — the system's gain is unstable right now, so we stabilize before we do anything else."

This does three things at once: it normalizes the obstacle, it teaches the precision concept experientially, and it supplies an immediate, source-validated antidote.

IVPart Four

The psychotherapeutic layer.

Two psychotherapies with RCT support — Pain Reprocessing Therapy (PRT) and Emotional Awareness and Expression Therapy (EAET) — already enact large parts of the Visuddhimagga's arc, whether or not their developers intended the parallel.

4.1 Reattribution

Revising the highest prior.

The single most important therapeutic move in primary chronic pain is reattribution — shifting the patient's explanation of the pain from "damage in my body" to "a learned, brain-generated, and therefore reversible signal."

In a secondary analysis of the PRT trial, the degree to which patients re-attributed their pain to mind- or brain-based processes predicted recovery. In predictive terms, reattribution lowers the precision of the catastrophic structural prior and reframes the sensory evidence as safe — which is exactly what is needed for the threat inference to relax.

Clinical application · reattribution and "stopping at the merely seen"

Reattribution is the cognitive entry point; "stopping at the merely seen" is its experiential completion.

The clinic teaches the patient to interrupt the cognitive series at the moment a bare sensation is about to be elaborated into "damage → danger → catastrophe," and to rest attention on the sensation as a constructed, changing, ownerless event.

Done repeatedly, this is not distraction; it is supplying the brain with disconfirming evidence against the threat prior.

4.2 Somatic tracking

Graded interoceptive exposure under safety.

PRT's central technique, somatic tracking, asks the patient to attend to the feared bodily sensation with curiosity and a felt sense of safety, neither avoiding it nor reacting to it.

This is graded exposure aimed at the prediction rather than the stimulus: by attending to the sensation while the predicted catastrophe fails to occur, the brain accumulates prediction error against the danger model and the precision on the threat channel falls. The contemplative source supplies the missing skill that makes this possible — the capacity, trained through concentration, to hold steady, precise, non-reactive attention on an unpleasant object.

Somatic-tracking elementVisuddhimagga correlateWhat it accomplishes computationally
Attend to the sensation directly
Directing attention to the chosen object.
Allocates precision to interoceptive error so it can be re-learned.
With curiosity, exploring its qualities
Investigating the object's changing features.
Reveals the percept as constructed and impermanent (anicca) → weakens "fixed/permanent" prior.
From a place of safety
Sīla: the safe, non-threatening base.
Lowers the threat context so error updates the prior toward "safe."
Without trying to change it
Non-aversion; dropping the second dart.
Removes the active-inference guarding that confirms danger.
4.3 Emotional processing

Re-inferring avoided affect.

EAET starts from the observation that primary and centralized pain is associated with avoided emotion, unresolved trauma, and interpersonal conflict, and that processing those experiences reduces symptoms. The cluster-randomized trial in fibromyalgia found EAET outperformed an active education control and, on several outcomes, the cognitive-behavioral standard of care.

The interoceptive-inference account explains the mechanism: avoided emotions and bodily symptoms are built from the same predictive machinery, so an unfelt emotion can be re-inferred by the brain as a somatic symptom.

Allowing the emotion to be felt, named, and expressed gives the system the interoceptive evidence to categorize the bodily state correctly — as grief, fear, or anger rather than as pain.

Neurocognitive translation

Vedanā, emotion, and the somatic re-inference

The Visuddhimagga's fine-grained attention to vedanā — the bare feeling-tone that precedes the named emotion — gives the clinician a tool EAET can use directly: training the patient to detect the raw valence of an experience before it is interpreted lets avoided affect be met at the level where the somatic mis-inference begins, rather than after it has already been re-coded as pain.

4.4 The therapeutic relationship

The alliance as shared precision.

The clinician is not outside the model. The therapeutic alliance can itself be read as active inference: a clinician's calm, confident, embodied conveyance of safety supplies high-precision top-down priors of safety that the patient's system can borrow before it can generate them alone.

This is the mechanism behind much of what we call rapport, expectancy, and even open-label placebo — and it is why the manner in which reattribution is delivered matters as much as its content. The contemplative tradition's insistence that the teacher embody the stability they teach is the same point.

