Hidden in the Criteria: Scene Construction and the Conditions of Retrieval

For fuck’s sake.

There it is again.

Good researchers.
Good data.
A difference that refuses to behave like a deficit is supposed to behave.

Give the autistic person words
and nothing moves.

Give her a picture
and the gap disappears.

Not narrows.

Disappears.

And everyone writes down autism.

That is the wound I keep finding.

Not bad science, exactly.
Something harder to remove.

A category large enough to swallow the architecture that produced the result.

Because somewhere inside those samples—almost certainly, invisibly, uncounted—are gestalt processors.

People for whom the picture is not merely more support.

People for whom a scene does not begin as details waiting to be assembled into a whole.

People for whom the whole may already be there.

People for whom words and pictures and bodies and rooms and relationships and time are not interchangeable routes towards the same stored object, but different entrances into it.

And nobody asked.

Nobody separated autism from processing architecture because the instrument had no variable called processing architecture.

So the difference enters the study as autistic.

The failed verbal retrieval is autistic.

The visual completion is autistic.

The delayed arrival is autistic.

The dependence on context is autistic.

The strange relationship between cue and access is autistic.

Everything goes into the same diagnostic mouth.

Then comes back out as deficit.

This is Hidden in the Criteria all over again.

Only now it is hidden in the sample.

A population is assembled because everyone shares a diagnosis, and the diagnosis is treated as though it names the mechanism underneath whatever the experiment subsequently finds.

But autism is not a processing architecture.

It is not a theory of memory.

It does not tell us whether experience was stored sequentially or simultaneously, whether language mediates retrieval or arrives afterwards, whether a visual field supplies context or constitutes the entrance itself, whether relational safety changes what can become available, whether the thing being called construction was ever constructed at all.

Those questions disappear when autistic is allowed to answer them in advance.

And this is the part that hurts.

Because the data keep trying to tell us.

Again and again, across literatures that do not speak to one another, the expected deficit changes shape when the conditions change. A capacity disappears here and returns there. A person who cannot produce the thing on command produces it through another axis. The instrument records inconsistency.

I see architecture.

I see people like me buried inside the mean.

I see the possibility that some portion of what autism research has spent decades describing as impaired memory, impaired narrative, impaired imagination, impaired self-projection, impaired whatever-comes-next may contain something much more specific:

a whole that cannot enter through the door the experiment keeps opening.

And because nobody thought to ask whether there might be different kinds of minds inside the autistic group, the door becomes invisible.

Only the failure remains.

That is the wound.

Not that the researchers failed to find us.

That we may have been there the entire time—
producing the anomaly,
bending the distribution,
responding differently to the picture,
not responding to the words,
leaving our fingerprints all over the data—
and the only name the research had available for what we were doing

was autism.

For fuck’s sake.

The evidence keeps arriving before the reader does.

The frustration above has a specific empirical source.

There is a moment in the literature on autism and memory that is easy to pass over because, within the existing framework, it looks like an ordinary methodological discrepancy. Read differently, it is considerably harder to dismiss.

Anger et al. (2019) showed autistic younglings a visual cue—a simple cartoon image—before asking them to recall a specific event. Under that condition, the statistical difference in event recall between the autistic group and the comparison group disappeared. Some differences in finer-grained sensory detail remained, but the expected group difference in specific event retrieval did not.

Elsewhere, Ciaramelli et al. (2018) offered what might conventionally be understood as more retrieval support: verbal prompts, structured questions, increasingly specific probes intended to help participants recover episodic detail. The comparison group improved substantially. The autistic group did not.

Two studies. One nominal population. One broadly related capacity. Opposite responses to support.

That distinction matters more than it initially appears to.

The literature has not ignored the discrepancy. It has done something more ordinary, and perhaps more consequential: it has contained it. Differences in method, cueing, task design, and retrieval support provide sufficient explanation for the findings to remain beside one another without requiring the larger construct to change. Each result can therefore be locally correct without the field having to ask what becomes visible when they are read together.

But the poem above begins from precisely that refusal to keep them separate.

