The calm that does not survive interruption

A session can look quiet for at least three different reasons. Attention may be actively held against a pull. The pull may lack its usual cue. Or the disposition that made the cue powerful may have changed. Only the third explanation predicts durable transfer.

Patañjali gives an unusually precise vocabulary for this ambiguity. Yoga Sūtra 2.4 describes the kleśas in four conditions: prasupta, dormant; tanu, attenuated; vicchinna, interrupted; and udāra, active or manifest. Vyāsa’s commentary sharpens the difference. A tanu affliction has been weakened through an opposing cultivation. A vicchinna affliction disappears while another tendency dominates, then reappears in its own conditions.¹

This is not a scale from zero to three. It is a set of causal states. Dormancy concerns availability without expression. Interruption concerns competition. Manifestation concerns current activity. Attenuation concerns a change in the disposition itself. A quiet interval therefore underdetermines what happened. Absence of expression is not evidence of attenuation.

Vyāsa’s example is psychologically sharp. Anger may not appear while attraction occupies the field, yet its absence does not make the person free of anger. The conditions for one tendency have merely displaced the conditions for another. When the relation changes, anger can return. What looks like success in a short observation window may therefore be an alternation among active dispositions rather than a weakening of any of them.¹

This yields three different meanings of “return to the breath”. In the first, the person exerts effort and wins a single contest. In the second, the cue remains active but loses because the session blocks the usual response. In the third, the cue recruits less priority, less elaboration and less action across repeated encounters. All three can produce the same button press. The decisive evidence is not one successful return, but whether the old trigger has less power to make return necessary.

That distinction also prevents a common mistake about effort. High effort is not failure. Early training may require deliberate correction, and the text explicitly values sustained practice. The error is to infer durable change from effortful success without testing what happens when effort is taxed. A perturbation creates that tax in a controlled way. It reveals how much stability belongs to the transformed disposition and how much was borrowed from the protected setting.

The distinction matters because the second chapter does not present concentration as an isolated skill. Kriyā-yoga is introduced as serving two linked purposes: cultivating samādhi and attenuating the kleśas. Later, the limbs are said to reduce impurity while allowing discriminative knowledge to become luminous. The text’s own causal target therefore includes a transformation of what drives cognition, rather than only a longer dwell time on an object.¹

A laboratory can operationalise only a small part of that claim. It can count orientation, dwell, recovery and action. It cannot read liberation from a reaction-time trace. Yet the textual distinction still improves experimental design. It forces the researcher to ask whether a measure captures a temporary state, an acquired skill, a context-dependent strategy or a changed disposition. Those categories are often blurred when a post-training average is treated as a single effect.

For the experiment to work, “capture” needs a modest operational meaning. Here it means an unchosen redirection of processing towards a cue, followed by measurable dwell or a competing action tendency. A tone can capture orientation without producing distress. A thought can capture dwell without moving the eyes. A cue can also trigger a hand movement before a person endorses the goal. This definition is intentionally narrower than kleśa.

flowchart LR
  A["The calm that broke at 6,40"] --> B["The calm that does not survive<br/>interruption"]
  B --> C["Five ways attention is captured"]
  C --> D["Why the limbs intervene at<br/>different depths"]
  D -. "re-test" .-> A
  classDef first fill:#0a1424,stroke:#087A98,stroke-width:2px,color:#e6f3ff
  classDef step fill:#0a1424,stroke:#3f7da0,stroke-width:1px,color:#e6f3ff
  class A first
  class B,C,D step

Five ways attention is captured

The five kleśas are not five kinds of distraction. They describe a deeper organisation in which misappraisal, appropriation, attraction, aversion and clinging help generate the thoughts and actions that repeatedly seize the mind.

Start with avidyā. The text calls it the field for the other four. Its examples are reversals of discernment: taking the impermanent as permanent, the impure as pure, the painful as pleasurable, and the non-self as self. This is not simply missing information. It is an active mis-taking of how things are. In the classical system, the decisive error concerns the relation between puruṣa and the operations of prakṛti.