VPart Five

The staged clinical protocol.

Here the seven purifications of the Visuddhimagga become an explicit, sequenced treatment program. You cannot revise a deep prior in a system whose gain is uncontrolled, and you cannot control gain in a system that does not yet feel safe. This is a spiral, not a staircase.

Stages of the protocol

Seven purifications · seven clinical stages.

Purification (source)Clinical stageComputational goal
0
1 · Purification of virtuesīla-visuddhi
Safety & regularity
Lower baseline threat precision; stabilize allostasis.
1
2 · Purification of mindcitta-visuddhi
Attentional precision training
Restore voluntary precision control.
2
3 · Purification of viewdiṭṭhi-visuddhi
Reattribution
Demote the structural-damage prior.
3
4 · Overcoming doubtkaṅkhāvitaraṇa
Consolidating the new model
Stabilize the brain-based model; resolve meta-uncertainty.
4
5 · Knowledge of path & not-pathmaggāmagga
Discriminating helpful from unhelpful responses
Distinguish adaptive coping from pleasant-but-confirming traps.
5
6 · Knowledge of the waypaṭipadā
Sustained reprocessing & exposure
Repeated disconfirmation; precision on threat channel falls durably.
6
7 · Knowledge & visionñāṇadassana
Restructured experience & relapse-resilience
Stable model reduction; self-prior no longer recruits threat context.
0
Stage 0 · Sīla

Safety & Regularity.

Source · Visuddhimagga I

"Virtue has the characteristic of … composing; its function is … to remove the danger of remorse; … its proximate cause is conscientiousness. The benefits are non-remorse, gladness, happiness, concentration …"

— Visuddhimagga I, on the benefits of virtue

Computational goal

Chronic threat keeps interoceptive and nociceptive precision globally elevated; the system cannot learn anything new while it is braced for danger. Stage 0 lowers baseline threat precision by regularizing the conditions of life — sleep, activity, nourishment, the manner of attending to the body — so that allostatic load falls and the system has the metabolic and predictive slack to learn.

In the room

The medical reassurance the work-up has earned.

Begin with the medical reassurance the work-up has earned: the symptoms are real and the body is not damaged or in danger. Establish behavioral regularity — consistent sleep and wake times, gentle graded activity, regular nourishment — not as lifestyle advice but as the substrate that lowers baseline arousal.

Introduce "guarding the sense-doors" in its clinical form: noticing and reducing the constant body-scanning, symptom-checking, and reassurance-seeking that keep threat precision high.

Clinical application · guarding the sense-doors, translated

The text's instruction to receive sense-contact without immediately fashioning it into "particulars" (elaborated, charged objects) becomes: when you notice the body, stop at the bare sensation — do not let attention spin it into a story about damage. Each time you do, you withhold the active-inference behavior (checking, bracing) that feeds the danger model.

Practice · Home practice

Grounding in the merely felt.

A brief daily practice: two or three minutes of attending to a neutral or pleasant bodily sensation (feet on the floor, breath, warmth of the hands), explicitly practiced as building a safe interoceptive base.

A simple log of body-checking and reassurance-seeking urges — noticed, not necessarily acted on — to make the active-inference loop visible.

1
Stage 1 · Samādhi

Attentional Precision Training.

Source · Visuddhimagga III

"… the centring of consciousness … evenly and rightly on a single object … So that state of … non-distraction, is what should be understood as concentration."

— Visuddhimagga III, defining samādhi

Computational goal

Before the threat prior can be revised it must become possible to direct and withdraw attention from bodily signals at will. Stage 1 trains voluntary precision control: focused attention (raising gain on a chosen anchor and returning to it) builds the capacity, and open monitoring (watching whatever arises without fixing on it) generalizes it. The five hindrances are the expected obstacles; each is met with its mapped antidote (§3.2).

In the room

Teach focused attention first, then open monitoring.

Teach a focused-attention anchor first — breath, sound, or a neutral body region — with the explicit framing that the skill being built is not relaxation but the ability to place and hold attention, notice when it is captured, and return it. The returning is the repetition that trains the control.

As stability grows, introduce open monitoring: letting sensations, thoughts, and feelings arise and pass without pursuit, which trains the flexible lowering of precision.