If autism is the only meaningful architecture available for describing the participants, then whatever happens inside the autistic sample becomes evidence about autism. Visual cueing helps autistic participants. Verbal retrieval support does not. Both findings remain attached to the diagnostic category, even though the divergence between them may be telling us about something the category itself does not name.

That possibility is where this piece begins.

Because if a capacity can become statistically indistinguishable from a comparison group when the route of access changes, then the question is no longer simply why autistic autobiographical memory is impaired, or why one intervention works better than another.

The prior question is more fundamental:

What, exactly, was being measured as absent before the right entrance was offered?

Everything that follows is the slower work of taking that question seriously—first in the data, then in the assumptions about capacity, time, relationship, and standardisation that allow the contradiction to appear without requiring the theory around it to move.

The contradiction at the centre of this piece does not depend on a single anomalous result. It appears when three findings, produced for different purposes and usually read within different conceptual frames, are allowed to sit beside one another.

The first is Anger et al. (2019). Autistic younglings were given visual cues—cartoon images—before being asked to retrieve specific autobiographical events. Under that condition, the expected group difference in event recall disappeared. Some differences in finer-grained sensory detail remained, but visual cueing changed the basic retrieval outcome enough that autistic participants became statistically indistinguishable from the comparison group on event recall.

The second is Ciaramelli et al. (2018). Here, retrieval was supported differently. Rather than supplying a visual entrance into the event, researchers used structured verbal probes intended to elicit additional episodic detail. This should matter if support is principally additive—if more scaffolding simply makes a difficult retrieval task easier. Yet the groups did not respond symmetrically. The comparison group benefited substantially from increased verbal support. The autistic group largely did not.

Already, something more interesting than autistic memory impairment is happening.

The capacity being measured is not responding simply to the presence or absence of assistance. It is responding differently to the form through which assistance arrives. A visual cue and a verbal probe may both occupy the methodological category of retrieval support, but that category tells us almost nothing about what either intervention actually does to the organisation of access.

Rubin et al. (2019) supplies the third piece from outside the autistic sample altogether. Working with typically developing adults, they asked what makes an autobiographical memory feel vivid, relived, and believable as something that actually happened. The strongest predictor was not the sheer quantity of detail retrieved, nor subjective emotional intensity. It was the clarity of spatial layout—the extent to which the remembered event cohered as a scene.

That finding changes how the first two can be read.

If scene coherence matters to the phenomenology of autobiographical remembering, and if autistic retrieval changes dramatically when a visual scene is supplied but not when additional verbal probing is supplied, then amount of support may be the wrong dimension along which to organise the discrepancy. The important difference may be architectural. One intervention supplies more language. The other supplies relations—objects already held in position, spatial organisation already available, a field within which the remembered event can begin to cohere.

Read separately, these remain three respectable findings about three somewhat different questions: visual cueing, verbal retrieval support, and the phenomenological organisation of autobiographical memory.

Read together, they begin to describe a pattern.

Not all entrances are equivalent.

That matters especially once we stop assuming that an autistic sample constitutes a single processing architecture. None of these studies was designed to identify gestalt processors within its autistic participants. None could therefore ask whether some participants approached autobiographical memory through wholes, relational configurations, or scene-level organisation rather than through the sequential accumulation of retrievable details. Whatever variation of that kind existed inside the samples could only appear in the data under the category the studies already possessed: autism.

This does not establish that gestalt processing explains the Anger–Ciaramelli discrepancy. The existing studies cannot answer that question because they did not measure it.

It establishes something prior, and methodologically more important.

The data do not behave as though support were a single quantity that can simply be increased until an impaired capacity performs better. They behave as though route matters—as though what becomes available depends partly upon the architecture through which access is attempted.

The literature already contains that possibility.

It has simply been distributed across separate studies, separate methods, and separate explanatory vocabularies, where each finding can remain locally intelligible without anyone having to ask what happens when the pieces are permitted to become a whole.

If route matters, then something more fundamental has to move with it.

The contradiction survives because psychological measurement rests on an assumption so ordinary that it rarely has to announce itself: capacity is what becomes available on demand.