Asmitā follows as an apparent identification of the power of seeing with the instrument of seeing. Rāga is the residue or following of pleasure. Dveṣa follows pain. Abhiniveśa is a deep clinging to continuity of life that is said to run even in the learned. Each term belongs to a soteriological and metaphysical argument. None should be replaced by “ego”, “reward”, “negative affect” or “survival circuit” as though a translation had been proved.¹

The list is dynamic rather than taxonomic. A misappraisal can make a role seem constitutive of the person. Threat to that role can then recruit attraction, aversion or clinging. The resulting action changes the person’s environment and rehearses the same interpretation. The commentary describes the kleśas as mutually supporting and as sustaining the stream of consequences. The kleśas are not five interruptions; they help explain why interruptions are regenerated.

Consider status checking. The surface event is a glance at a metric. Underneath it may be the belief that the metric offers stable security, an appropriation of professional rank as self, pleasure when the number rises, aversion when it falls, and fear of becoming irrelevant. Removing the app can interrupt the behaviour. It does not by itself revise the organisation that made another metric capable of replacing it. The classical analysis therefore invites a depth question: which dependency was actually altered?

A narrow bridge to attention science

Modern attention research offers a useful, limited question: why does the current goal lose? Awh, Belopolsky and Theeuwes argued that selection cannot be explained by a simple contest between voluntary goals and physical salience. Past selection history contributes to an integrated priority map.³ Anderson, Laurent and Yantis showed that a stimulus previously associated with reward can capture attention later even when it is no longer relevant and is not the most visually salient item.⁴ Habit research likewise finds that responses learned in recurring contexts can be triggered by those contexts despite a changed current intention.⁵

This does not show that reward history is rāga. It shows why a repeated pull can outlive the goal that created it. The bridge preserves one relation: prior interaction changes the probability that a cue will recruit processing later. It fails at ontology, moral significance, first-person character and the classical purpose of liberation. A bounded comparison earns its value by naming what it cannot carry.

The kleśas can guide a different set of questions without becoming modern variables. What has been mistaken as stable or satisfying? Which role or mental operation has been appropriated as “me”? What pleasant outcome is being rehearsed? What pain is being escaped before it arrives? What continuity is treated as non-negotiable? These are inquiry prompts, not a scoring rubric.

Why the limbs intervene at different depths

The eight limbs are often presented as a staircase ending in concentration. A better causal picture is a set of nested interventions. Each limb changes a different part of the loop through which attention is prepared, captured, sustained and interpreted.

Yama and niyama operate before a meditation object is selected. They govern conduct, restraint, observance and the repeated actions through which a life acquires its cues. The text’s instruction of pratipakṣa-bhāvanā, cultivating an opposing orientation when harmful thoughts arise, is especially important because Vyāsa also uses opposing cultivation to explain attenuation. A practitioner who stops repeatedly enacting hostility is more than becoming better at ignoring hostile thoughts during a session. The contexts, rehearsals and consequences that feed those thoughts may change.

Suppose a person practises attention every morning but spends the rest of the day repeatedly opening provocative content, composing imaginary arguments and rewarding each surge with further engagement. The session trains return. The day trains capture. A limb that changes conduct alters the number and kind of learning trials delivered outside formal practice. This is why an ethical restraint can have attentional consequences without being reducible to an attention technique.

This is a practitioner synthesis, not a claim that yama is behaviour therapy. Its point is causal: conduct can alter the training distribution. If a person spends the day amplifying comparison, stimulation and deception, a quiet evening session inherits that history. The practice environment begins long before the cushion.

Āsana supplies a stable and easeful bodily configuration. Prāṇāyāma changes the organisation of breathing after posture is established. Pratyāhāra changes the relation between the senses and their objects. Dhāraṇā binds citta to a place. Dhyāna is the continuity of the same cognition there. Samādhi is described as the object alone becoming manifest, as though the form of the meditator had receded. The three together become saṃyama.²

These definitions are not interchangeable stages of “more focus”. Posture concerns bodily stability. Breath concerns a regulatory process. Withdrawal concerns sensory coupling. Dhāraṇā concerns placement. Dhyāna concerns continuity. Samādhi concerns a transformed relation among knower, knowing and object. The system also retains discriminative knowledge as its deeper horizon. A performance-only reading cuts away precisely what makes the architecture distinctive.