Use the hindrance map live. When attention destabilizes, name the failure mode (restlessness-worry, sloth-torpor, aversion) and apply its antidote rather than pushing through. This both rescues the practice and teaches precision experientially.

Practice · Home practice

Graded daily practice.

Start at five minutes, build deliberately: focused attention on the anchor, then a short period of open monitoring. The aim is consistency, not duration.

A one-line note afterward on which hindrance showed up trains meta-awareness (yoniso manasikāra) and gives the clinician the data to tailor the next session.

2
Stage 2 · Purification of View

Reattribution.

Source · Visuddhimagga XVIII

"… when he sees correctly … there comes to be purification of view … he sees 'This is mere mentality-materiality; there is no being, no person here.'"

— Visuddhimagga XVIII, purification of view

Computational goal · the decisive intervention

Demoting the high-precision prior that the pain signals tissue damage and danger, and installing in its place the (accurate) generative model that the pain is a learned, brain-produced, reversible signal.

Because it is a high-level prior, it is revised through both explicit understanding (the pain-neuroscience explanation) and accumulated experiential evidence (Stage 1's attention skill now applied to the symptom itself).

In the room

Collaborative evidence-gathering, not assertion.

Walk through the patient's own history for the fingerprints of a brain-generated symptom: onset under stress, spread or migration inconsistent with anatomy, variability with mood, attention, and context, symmetry, and the absence of structural explanation. Each datum is prediction error against the damage model.

Pair the explanation with the embodied safety of the alliance (§4.4) so the new prior is delivered at high precision.

Clinical application · "purification of view" is reattribution completed

The text's purification of view — seeing experience as "mere mentality-materiality," constructed and ownerless — is the deep version of the move the clinic makes when it teaches the patient to see the pain as a constructed signal rather than a fact about a damaged body. The clinical version is targeted and partial; the structure is identical.

Practice · Home practice

Begin somatic tracking.

Short, daily: attend to a tolerable pain sensation from the trained place of safety, with curiosity, exploring its edges and changes, without trying to make it leave.

The patient is instructed to notice, each time, that the predicted catastrophe does not occur — this is the disconfirming evidence that revises the prior.

3
Stage 3 · Overcoming Doubt

Consolidating the New Model.

Source · Visuddhimagga XIX

"… when he has thus discerned … he passes beyond uncertainty … This is called purification by overcoming doubt."

— Visuddhimagga XIX, overcoming doubt

Computational goal

A newly installed prior is fragile; doubt (vicikicchā) — corrosive meta-uncertainty about whether the brain-based model is true — is the predictable threat to it, and a major driver of drop-out and relapse.

Stage 3 stabilizes the generative model the way the source prescribes: not by argument but by accumulated first-person evidence that the patient gathers and credits.

In the room

Make the patient the scientist of their own case.

Review the somatic-tracking and symptom logs for instances where the pain changed with attention, context, emotion, or expectation — each is direct, personally-owned evidence that the symptom is modulable and therefore brain-based.

Address doubt explicitly when it arises; treat it as an expected hindrance, not a verdict. Setbacks are reframed as the system testing the old prior, not as proof the model was wrong.

Practice · Home practice

The evidence journal.

Continue somatic tracking, now with a deliberate evidence journal: each entry records one observation of the symptom varying with a non-structural factor.

Over weeks this becomes a body of self-generated proof that consolidates the new model at high precision.

4
Stage 4 · Path & Not-Path

Discriminating Helpful from Unhelpful Responses.

Source · Visuddhimagga XX

"… he comes to know … 'This is the path, this is not the path' … This is called purification by knowledge and vision of what is the path and what is not the path."

— Visuddhimagga XX, path and not-path

Computational goal · a subtle clinical teaching

At this stage the practitioner encounters experiences that feel like progress — calm, lightness, even bliss — but that can become subtle attachments that stall the work.

The computational reading is that not all error-reduction is therapeutic: some pleasant, reassuring responses (over-reliance on a coping ritual, a relaxation that becomes a new avoidance) reduce error in the moment while quietly confirming that the body still needs managing.

In the room

The directional test.