A capacity is treated as a standing property of the person, available to report for duty when the examiner asks. The task is presented. The cue is given. The clock starts. Something becomes available, or it does not. What appears inside that window is scored, compared, and eventually returned as a statement about what the person can do.

The movement from available here to capacity possessed is so deeply built into the structure of assessment that it can disappear as an inference altogether.

But the cueing literature makes that inference difficult to sustain.

Anger et al. (2019) did not add autobiographical memory capacity to participants who previously lacked it. A cartoon image did not teach scene construction in the moments before retrieval. Something about the conditions of access changed, and with that change came a different performance from the same population on the same broad capacity.

Ciaramelli et al. (2018) makes the distinction sharper rather than softer. More retrieval support did not produce the same movement. Verbal probing benefited the comparison group substantially whilst leaving the autistic group largely where it had been. Whatever separates these outcomes, it cannot be described adequately as more help produces more capacity. The route through which retrieval is invited is doing work of its own.

That leaves us with two variables where the measurement architecture has often presumed one.

There is capacity—whatever organisation of memory, scene, relation, and experience exists within the person.

And there is availability—whether that organisation can become traversable here, now, through this cue, in this room, under these conditions.

Those are not the same claim.

For a gestalt processor, the distinction may be especially consequential. If experience is held relationally, as configurations whose elements acquire meaning through the whole rather than as independently retrievable pieces waiting to be assembled, then a cue does more than provide assistance. It determines where, and perhaps whether, an entrance into that configuration can be found. A verbal probe asking for another detail may therefore be no closer to a visual cue supplying scene organisation than repeatedly knocking harder is to discovering that one has been standing at the wrong door.

This is precisely what disappears when processing architecture is absorbed into the larger category of autism. The instrument records whether the answer arrived. The analysis records a group difference. Neither necessarily has a variable capable of describing how the person had to reach what she knew.

Once that distinction becomes visible, failure under a particular condition can no longer bear quite the theoretical weight placed upon it.

Non-arrival is evidence of non-arrival.

It may reflect diminished capacity. It may reflect difficulty accessing an intact capacity. It may reflect an interaction between capacity, cue, processing architecture, time, and conditions that these studies were never designed to separate. The existing evidence cannot decide among those possibilities because the instruments collapse them at the point of measurement.

That is the narrower claim, but it produces the larger problem.

If availability varies with the route through which a capacity is approached, then a single testing window cannot tell us, by itself, whether we have measured the limits of the capacity or merely the limits of the entrance we happened to provide.

The entrance may have failed.

The whole may still be there.

Once capacity and availability are separated, another assumption becomes visible.

The clock is doing theoretical work.

A testing window appears to be nothing more than administration—a practical boundary around the thing being measured. The participant begins here. The task lasts this long. The response is recorded before the window closes. None of that is inherently problematic. Research requires boundaries, and chronos—clock time, sequential time, the time in which one event follows another and duration can be measured—is extraordinarily good at providing them.

The problem begins when the boundary around the measurement becomes a boundary around the capacity.

A testing window does not merely ask whether autobiographical memory exists. It asks whether some part of that memory can become available now: inside this bounded interval, following this particular cue, on this particular pass, without needing to return later, approach from another direction, or wait for something not yet present in the conditions of retrieval.

Chronos can tell us exactly how long the participant was given.

He cannot, by himself, tell us what that duration means.

That distinction matters because there is another temporal organisation involved here—kairos, not simply as “more time,” and certainly not as an alternative quantity of minutes, but as the time of conditions becoming sufficient for something to happen. I have elsewhere called this meaning-time: the temporality in which an experience can already have occurred, already have been registered, already have altered the person carrying it, whilst its meaning, language, or traversability has not yet become available (Hoerricks, 2026a).

Kairos does not run slower than Chronos.

She answers a different question.

Chronos asks: how much time has passed?

Kairos asks: has the moment in which this can become available arrived?

For many forms of experience, those questions coincide closely enough that the distinction disappears. I am asked what happened yesterday; I remember; I answer. Event, access, language, and report arrive near enough together that chronological sequence appears to describe the whole process.

But they do not always coincide.