The sequence is ordered, but it is not a disposable ladder. The third chapter calls dhāraṇā, dhyāna and samādhi internal relative to the preceding limbs, then calls even these external relative to seedless samādhi. “Inner” is therefore relational, not a declaration that outer conditions have become irrelevant. A conflict discovered in meditation may require a change in conduct. A bodily instability may make continuity impossible. A sensory habit may keep recreating the object that concentration is asked to defeat.

The limbs are complementary because the capture loop has several entry points. One intervention changes what is repeatedly enacted. Another changes physiological disturbance. Another changes whether a sense cue is pursued. Another changes placement and continuity. The architecture becomes intelligible when it is viewed as coordinated causal depth rather than as eight badges of advancement.

A ninety-second interruption

Return to the notification tone. A layered account does not imagine eight miniature steps firing in order. It asks where the system has already been changed. A conduct-level commitment may have removed non-essential alerts. A stable posture prevents the tone from recruiting a cascade of restless movement. Natural, unforced breathing supplies a non-threatening rhythm. Pratyāhāra-like restraint prevents immediate pursuit of the sound. Dhāraṇā restores the chosen place. Dhyāna concerns whether that return becomes continuous rather than a repeated tug-of-war. Discernment determines whether the imagined message is treated as urgent fact.

What modern experiments can and cannot add

Lutz and colleagues distinguished focused-attention practices from open-monitoring practices, arguing that they train partly different regulatory skills.⁶ Jha and colleagues found that different training formats were associated with changes in different attention subsystems rather than one general faculty.⁷ Intensive studies have reported changes in attentional blink, perceptual discrimination and sustained attention, but their dose, retreat setting and task specificity limit transfer to ordinary short practice.⁸⁹

A particularly relevant brief experiment compared monitoring-only instructions with monitoring plus acceptance. Four short sessions reduced mind-wandering more in the combined condition, suggesting that how experience is related to may matter alongside what is monitored.¹⁰ This result does not establish Patañjali’s architecture. It does challenge a concentration-only assumption.

A meta-analysis of 111 randomised controlled trials reported small-to-moderate effects of mindfulness-based interventions on several cognitive outcomes, including executive attention and aspects of working memory, inhibition and sustained attention. It did not find reliable benefits for every outcome, and the underlying interventions and controls varied.¹¹ Reviews comparing mindfulness programmes with active controls likewise find smaller and less consistently significant effects than comparisons with passive controls.¹²

Construct validity is the deeper problem. “Mindfulness” can refer to a state report, a trait questionnaire, a family of practices, a clinical programme or an ethical orientation. “Attention” can refer to orienting, alerting, conflict monitoring, sustained detection, working-memory filtering or subjective mind-wandering. A gain in one pairing does not travel automatically to another. Critical reviews have therefore called for tighter definitions, active controls, better adverse-event reporting and less reliance on broad labels.¹⁵

The rebound design responds to that problem by choosing one event, one perturbation and one transfer claim. It does not ask whether a person became more mindful. It asks whether a specified cue recruited less processing after support was challenged. This narrower question is less impressive, but more answerable.

stateDiagram-v2
  [*] --> Claim
  state "Figure 1. A training path<br/>through recurrent vortices" as Claim
  state "Figure 2. Two calm sessions,<br/>two mechanisms" as Test
  state "Figure 3. The root is not<br/>another branch" as Revision
  Claim --> Test: expose mechanism
  Test --> Decision: evidence survives
  Test --> Revision: rival survives
  Revision --> Claim: narrow or rebuild
  Decision --> [*]

The rebound test

If interruption and attenuation are different mechanisms, they should make different predictions. The smallest useful test is to match apparent calm, introduce a neutral perturbation, and measure what returns.

The proposed instrument is the capture rebound test. It compares two conditions with the same practice duration. The concentration-only condition trains target placement and return. The staged analogue also changes cue exposure, behavioural commitment, bodily settling and the relation to the arising pull. It is called an analogue because it does not instantiate or test the full eight-limbed path.

Four observations matter. The first is capture frequency late in the protected session. The second is capture frequency immediately after a standard perturbation. The third is recovery latency, measured from capture to stable return. The fourth is transfer during an ordinary task with a related but not identical cue. A training effect confined to the protected session may be useful, but it is weak evidence of attenuation.