Help the patient distinguish responses that disconfirm the threat prior (engaging life, approaching feared movement, meeting emotion) from those that merely soothe while confirming it (elaborate symptom-management routines, reassurance rituals, "safe" avoidance dressed as self-care).

The test is directional: does this response move toward the predicted catastrophe and find it absent, or does it organize life around preventing the catastrophe?

Clinical application · pleasant traps are still active inference

A relaxation technique used to flee a sensation is, computationally, the same guarding move as bracing — it just feels better. The source's warning against attachment to pleasant meditative states is the same caution: do not mistake error-reduction for prior-revision.

Practice · Home practice

Graded re-engagement as experiments.

Re-engagement with avoided activities and movements, chosen collaboratively, each framed as an experiment that tests the danger prediction.

The patient tracks which of their own coping responses are approach (path) and which are subtle avoidance (not-path).

5
Stage 5 · Knowledge of the Way

Sustained Reprocessing & Exposure.

Source · Visuddhimagga XXI

"… contemplation of rise and fall … of dissolution … he sees how all formations … are impermanent, painful, not-self … and his mind … retreats, retracts and recoils … [then] becomes dispassionate."

— Visuddhimagga XXI, the insight knowledges of the way

Computational goal

The long middle of recovery: repeated, varied disconfirmation until the precision on the threat channel falls durably and the symptom loses its grip.

The source's detailed "insight knowledges" — the progressive, sometimes turbulent stages in which formations are seen as impermanent, unsatisfactory, and not-self — describe the experiential texture of sustained prior revision, including the predictable phase where things feel worse before dispassion sets in.

In the room

"It got louder before it got quieter."

Sustain and broaden somatic tracking and graded exposure across contexts and activities, increasing variability so the new "safe" prediction generalizes rather than staying tied to the consulting room.

Normalize the difficult phase the source documents: as the old prior is repeatedly violated, patients often report a period of heightened symptom salience or emotional turbulence. Framing this as expected — the system relinquishing an entrenched habit of prediction — prevents it from being read as relapse.

Neurocognitive translation · impermanence as repeated prediction error

The contemplative observation that all sensations visibly arise and pass is, for the pain patient, the most therapeutic observation available: a sensation watched closely is seen to change, and a changing sensation cannot be the fixed evidence of fixed damage.

Every observed fluctuation is prediction error against the "permanent/structural" prior.

Practice · Home practice

Daily practice integrated into ordinary life.

Concentration, somatic tracking, open monitoring of impermanence, and lived behavioral exposure. The home program increasingly looks like ordinary life re-entered, with the skills carried into it rather than practiced apart from it.

6
Stage 6 · Knowledge & Vision

Restructured Experience & Relapse-Resilience.

Source · Visuddhimagga XXII

"… with the fading away of … clinging … the mind is liberated … 'Birth is destroyed … there is no more of this to come.'"

— Visuddhimagga XXII, on the fruition of the path

Computational goal · stable model reduction

The catastrophic threat prior is no longer dominant, the self-model no longer reflexively recruits a threat context around bodily sensation, and the patient meets sensation with low anticipatory overlay.

Recovery is durable not because symptoms can never recur but because the patient now has a different relationship to the machinery that produces them — they can recognize and decline the second dart.

In the room

The patient leaves as their own clinician.

Consolidate gains into an explicit relapse-resilience plan built from the model itself: recognize a flare as the old prior being re-tested (often under stress or threat), apply the now-familiar sequence (safety → precision → reattribution → tracking), and decline the active-inference loop of guarding and catastrophizing.

The patient leaves understanding their own nervous system well enough to be its clinician.

Neurocognitive translation · not-self and the dissolved threat context

The deepest source teaching — that there is no solid, separate self who owns and is endangered by the sensations — has a precise clinical payoff even in its mild, secular form: when sensation is met without the prior "this is happening to me and threatens me," the high-precision threat context that amplified it is simply absent. The pain, deprived of its amplifying frame, subsides.

Practice · Home practice

A maintenance practice the patient owns.

A short daily period of attention training and open monitoring, plus the explicit relapse-resilience protocol kept available for flares. Booster sessions as needed.

VIPart Six

The AI & computational layer.