Sometimes an experience becomes traversable only on return. Sometimes a later image supplies relations that a verbal question could not. Sometimes language arrives after the body has already known. Sometimes the meaning of an event becomes available months or years after the event itself—not because the memory spent those years slowly constructing itself, but because the conditions from which it could be entered had not yet converged.

For a gestalt processor, that distinction may be especially important. If what is held is relational and whole, rather than stored as a sequence of independently retrievable elements, then waiting is not necessarily waiting for additional pieces to accumulate. It may be waiting for an axis from which the whole becomes traversable.

This is where the testing window ceases to be temporally neutral.

The protocol operates in Chronos because it must. The category error comes afterwards, when performance inside Chronos is interpreted as though it exhausts what can exist in Kairos—when not available during the testing interval quietly becomes not possessed by the person.

The laboratory version is easy to recognise once stated. A cue is given. Thirty seconds pass. A minute. Another prompt. The response does not arrive. The window closes, and the absence inside that window becomes data.

Outside the laboratory, autistic people encounter the same conversion constantly.

Someone asks what happened.

What did you feel?

Why did that upset you?

What did you mean?

The question arrives in Chronos and expects its answer there too. When meaning has not yet become traversable, the delay is easily read as evasiveness, lack of insight, poor autobiographical memory, weak emotional awareness, or simply not knowing. Later—sometimes minutes later, sometimes after the conversation has ended, sometimes after sleep, movement, writing, another context, another person, another year—the whole becomes reachable.

Nothing necessarily came into existence at that later moment.

Something became available.

That is what a single testing window cannot distinguish.

The problem, then, is not that psychology uses clocks. It could hardly do otherwise.

The problem is allowing the clock that measures the observation to become, without argument, a theory of the phenomenon being observed.

Chronos can tell us when the answer arrived.

He cannot tell us that whatever did not arrive was never there.

If the clock is not merely the container around the measurement, neither is the room.

Follow the same logic one dimension further and another background condition begins to move into the foreground. Across this literature, researchers have varied the cue: visual versus verbal, cued versus uncued, lower versus higher retrieval support. They have asked, quite reasonably, what happens when the route into memory changes.

What they have not varied is the relationship within which retrieval is being asked to occur.

Is the examiner familiar?

Has the participant met her before?

Has enough time passed for the interaction itself to become predictable?

Does the participant know what happens when she cannot answer immediately, asks for clarification, moves away from the expected form of response, or needs to return to something later?

Has the nervous system had any reason yet to learn that this particular person, in this particular room, is safe enough to stop monitoring the encounter and allow attention to go elsewhere?

These are not incidental questions if retrieval depends upon more than cue modality. Yet they largely disappear under standardisation—not because researchers have demonstrated that they do not matter, but because the protocol needs them to remain sufficiently constant that they can be treated as background.

That is methodologically understandable.

It is not theoretically neutral.

In the five-dimensional account I have developed elsewhere, I use relational field to name something more specific than rapport and more consequential than comfort: the organisation of presence between a person and what surrounds her, including whether another person can be held as stable enough that meaning does not have to keep reorganising around uncertainty, vigilance, or threat (Hoerricks, 2026a). The field is not another cue arriving alongside the visual or verbal one. It is the condition within which either cue has to become usable.

That distinction matters here.

If relational field contributes to whether autobiographical material can become traversable, then examiner familiarity cannot simply be classified as noise around the measurement. Safety cannot be reduced to a pleasant addition that might improve performance slightly. Trust cannot be treated as an unspecified quantity of “support” interchangeable with another prompt.

They belong to a different order of the problem.

A cue offers an entrance.

The field may determine whether an entrance can be used.

And we do not presently know how much work that distinction is doing in this literature, because the studies were not designed to find out.

This is where the wound named at the beginning of this piece opens again. If there are gestalt processors inside these autistic samples—and there almost certainly are—then there may also be participants for whom relational organisation is not contextual decoration around cognition but part of how cognition becomes available at all. Yet processing architecture is not identified, relational field is not manipulated, and the resulting performance is aggregated under autism. What may be an interaction between processing architecture, relational stability, cue, and access becomes another characteristic of the diagnostic group.

The field disappears.

The deficit remains.

