The gate that more signs cannot cross

Most teaching follows a reliable rule: when a learner fails, add information. Explain the term, show a diagram, supply an example, correct the misconception and test again. The rule works because the target is usually an object or skill that can be represented more adequately. A student who cannot solve a quadratic may need a missing transformation. A technician who misconfigures a network may need the correct dependency. Better representations improve control.

Now change one variable. Suppose the target concerns the learner’s present relation to representations themselves. The learner is more than trying to know which thought is correct. She is trying to notice how a thought becomes the unquestioned frame through which everything else is taken. A further explanation may still help, but it can also become one more item inside the same frame.

The hidden assumption in ordinary instruction is that the target can be made present as an object for a subject. Call this the objectifying assumption. It is not false in general. It is the basis of science, engineering and much contemplative scholarship. Trouble begins only when the exercise asks how subject, object and representation are being constituted together.

Zen and Mahāmudrā do not solve this problem by abolishing concepts. Both traditions have vocabularies, texts, commentaries, arguments, institutions and demanding forms of training. The issue is narrower. At certain points, a concept must stop pretending to be the reality it organises. Contemporary meditation research calls one aspect of this change dereification: a thought is experienced as a mental event rather than simply as what it represents (Dunne, Thompson and Schooler, 2019).

This gives us a first distinction. A first-order method changes the represented content: a better belief, map or procedure. A second-order method changes the conditions under which representations are selected, trusted and released. The same sentence can do either job. “Notice the thought as a thought” can become a proposition to remember, or it can operate as a cue that alters the next moment of cognition.

Three layers are easy to confuse. The content layer contains what the representation says. The control layer determines when that representation guides attention and action. The evidence layer records what changed when the representation was introduced, removed or contradicted. A learner can possess accurate content while retaining a rigid control policy. She may know that thoughts are transient, repeat the doctrine on demand and still obey every evaluative thought as fact. Conversely, a temporary reduction in verbal thought may occur without any durable improvement in control.

A familiar engineering case makes the distinction concrete. A debugger can print every variable in a running program and still fail to find a race condition because the act of tracing changes the timing. More instrumentation produces better descriptions of the wrong execution regime. The remedy is not anti-measurement. It is a different intervention: deterministic replay, schedule perturbation or a causal trace that preserves the relevant relation. When observation participates in the failure, the method must be redesigned around that participation. Reflexive cognition creates an analogous, though not identical, problem. The representation used to inspect the process can become part of the process being inspected.

flowchart LR
  A["Two students give the same answer"] --> B["The gate that more signs<br/>cannot cross"]
  B --> C["Two disciplines of indirectness"]
  C --> D["A mechanism without a mystical<br/>shortcut"]
  D -. "re-test" .-> A
  classDef first fill:#0a1424,stroke:#007D68,stroke-width:2px,color:#e6f3ff
  classDef step fill:#0a1424,stroke:#3f7da0,stroke-width:1px,color:#e6f3ff
  class A first
  class B,C,D step

Two disciplines of indirectness

The easiest way to compare Zen and Mahāmudrā is also the least useful: say that both point beyond thought. The sentence compresses too much. It hides the different histories, doctrines, training environments and standards of verification that give the practices their meaning. A disciplined comparison must first reconstruct each source structure on its own terms.

Kōan work is not an irrationality machine

Popular retellings often describe a kōan as a paradox designed to exhaust reason until a non-rational breakthrough occurs. That picture catches one possible experience of impasse, but historians have shown why it is inadequate. Kōans belong to textual, ritual, pedagogical and institutional practices. They are quoted, commented upon, performed, contested and tested within curricula. Zen’s rhetoric of transmission outside words coexists with a remarkably large written canon (Hori, 2000) (Foulk, 2000) (Heine and Wright, 2004).

This matters causally. If the kōan were only meaningless noise, any confusion should work. If it were only a riddle, a stored answer should work. Its actual use can combine inherited language, embodied presentation, interview, repeated return, ethical formation and public testing. The point of impasse is not to destroy reason, but to make a particular strategy unable to hide behind its own fluency.

Language in this setting can be descriptive, performative and diagnostic at once. A case may preserve a remembered encounter, stage a relation for the present student and supply material through which a teacher probes fixation. The response is therefore not scored like a multiple-choice item. Timing, adaptation and conduct can matter because the exercise asks whether the response remains alive under changed conditions. This also creates obvious risks. Institutional approval can reward conformity, charisma or theatre. Historical density does not guarantee epistemic reliability. It simply prevents us from reducing the practice to a generic paradox generator.

A scientifically useful reconstruction keeps both facts in view. The curriculum may deliberately constrain ordinary solution search, and its social setting may shape what counts as success. The first fact motivates an intervention hypothesis. The second demands controls for expectation, authority and learned display. Removing the social setting makes the practice easier to model but may remove part of the mechanism. Treating the social setting as infallible makes the mechanism impossible to test.

The Platform Sutra, itself a historically constructed text rather than a transparent transcript, became central to discourse about sudden awakening and seeing one’s nature (McRae, trans., 2000). Dōgen’s Genjōkōan pushes against a spectator model of practice. Its famous line, “To study oneself is to forget oneself”, turns inquiry away from a self inspecting external dharmas and towards a transformation in the relation among self, practice and world (Dōgen, Sōtō Zen Text Project trans.).