Predictive processing is not only a theory of the brain; it is a family of algorithms, which makes it a natural bridge to the digital tools increasingly used in pain care. AI tools are instruments that support a clinical relationship — not replacements for it.

6.1 Bayesian foundations

The brain as a Bayesian inference machine — and what that licenses.

If pain is perceptual inference — prior, likelihood, precision, posterior — then each term is, in principle, a measurable, modelable quantity. Computational models of pain already formalize evoked and spontaneous pain as predictive-coding processes, and individual differences in how strongly expectation shapes pain have been captured in explicit Bayesian models.

This is what makes a digital layer more than a gadget: the same mathematics describes the disorder, the therapy, and the tool.

AI / computational layer · mapping the model to system functions

Where each term lives.

Prior → the patient's beliefs and expectations about their pain, the target of reattribution.

Likelihood → the incoming sensory and contextual evidence.

Precision → attention and the trust placed on bodily signals, the target of concentration training.

Posterior → the felt experience.

A well-designed digital tool intervenes on the prior (psychoeducation, reattribution support) and on precision (attention training, biofeedback), and measures the posterior (symptom report) and its modulators.

6.2 Tools that earn their place

Where computational tools earn their place.

Tool classModel targetSource / clinical correlate
Conversational psychoeducation & reattribution support LLM-based
Lowers precision of the structural-damage prior; available between sessions.
Stage 2 reattribution; repeated explanation as the alliance extends.
Guided attention / somatic-tracking apps adaptive pacing
Trains voluntary precision control; scaffolds Stage 1 and Stage 5 practice.
Samādhi training; somatic tracking.
Interoceptive biofeedback HRV, respiration
Makes interoceptive precision visible and trainable; closes the active-inference loop adaptively.
Purification of mind; allostatic regulation.
Real-time neurofeedback e.g. targeting cingulate activity
Trains down-regulation of the affective pain network; demonstrated to reduce pain ratings.
Direct precision control over the threat channel.
Ecological momentary assessment EMA
Measures the posterior and its modulators (mood, context, attention) → fuels the Stage 3 evidence journal.
Overcoming doubt through accumulated personal data.
Immersive VR exposure
Graded disconfirmation of movement-danger predictions in a controlled, varied environment.
Stage 4–5 exposure; generalizing the safe prediction.
AI / computational layer · the genuine opportunity

Closed-loop personalization.

The deepest contribution of the computational layer is closed-loop adaptivity: a system that measures the patient's state (EMA, biofeedback), infers which stage and which hindrance is active, and adjusts what it offers — stabilizing when precision is unstable, exposing when the system is safe enough to learn.

This is the digital instantiation of the clinician's moment-to-moment titration, and it is where AI is most defensible: not generating insight, but extending the reach and responsiveness of a model the clinician owns.

6.3 Limits, risks & the second-dart problem for machines

When the tool becomes the not-path.

A predictive-processing lens also sharpens the cautions. A conversational agent that over-reassures can become a reassurance ritual — an active-inference loop that confirms the body needs constant monitoring (the very Stage-4 trap, now automated and available at 3 a.m.).

A biofeedback display can itself become an object of hypervigilance, raising the precision it was meant to lower. Symptom-tracking can shade into symptom-focus.

The design principle that follows directly from the model: digital tools should be built to make themselves progressively unnecessary — to hand precision control back to the patient — not to maximize engagement. Engagement-maximizing design and recovery are, here, in direct tension.

Clinical application · design heuristic

Ask of any feature: does this supply disconfirming evidence against the threat prior and build the patient's own capacity (path), or does it soothe while confirming that the body must be managed (not-path)?

The Visuddhimagga's discrimination of path from not-path is, unexpectedly, a product-design principle for digital therapeutics.

VIIPart Seven

Putting it together.

An illustrative composite case, and a session-by-session skeleton for a 12–16 session course.

7.1 A composite case

An illustrative composite case.

The following is a composite, not a real patient, assembled to show the protocol in motion.

A woman in her forties presents with two years of widespread, migrating musculoskeletal pain and fatigue, extensively investigated, with no structural explanation and a diagnosis of fibromyalgia. Onset followed a period of acute work and family stress. She scans her body constantly, has stopped exercising for fear of "damage," and has begun to doubt that anything will help.