This also complicates what standardised administration is doing. A fixed script, an unfamiliar or minimally familiar examiner, a bounded encounter, and a single administration window are not necessarily hostile conditions. Nor should we simply declare them universally low-field for every autistic person. But they are conditions deliberately designed to minimise the influence of relationship upon measurement.

That design makes sense only if relationship is extraneous to the capacity being measured.

If it is not—if relational stability participates in making some forms of retrieval available—then standardisation has done something methodologically peculiar. It has controlled a potentially constitutive variable before establishing whether that variable belongs inside or outside the construct.

And because relational field is held outside the model, the resulting studies cannot tell us whether a failure of retrieval reflects diminished scene-construction capacity, an inaccessible route, insufficient field stability, an interaction among them, or something else altogether.

The literature has not demonstrated that relational field is the missing variable.

It has done something more interesting.

It has left us no experimental basis for ruling it out.

The room, like the clock, was never empty.

It was part of the instrument all along.

This is where the methodological problem becomes more serious.

Standardisation exists for good reason. Hold relevant conditions constant, administer the same task in the same way, reduce unwanted variation, and the resulting measurements become comparable. If the procedure produces similar results across administrations, examiners, and samples, we gain confidence that we are measuring something stable rather than chasing noise.

That is reliability.

And reliability matters.

But one can be reliably wrong.

A scale that adds five kilograms every time is reliable. A clock that loses exactly seven minutes each day is reliable. Repetition tells us that the error is stable; it does not make the measurement accurate.

Psychological measurement has the same problem, except that the error is harder to see because the thing being measured cannot be placed independently on another scale. We infer the construct through the conditions we create for it to appear. If those conditions systematically alter access to the construct, then repeated measurement can reproduce the same distortion with remarkable consistency.

The result becomes reliable precisely because the conditions producing it are reliable.

That distinction matters enormously here.

If autobiographical retrieval varies not only with underlying memory capacity but with cue architecture, processing architecture, time of availability, and relational field, then a standardised protocol is not simply measuring the person whilst holding the room still. It is measuring the person-in-these-conditions and then deciding, analytically, which portion of the resulting performance belongs to the person.

The instrument may be highly reliable in doing so.

The question is whether that attribution is accurate.

And beneath accuracy sits the more technical problem of construct validity: whether the measurement actually warrants the claim subsequently made from it. If a participant cannot retrieve a specific autobiographical event during a bounded verbal task with an unfamiliar examiner, the observation itself may be perfectly accurate. She did not retrieve the event under those conditions. The error enters when that observation is allowed to travel—to become evidence of diminished autobiographical memory capacity, impaired scene construction, or some stable property of autistic cognition without first separating capacity from the conditions governing its availability.

The measurement has moved from:

it did not appear here

to:

she does not possess it there.

That movement is where accuracy can be lost even whilst reliability remains intact.

And standardisation can make the problem harder to detect.

If relational field is constitutive of access for some participants, then holding relational conditions constant does not remove their influence. It holds their influence constant too. If a particular cue architecture systematically fails to provide an entrance for some processing architectures, administering that cue consistently does not neutralise the mismatch. It reproduces the mismatch. If a bounded testing window repeatedly closes before meaning becomes traversable, using the same window across participants does not make time irrelevant.

It makes the temporal demand reliable.

This is particularly consequential when everything is subsequently aggregated under autism.

Imagine that a sample contains several processing architectures, including gestalt processors for whom visual, relational, spatial, or whole-scene access differs substantially from sequential verbal retrieval. The protocol does not identify that architecture. It administers the same entrance to everyone, records who can enter through it, aggregates the resulting differences, and reports an autistic group effect.

Run the study again.

Use the same protocol.

Recruit another autistic sample containing an unknown mixture of processing architectures.

Find the same group difference.

Replicate it ten times.

The finding becomes extraordinarily reliable.

And it may still be wrong about why the difference exists.

This is the distinction I want to protect, because none of it requires denying the observed difficulty. Autistic people can struggle profoundly with retrieving autobiographical detail on demand, particularly under ordinary, unaccommodated conditions. That difficulty matters—in classrooms, therapy rooms, clinical assessments, police interviews, courtrooms, relationships, and every other place where credibility can depend upon producing a coherent account on someone else’s schedule.