Each observation needs an event rule. A capture might be a key press made when the participant first notices departure, an eye movement towards a prohibited cue, a browser switch, or a pre-defined interruption in task performance. Recovery should require more than touching the target once. For example, define stable return as twenty uninterrupted seconds of task engagement. Decide the rule before seeing the condition labels. Otherwise, an analyst can unconsciously make recovery easier to find in the preferred condition.

The hierarchy of evidence should remain visible. Self-report can locate first-person pull and effort. Behaviour can show enacted capture. Latency can reveal cost. Transfer can show context dependence. None is a privileged truth channel. Agreement strengthens the case; disagreement is information. A participant who reports calm while repeatedly switching tasks may have poor introspective access, a different interpretation of “capture”, or a demand to appear successful.

A minimal model

Let the probability of capture be determined by five illustrative quantities. s is immediate sensory salience. h is learned selection history. i is identity relevance. t is threat or avoidance load. c is currently available stabilising control. The logistic function converts their difference into a probability between zero and one:

The equation is not a psychological law. It compresses the causal proposal. Concentration-only training mostly raises c during the exercise. A staged intervention may raise c less dramatically while also lowering one or more pull terms. A perturbation temporarily reduces available control. The conditions then diverge.

Define the simplest index as:

A high positive CRI means calm was fragile under the chosen perturbation. A low CRI means it was more stable. Neither result establishes a classical kleśa state. A low CRI could arise from habituation to the tone, fatigue, strategic expectation or dishonest reporting. That is why transfer and rival controls remain necessary.

The simulator encodes a testable prediction, not a result. Increasing concentration can produce excellent late-practice performance while leaving a large rebound. Reducing learned and interpretive pull can produce a smaller difference between practice and perturbation. A finding in real participants would be informative only if the perturbation, reporting rule and exposure were matched.

from math import exp


def sigmoid(x: float) -> float:
    return 1.0 / (1.0 + exp(-x))


def capture_probability(s: float, h: float, i: float,
                        t: float, c: float) -> float:
    """Synthetic probability, not a validated psychological model."""
    return sigmoid(s + h + i + t - c - 1.5)


def compare(s=1.1, h=1.2, i=0.8, t=0.7, c=1.2, depth=1.0):
    # Concentration raises control strongly, but perturbation removes much of it.
    co_late = capture_probability(s, h, i, t, c + 1.4)
    co_post = capture_probability(s, h, i, t, c + 0.3)

    # Staging reduces pull terms and makes control less perturbation-sensitive.
    s2 = s * (1 - 0.15 * depth)
    h2 = h * (1 - 0.35 * depth)
    i2 = i * (1 - 0.25 * depth)
    t2 = t * (1 - 0.25 * depth)
    st_late = capture_probability(s2, h2, i2, t2, c + 0.8)
    st_post = capture_probability(s2, h2, i2, t2, c + 0.35)

    return {
        "concentration_only": (co_late, co_post, co_post - co_late),
        "staged_analogue": (st_late, st_post, st_post - st_late),
    }


if __name__ == "__main__":
    for condition, values in compare().items():
        late, post, rebound = values
        print(condition, {
            "late_capture": round(late, 3),
            "post_capture": round(post, 3),
            "capture_rebound_index": round(rebound, 3),
        })

A safe research-observation protocol

The smallest implementation is a low-intensity, counterbalanced observation study. It is not treatment and it is not a substitute for instruction in a living Yoga tradition. Use ordinary breathing only. Do not add breath retention, sensory deprivation, sleep restriction, fasting, trauma exposure or prolonged sitting.

Run twelve brief sessions across three weeks. Randomise the order of six target-return sessions and six staged-analogue sessions so that simple improvement over time does not favour one condition. Each session lasts eight minutes. At minute six, play the same soft tone. The participant records whether attention was captured, how long elaboration continued, and when stable return occurred. Later that day, a preselected ordinary cue provides the transfer probe.

The target-return condition gives one instruction: notice departure and return to a neutral target. The staged analogue adds four observable preparations: remove non-essential cues, make one narrow behavioural commitment not to enact the target habit during the block, settle posture and natural breathing for one minute, and label the cue’s pull without treating it as an instruction. These elements are inspired by intervention depth, not claimed as replacements for yama, niyama, āsana, prāṇāyāma or pratyāhāra.