Mahāmudrā is not the absence of method

Mahāmudrā is a multivalent term with Indian and Tibetan histories. In Kagyu presentations it can include gradual preparation, calm abiding, insight, direct introduction and repeated familiarisation. Sudden and gradual language therefore does not map cleanly onto method and no method (Jackson, 2011). Dakpo Tashi Namgyal’s Moonbeams of Mahāmudrā is a large, ordered manual, not a celebration of unstructured spontaneity (Dakpo Tashi Namgyal, 2019).

A direct instruction can be brief because its causal function is indexical. It points the practitioner towards an aspect of present cognition rather than describing a remote object. “Look at the mind” is not useful if treated as a command to find a thing with shape and location. It may become useful when it reveals that the search itself is another appearance. Yet even this account is incomplete. Recognition, in tradition-internal terms, is not secured by one striking state. Stabilisation, conduct, lineage interpretation and repeated testing matter.

Directness describes the intended relation between instruction and recognition, not the absence of preparation, language or correction. A finger can point directly only because the learner can follow pointing, distinguish the intended field and resist staring at the finger. The short instruction rests on a long ecology of capacities.

Three false completions should be excluded. The first is blanking: reducing content until little remains and calling the absence insight. The second is objectified witnessing: imagining a detached observer that stands behind experience as a subtle thing. The third is verbal capture: learning the preferred report and treating fluency as recognition. Tradition-internal accounts distinguish these in ways that cannot be replaced by a single laboratory variable. Research can nevertheless build minimal controls for content loss, observer imagery and report rehearsal.

The role of a teacher also differs from that of a database. A database can preserve instructions. A teacher can vary the instruction in response to how it is being misconstrued. That adaptive correction may be one reason brief language works. It also concentrates authority. Any contemporary system borrowing the form of direct pointing must therefore separate interpretive expertise from permission to intensify practice, and must make stopping easier than continuation.

The two traditions also disagree internally. Zen is not one pedagogy. Mahāmudrā is not one map. Historical scholarship warns against treating reports of immediate, pure experience as culturally unmediated data (Sharf, 1995). That warning does not prove that non-conceptual experience is impossible. It changes the burden of evidence: a report must be studied as experience, language, training history and social act at once.

We can now state the comparison precisely. A kōan can create a controlled failure of the strategy “turn the matter into a proposition I can possess”. A Mahāmudrā instruction can invite recognition of cognition without making awareness another represented object. In both cases, the decisive operation may be second-order. The instruction changes the learner’s policy for using representations, rather than only the representations available to the learner.

A mechanism without a mystical shortcut

A causal account needs more than the word “beyond”. It must say what changes, what remains, what would count against the claim and what rival process could produce the same report. The following mechanism is deliberately modest. It does not explain enlightenment. It explains why a class of tasks may respond non-monotonically to more conceptual instruction.

Start with three operations. Saturation makes the habitual strategy visible by letting it run until its limits appear. Puncture interrupts the automatic identification of a representation with its target. Stabilisation preserves clarity after the interruption, so the result is not mere confusion. Kōan impasse often foregrounds saturation and puncture. Mahāmudrā presentations often foreground direct reorientation and stabilisation. Both can contain all three.

The operations make different predictions. Saturation should increase search effort before a strategy switch. Puncture should reduce the causal influence of a previously dominant label, even when the label remains perceptually available. Stabilisation should preserve discrimination, confidence calibration and responsiveness after that reduction. If only effort rises, we have difficulty without reorganisation. If label influence falls together with accuracy, we have suppression. If a participant learns a new sentence but cannot transfer the effect to unfamiliar stimuli, we have semantic retraining.

This sequence also explains why “sudden” need not mean uncaused. A phase transition can be abrupt at the level of observed behaviour while depending on long preparation. Water does not become mysterious at the boiling point because temperature changed gradually before the visible transition. The analogy has limits: contemplative recognition is not thermodynamics, and no scalar temperature is known. Its value is methodological. We should measure the accumulating conditions as well as the reported discontinuity.

The smallest useful formal model

Let L be instruction load, from zero to one. Let S be the scaffold benefit supplied by concepts. Let G be representational grip, and let C be clarity. For an ordinary analytic task, performance can increase with S. For a representation-limited task, performance also depends on the ability to release G.

The threshold τ is the load at which representation begins to dominate the task. The model assumes that conceptual scaffolding is initially useful, that grip rises after a threshold, and that clarity is independently maintained. Remove the third assumption and the falling curve could mean fatigue or suppression. Remove the first and the model becomes romantic anti-intellectualism. The point of formalism is to expose those assumptions.

What scientific evidence can and cannot add

Several research programmes offer partial handles. Sustained non-propositional meta-awareness has been proposed as awareness of mental contents without needing a further explicit proposition about each content (Dunne, Thompson and Schooler, 2019). Enactive work has modelled non-dual practice through non-adaptivity and non-propositional meta-awareness, while explicitly acknowledging that operationalisation remains incomplete (Meling, 2022). Predictive-processing accounts suggest that some deconstructive practices reduce temporally deep, counterfactual modelling, but these are theoretical interpretations rather than established mechanisms (Laukkonen and Slagter, 2021).