Presentation featureModel readingStage that addresses it
Stress-linked onset, no structural cause, migration, variability
Brain-generated (nociplastic) pain; high-precision threat prior.
Stage 2 — reattribution
Constant body-scanning, reassurance-seeking
Active-inference guarding confirming the danger model.
Stage 0 — guarding the sense-doors
Fear-driven avoidance of exercise
Movement-danger prediction never disconfirmed.
Stages 4–5 — graded exposure
Avoided stress and emotion around onset
Affect re-inferred as somatic symptom.
EAET-style emotional processing (§4.3)
"Nothing will work"
Doubt (vicikicchā) threatening the new model.
Stage 3 — overcoming doubt via personal evidence
7.2 Session skeleton

A session-by-session skeleton.

A scaffold, not a script; pacing follows the patient's precision and safety, and stages recur.

  1. Sessions 1–2 — Assessment, validation, and Stage 0. Confirm the work-up is complete; deliver the "real but not damaged" framing; establish regularity and begin reducing body-scanning and reassurance-seeking.
  2. Sessions 3–4 — Stage 1. Introduce focused attention and the precision concept; begin the hindrance map; brief daily practice.
  3. Sessions 5–7 — Stage 2. Collaborative reattribution from the patient's own history; begin somatic tracking from a place of safety.
  4. Sessions 8–9 — Stage 3 and emotional processing. Build the evidence journal; address doubt; open the emotional / interpersonal material linked to onset.
  5. Sessions 10–12 — Stages 4–5. Discriminate approach from avoidance; graded behavioral exposure as experiments; broaden across contexts; normalize the "louder before quieter" phase.
  6. Sessions 13–16 — Stage 6. Consolidate the restructured relationship to sensation; build the relapse-resilience plan; taper to maintenance and boosters.
One page the patient can keep

The pain is real. It is produced by my nervous system, not by damage, and it can change.

My job is not to fight the sensation but to stop adding the second dart — the fear, the guarding, the story.

Safety first, then steady attention, then meeting the sensation with curiosity from that safe place.

Every time I notice the predicted catastrophe doesn't come, my brain learns the body is safe.

A flare is the old pattern being tested, not a return of damage. I know the sequence. I can come back to it.

VIIIPart Eight

Cautions, boundaries & honest limits.

Seven cautions for the practitioner adopting this framework — what it is, what it is not, and where the practitioner's responsibilities remain undiminished.

  1. Medical first. Psychophysiologic formulation follows, and never replaces, appropriate medical assessment. Re-evaluate if the picture changes or red flags emerge.
  2. Not all chronic pain is nociplastic. This model targets primary / centralized / nociplastic pain and functional somatic disorders. Pain with ongoing nociceptive or neuropathic drivers needs those addressed; the two can coexist.
  3. Reattribution is not blame. "Brain-generated" never means "imagined," "your fault," or "just stress." Mishandled, it ruptures the alliance and raises threat. The manner of delivery is part of the mechanism.
  4. The analogy is a tool, not a doctrine. Reading the Visuddhimagga as a predictive-processing model is a productive heuristic, not a claim that Buddhaghosa anticipated computational neuroscience or that the traditions are identical. The contemplative goal is liberation, not symptom relief; we borrow its phenomenology and methods, respectfully and partially.
  5. Trauma needs competence. Emotional-processing work can surface trauma; do it within your scope or refer. Stabilization (Stage 0–1) precedes processing.
  6. Contemplative practice has adverse effects. Intensive attention training can destabilize some patients (anxiety, dissociation, re-traumatization). Use graded, supported, clinically-paced practice — not retreat-intensity practice — and monitor.
  7. Evidence is strong but still maturing. PRT and EAET have meaningful RCT support; the predictive-processing account of their mechanism is well-motivated but partly theoretical, and the contemplative integration proposed here is a synthesis, not itself a trialed protocol. Hold it as a guiding model to be tested, not settled fact.
A closing note

The Visuddhimagga's title is a promise about purification — the progressive removal of what distorts clear seeing. Read clinically, the distortions it removes are over-trusted priors: the brain's insistence that a safe body is dangerous, that a constructed signal is a structural fact, that a passing sensation belongs permanently to a threatened self.