Rivals, failures and safety

At least six rival explanations can imitate attenuation: participants may expect the staged condition to work; its minute of settling may simply relax them; fewer phone interactions may reduce exposure; repeated tones may habituate; participants may learn the reporting demand; or better sleep may improve control across both conditions. Measure these where possible. Use an active comparison with equal instructor contact and time. Change the transfer cue. Pre-register which outcome has priority.

A fair null result is a successful experiment. If both conditions show the same rebound after expectancy, order and exposure are controlled, the concentration-only explanation survives. If both improve only for the trained tone, the result points to habituation or cue-specific learning. If the staged condition improves transfer but not immediate rebound, it may be changing action policy rather than orienting. The design earns value by separating these outcomes instead of forcing them into a victory narrative.

There is also a dangerous false positive. A person can report fewer thoughts because attention has narrowed into fear, freeze or dissociative detachment. Another can achieve lower capture while sleep deteriorates and ordinary functioning becomes brittle. Meditation-related adverse effects are reported across experimental, observational and clinical literatures, although prevalence estimates vary sharply with definitions, study design and sampling.¹³¹⁴ That variation is a reason for better monitoring, not a single universal risk number.

A positive result in the rebound test permits a modest conclusion: the staged condition changed measured stability under a specified perturbation and context. A null result could mean the causal thesis is wrong, the analogue was too weak, the cue was badly chosen, or the measure was insensitive. An adverse result overrides both. The instrument cannot establish samādhi, metaphysical insight, consciousness, moral transformation or the extinction of a kleśa.

sequenceDiagram
  participant Q as Question
  participant M as Mechanism
  participant E as Evidence
  participant D as Decision
  Q->>M: The calm that broke at 6,40
  M->>E: Figure 1. A training path through recurrent vortices
  E-->>M: compare rival explanations
  E->>D: Test recurrence before calling a practice deeper
  D-->>Q: revise scope or act

Test recurrence before calling a practice deeper

Do not ask only whether attention stayed still. Ask what had to be protected, suppressed or removed for stillness to appear, and what happens when ordinary life returns.

For practitioners, the decision is whether to respond to instability with more force or with a wider intervention. If concentration improves only inside a silent, tightly controlled session, increasing duration may deepen a context-bound skill. The next useful step may instead be to change cue exposure, repeated conduct, embodied preparation, sensory pursuit or the interpretation that turns a signal into an order.

The practical diagnostic is the shape of failure. Immediate drift with little distress may call for clearer placement. Strong bodily agitation may make posture and gentle regulation relevant. Stable practice followed by compulsive re-entry points towards context and habit. Calm accompanied by moral or relational deterioration contradicts the larger path even if attention scores rise. The deepest dosing error is to add intensity when the intervention is aimed at the wrong depth.

For researchers, the decision is whether perturbation and transfer belong in the primary outcome. A study that reports only mean mind-wandering during practice cannot distinguish high control from reduced pull. A matched interruption, recovery measure and ordinary-context probe add that discrimination at low cost.

For teachers and tradition scholars, the decision is whether to describe the limbs as a concentration curriculum or as a coordinated transformation of life and cognition. The latter reading better preserves the text’s link between practice, kleśa attenuation and discernment. It also makes room for an honest research boundary. Reaction-time measures can examine one consequence of training without pretending to adjudicate puruṣa, kaivalya or the status of awareness.

For system designers, the same logic offers a general lesson without turning Yoga into engineering. A system can remain stable because an operator holds it inside a narrow envelope, because disturbances are filtered, or because the system’s response dynamics changed. Stress testing and transfer distinguish those explanations. The comparison preserves a method of causal inquiry, not an ontology of mind.

Patañjali’s distinction between vicchinna and tanu makes a demanding contribution to attention research. It warns that a tendency can vanish from view without losing its power to return. The eight limbs then suggest why change may need to reach conduct, body, breath, sense relation, placement, continuity and discernment together. The practical consequence is simple: stop treating uninterrupted minutes as the whole achievement. Design practice and evidence around recurrence.