Survey research on so-called minimal phenomenal experiences finds recurring dimensions in reports, but factor structure is not a window onto metaphysical essence (Gamma and Metzinger, 2021). Laboratory insight studies show that sudden solutions can have distinct behavioural and neural correlates (Jung-Beeman et al., 2004). Yet impasse models do not describe every solver, and the feeling of an “aha” can accompany false solutions (Fedor et al., 2015) (Danek and Wiley, 2017).

stateDiagram-v2
  [*] --> Claim
  state "The last signpost" as Claim
  state "Representational grip" as Test
  state "Not every difficulty is<br/>profound resistance" as Revision
  Claim --> Test: expose mechanism
  Test --> Decision: evidence survives
  Test --> Revision: rival survives
  Revision --> Claim: narrow or rebuild
  Decision --> [*]

Test the limit without manufacturing mystery

The thesis becomes useful only when it changes experimental design. The common mistake is to ask whether a participant reports a profound state after an unusual instruction. That design rewards expectation, vocabulary and performance. A better test holds the target narrow, adds matched controls and asks whether the proposed intervention changes a causal signature that a rival process should not produce.

The method-resistance assay

The assay uses two tasks. The analytic control requires explicit representation: participants classify short arguments by rule, and additional explanation should help. The label-capture task requires discrimination while a conflicting word competes with the stimulus. Participants hear a high or low tone, then briefly see the congruent or incongruent word. They classify the tone and rate confidence.

Across counterbalanced blocks, participants receive one of three interventions. The description condition adds increasingly detailed conceptual instruction. The interruption condition introduces a brief cue that prevents treating the flashed word as evidence. The meta-awareness condition trains the participant to notice the word and any inner label as events while maintaining pitch discrimination. The dependent measures are tone accuracy, congruency cost, response time, confidence calibration, self-reported blankness and demand-awareness probes.

A credible first study would preregister the crossover threshold, exclusion rules and primary contrast before collecting data. It would use a within-participant design so each person supplies an analytic control and a label-capture curve. Block order would be counterbalanced. The scripts would avoid Buddhist terminology, and the analyst coding open reports would be blind to condition. A relaxation control would match breathing pace and session duration without the dereification cue. A simple executive-control task would test whether any benefit is merely improved distractor inhibition.

The most informative outcome is an interaction, not a group average. We want to know whether instruction load has a positive slope for the analytic control and a curved or negative slope for label capture after the threshold. We then ask whether the release condition selectively reduces incongruency cost while preserving sensitivity. Trial-level hierarchical models can estimate individual differences without turning one noisy mean into a universal law. The preregistration should count a null result as useful: it would show that the narrow task does not require a special method under the tested conditions.

Reports remain secondary. Participants can describe whether the word felt compelling, distant, absent or confusing, but those categories should be linked to behaviour rather than treated as verdicts. Training exposure, meditation history and familiarity with non-dual vocabulary should be recorded because they can alter both strategy and language. The experiment succeeds when it makes a stronger claim harder to make accidentally.

The predicted signature is not a mystical report. It is a selective crossover. Description should improve the analytic control. In the label-capture task it should help initially, then plateau or increase interference. Interruption plus stable meta-awareness should reduce interference without lowering tone accuracy or confidence calibration. If all conditions improve monotonically, the task is probably ordinary skill learning. If interference falls because accuracy collapses, the intervention produced suppression, not flexible release.

from dataclasses import dataclass
from math import exp

@dataclass(frozen=True)
class Trial:
    task_demand: float
    instruction_load: float
    release_cue: float
    stability: float
    clarity: float

def clamp(x: float) -> float:
    return max(0.0, min(100.0, x))

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

def score(t: Trial) -> dict[str, float]:
    d, load, cue, stable, clear = (
        t.task_demand, t.instruction_load, t.release_cue,
        t.stability, t.clarity
    )
    scaffold = 1.0 - exp(-load / 24.0)
    grip = clamp(0.72 * load + 0.28 * d - 0.52 * cue - 0.38 * stable + 28.0)
    clarity_window = exp(-((clear - 70.0) / 28.0) ** 2)
    release = 100.0 * scaffold * (1.0 - sigmoid((grip - 48.0) / 8.0))
    release *= clarity_window * (0.35 + 0.65 * stable / 100.0)
    analytic = clamp(18.0 + 0.52 * load + 0.20 * d + 0.12 * stable - 0.05 * cue)
    blankness = clamp(1.15 * max(0.0, 58.0 - clear) + 0.25 * cue - 0.30 * stable)
    return {"analytic": analytic, "release": clamp(release),
            "grip": grip, "blankness": blankness}

if __name__ == "__main__":
    trial = Trial(55, 45, 45, 65, 70)
    print(score(trial))

The reference implementation makes the synthetic assumptions inspectable. Change one coefficient at a time, plot the curves and test whether the claimed crossover survives.

A configured system must preserve authority and safety