The observation is real.

What remains unsettled is its attribution.

A person may genuinely fail to retrieve something under standardised conditions. A study may accurately record that failure. A research programme may reliably reproduce it across decades. And the explanatory framework may still be wrong about what the failure belongs to.

Capacity.
Access.
Cue.
Clock.
Field.
Processing architecture.

Some interaction among all five.

If the instrument cannot distinguish among them, then consistency cannot perform that distinction afterwards.

Reliability tells us that the pattern repeats.

Accuracy asks whether we have located the pattern correctly.

Construct validity asks whether the name we have given it is one the evidence can actually bear.

Those are different achievements.

And a field can become extraordinarily good at the first whilst quietly failing the other two.

This is where the argument becomes difficult for me to keep at the distance of methodology.

The frustration at the beginning of this piece—the for fuck’s sake of finding another literature full of careful observations that seem to stop one conceptual move before the thing I recognise—is not abstract frustration. I recognise this pattern because I have lived inside it.

Intensely.

I know what it is to have an experience remain whole without being available from the direction another person is asking me to enter it. I know what it is to be asked for a detail and find nothing there, then encounter some apparently peripheral thing—a place, a bodily relation, a phrase, a movement, the presence of another person—and have not merely the requested detail but the whole field become traversable at once.

From outside, those moments can look like retrieval failure followed by successful retrieval.

From inside, that is not what they feel like.

Nothing was assembled.

Something opened.

That distinction is difficult to demonstrate experimentally because the conditions through which I have come to recognise it are almost exactly the conditions a standardised protocol removes. No fixed cue. No requirement that retrieval complete itself in one sitting. No assumption that the first available axis is the correct one. No prohibition against returning months later when something else has become reachable.

I have, however, documented one such return without initially intending it as evidence about memory at all.

Two pieces, published two months apart, return to the same evening. The same alcove. The same encounter (Hoerricks, 2026b; Hoerricks, 2026c).

The first approaches the evening through field. What was present between us. The room as participant rather than backdrop. Witnessing. The way desire had already reorganised the encounter before either of us possessed language capable of saying what was happening.

That piece was complete.

I did not publish it believing that I had produced an incomplete account waiting for missing details to be recovered later. Nothing in me experienced the scene as unfinished.

Then, two months later, I returned.

Not because someone asked me to remember more.

Because another axis had become available.

This time the entrance was bodily and mechanical: weight, contact, gravity, pressure, height, stance, the specific negotiation required when two bodies meet and their dimensions materially alter what touch can do. Suddenly I could move through the same remembered event from somewhere else.

And this is the part that matters.

The second account did not correct the first.

It did not replace it.

It did not fill holes left by inadequate retrieval.

The earlier scene remained exactly as whole as it had been. What changed was that another set of relations within that whole had become traversable. Later language could now move through something that had already been present, simultaneously, at the time of the event.

I described it then as later language brought to sit beside an earlier simultaneity.

That was phenomenological description before I realised it was also a problem for the literature I am reading here.

Because what would a standard scene-construction task have made of me?

Ask from the first axis and I might produce the room, the relational field, the witnessing, the organisation of desire.

Ask for weight and mechanics at that same moment and perhaps I cannot give them to you.

Score the omissions.

Prompt me verbally.

Ask again.

Increase retrieval support.

Watch nothing happen.

Then conclude something about the specificity, completeness, or construction of my autobiographical memory.

Two months later, I write the other piece.

What changed?

That question is the reason I keep arriving at this literature with something considerably stronger than academic curiosity.

If the later account contains material unavailable during the earlier retrieval, the conventional model has an easy explanation: additional retrieval occurred. More of the scene was reconstructed. Previously inaccessible details became available and were incorporated into a richer representation.

Perhaps.

But that description does not capture the phenomenology I am trying to name.

I did not experience the second return as finding additional pieces and adding them to the first account. I experienced a change in where I could stand inside the same whole. The relations available from that position changed. The scene did not become larger. It became differently traversable.