Modern pain neuroscience arrived, by an entirely different road, at the same destination: that suffering is amplified by prediction, and that prediction can be retrained. This handbook is an attempt to let the two maps inform a single, careful clinical practice.

Bibliography

Selected references.

References are drawn from the peer-reviewed literature located via PubMed and primary journals. Source citations to the Visuddhimagga follow the chapter and section conventions of the Ñāṇamoli translation (Buddhist Publication Society, rev. ed. 2011).

Predictive coding, active inference, and pain

Tabák Z, et al. Why harmless sensations might hurt in individuals with chronic pain: about heightened prediction and perception of pain in the mind. Front Psychol. 2016;7:1638. PMC5078757
Song Y, et al. Hierarchical predictive coding in distributed pain circuits. Front Neural Circuits. 2023;17:1073537. PMID 36937818
Yang Z, et al. Predictive coding models for pain perception (evoked and spontaneous pain; mean-field ACC model). J Comput Neurosci / IEEE. 2021. PMID 33595765 · PMC8025043
Hird EJ, et al. Sensitivity to pain expectations: a Bayesian model of individual differences. Cognition / PubMed. 2019. PMID 30243037
Lersch FE, et al. Analgesia for the Bayesian brain: how predictive coding offers insights into the subjectivity of pain. Curr Pain Headache Rep. 2023. PMID 37421540

Interoception, allostasis, and emotion

Seth AK, Friston KJ. Active interoceptive inference and the emotional brain. Philos Trans R Soc Lond B. 2016;371(1708):20160007. PMID 28080966
Seth AK. Interoceptive inference, emotion, and the embodied self. Trends Cogn Sci. 2013;17(11):565–573. PMID 24126130 (see also PMID 23663284)
Barrett LF, Quigley KS, Hamilton P. An active inference theory of allostasis and interoception in depression. Philos Trans R Soc Lond B. 2016;371(1708):20160011. PMID 28080969
Barrett LF. The theory of constructed emotion: an active inference account of interoception and categorization. Soc Cogn Affect Neurosci. 2017;12(1):1–23. PMID 27798257

Meditation and the predictive mind

Laukkonen RE, Slagter HA. From many to (n)one: meditation and the plasticity of the predictive mind. Neurosci Biobehav Rev. 2021;128:199–217.
Lutz A, Mattout J, Pagnoni G. The epistemic and pragmatic value of non-action: a predictive coding perspective on meditation. Curr Opin Psychol. 2019;28:166–171.

Clinical trials and treatment models

Ashar YK, Gordon A, Schubiner H, et al. Effect of pain reprocessing therapy vs placebo and usual care for patients with chronic back pain: a randomized clinical trial. JAMA Psychiatry. 2022;79(1):13–23. PMID 34586357
Ashar YK, Lumley MA, Perlis RH, et al. Reattribution to mind-brain processes and recovery from chronic back pain: a secondary analysis of a randomized clinical trial. JAMA Netw Open. 2023;6(9):e2333846.
Lumley MA, Schubiner H, Lockhart NA, et al. Emotional awareness and expression therapy, cognitive-behavioral therapy, and education for fibromyalgia: a cluster-randomized controlled trial. Pain. 2017;158(12):2354–2363.
Lumley MA, Schubiner H. Emotional awareness and expression therapy for chronic pain: rationale, principles and techniques, evidence, and critical review. Curr Rheumatol Rep. 2019;21(7):30. PMID 31123837

Primary source

Buddhaghosa, Bhadantācariya. The Path of Purification (Visuddhimagga). Trans. Bhikkhu Ñāṇamoli. Kandy: Buddhist Publication Society; revised online ed., 2011.

"Bare phenomena that roll on."

This handbook is a synthesis, not a trialed protocol. It is offered as a clinically-grounded model to be examined, tested, and refined in practice. Clinical decisions remain the responsibility of qualified practitioners, working with patients whose structural disease has been appropriately excluded.

© Karuna Labs Inc.  ·  The Path of Purification  ·  v1.0