This is why the word construction catches against my experience.

Construction implies an assembly problem. Pieces exist, pieces are retrieved, pieces are related, and sufficient successful assembly produces a coherent scene.

But what if coherence can precede traversability?

What if the whole can already exist before all of its internal relations are independently available to language?

What if a failed request for detail tells us not that the scene has not been constructed, but that the axis from which that detail becomes reachable is not presently available?

Then the apparent sequence—

failure → cue → retrieval → completed scene

—may sometimes be describing something quite different:

whole → inaccessible axis → entrance → traversability.

That possibility is what I keep seeing beneath the data.

And it is why finding studies in which autistic performance changes radically with cue architecture produces that involuntary for fuck’s sake.

Because I know this shape.

I know what it is to have someone standing outside the whole, asking me for the piece that would prove to them that the whole exists.

I know what it is for the piece not to come.

And I know what it is for the whole to remain there anyway.

None of this proves that the participants in Anger et al. (2019) or Ciaramelli et al. (2018) experienced memory as I do. The studies cannot tell us that. They did not identify gestalt processors, did not ask about whole-first phenomenology, and were not designed to distinguish construction from traversability.

That limitation is precisely the point.

The literature has interpreted differences in what becomes available as differences in the capacity to construct a scene without first establishing whether construction is what every participant is doing.

My own case cannot settle that empirical question.

It can make the missing question visible.

Perhaps the relevant question for at least some autistic people is not:

Can this person construct a coherent scene?

Perhaps it is:

From what axis does an already-coherent scene become traversable?

Those questions are not refinements of one another.

They describe different architectures of mind.

And if gestalt processors are hidden inside samples organised only by autism, the instrument may repeatedly ask the first question of people whose experience is organised by the second—record the mismatch faithfully, reproduce it reliably, and call what remains a deficit.

For fuck’s sake.

I have been trying to tell you where the scene went.

It did not go anywhere.

I did not arrive at any of this from inside the autobiographical-memory literature.

I came from the other direction.

For years I have been re-reading the instruments autism research uses to diagnose, classify, and describe—ADOS-2, ADI-R, SCQ, Vineland, CBCL—and finding, again and again, behavioural observations coded as evidence of autistic deficit sitting directly on top of something that becomes differently legible once gestalt processing is available as an explanatory architecture.

Scripts.
Delayed recombination.
Context-dependent access.
Whole units of meaning retrieved intact.

Language that appears inflexible when analysed word by word but becomes coherent when the unit of analysis is allowed to become larger.

Behaviour recorded accurately and interpreted through an architecture that had no category for the thing producing it (Hoerricks, 2026d).

That was Hidden in the Criteria.

What unsettles me about the scene-construction literature is that I have now arrived at the same problem from somewhere I was not looking for it.

There is almost no citation traffic between these bodies of work. The researchers studying autobiographical memory and scene construction are not generally asking questions about gestalt language processing. The GLP literature has not, in turn, organised itself around autobiographical memory, allocentric navigation, or scene construction.

And yet, when placed beside one another, the patterns begin to rhyme.

A capacity that changes dramatically depending upon the route through which it is approached.

Wholes whose availability cannot be predicted by the availability of their apparent parts.

Context doing more work than the experimental model knows how to name.

Performance interpreted as capacity.

Difference aggregated under autism before processing architecture has been separated from diagnosis.

Two literatures can therefore produce observations that point towards the same underlying architectural possibility without either literature being capable of naming it. Not because the researchers are careless. Because each field inherits the variables its predecessors taught it to see.

That absence of cross-citation matters to me.

It suggests that I am not simply finding another example inside a theory I already favour. Independent research traditions, built around different questions and different constructs, have produced findings that become unexpectedly coherent when read through the distinction between whole and part, capacity and availability, construction and traversability.

But it also produces another, rather more personal, for fuck’s sake.

Because I keep reading these studies and thinking:

I could have helped you.

Not by arriving after data collection to provide an autistic quotation for the discussion section.

Not by sitting on an advisory panel whose contribution begins after the hypotheses have already hardened into variables.

Not by being asked whether the participant information sheet sounds respectful, whether the language is acceptable, whether autistic people will tolerate the procedure, or whether the conclusions should be softened before publication.

Earlier.

At the table where the study becomes a study.

Because an autistic gestalt processor on the research team might have looked at “visual cue” and “verbal retrieval support” and refused the assumption that these were merely different quantities of the same thing.

She might have asked whether the picture was supplying information at all—or supplying an organisation within which information could become reachable.

She might have asked what happened before the clock started.

Whether the examiner was familiar.

Whether the participant who appeared unable to retrieve the scene could write it later.

Whether returning tomorrow changed anything.

Whether the same person who failed under verbal probing could enter through movement, spatial layout, drawing, objects, relation, or another whole.

She might have looked at variability inside the autistic group and asked whether the variance was noise.

She might have asked whether autism was carrying too much explanatory weight.

She might, most importantly, have heard the phrase scene construction and said:

Are we certain they are constructing it?

That question has to enter before the protocol is built.

Once the study has operationalised scene construction as performance under a particular cue, inside a particular window, with processing architecture unmeasured and relational conditions standardised out of analytic visibility, lived-experience consultation cannot repair the design afterwards. The categories have already been made. The possible answers have already been constrained. The instrument has already decided what kind of mind it expects to meet.

This is why autistic researchers, and gestalt processors specifically when gestalt processing is plausibly implicated, need to be full intellectual participants in research teams rather than evidence of participatory inclusion.

Representation is not the point.

Epistemic access is.

A researcher who lives inside an architecture the dominant model does not contain does not merely bring another perspective on the same object. She may notice that what everyone else has agreed is the object is actually an artefact of where they have been standing.

That does not make her interpretation automatically correct.

Lived experience is not methodological immunity. Autistic researchers can overgeneralise, become attached to theories, mistake personal phenomenology for population-level mechanism, and be wrong like anybody else.

But that is precisely why she belongs inside the research team.

Her propositions can be challenged.
Operationalised.
Compared against alternatives.
Built into hypotheses.
Allowed to fail.
Replicated.

What cannot happen is the testing of a question nobody in the room knew needed asking.

And that is what keeps catching in me as I read this literature. Not you should have listened to autistic people as a ceremonial conclusion.

Something much more practical.

You needed somebody in the room who could see this before you built the room.

Because the data are good. That is almost what makes it hurt.

The observations are there. The discrepancies are there. The strange responsiveness to cue type is there. The instability between what should supposedly be one capacity under different conditions is there.

I could have helped you ask what that meant.

Another gestalt processor might have asked something entirely different and seen something I cannot see.

That is what a research team is for.

Not window dressing.

Not representation after the architecture has already been built.

Different architectures of mind participating in the construction of the question itself.

Otherwise we will keep doing this.

We will recruit autistic people into samples, collapse whatever processing differences exist among them into autism, produce careful measurements of the resulting variation, and then wonder why the data keep refusing, at their edges, to behave like the theory says they should.

For fuck’s sake.

We were already in the data.

We needed to be in the room.

If this argument holds, assessment has to become capable of separating capacity from access. Multiple administration windows. Examiner familiarity treated as a variable rather than background. Different cue architectures offered as genuinely different entrances rather than different quantities of support. Processing architecture identified rather than silently absorbed into autism.

None of this is particularly exotic. It follows directly from taking the literature’s own inconsistencies seriously.

But there is another question I do not want to lose beneath the pleasure of designing a better experiment:

Who asked for another experiment?

Sher Griffin, writing recently about what she calls The Autistic Gap, names the distance between increasingly affirming accounts of autism and systems still organised around demonstrable impairment. People can spend enormous personal resources building the conditions under which they remain functional, only to have those successful accommodations returned as evidence that they require no support. Remove the privately purchased scaffolding and the disability becomes immediately legible again (Griffin, 2026).

That matters here because autistic people do not principally need research to become more sophisticated at describing why we struggle.

We need material help.

Income. Housing. Healthcare. Communication access. Sustainable work. Support that does not disappear in adulthood. Environments that do not require us to become visibly broken before our needs become credible. Sher’s argument is particularly sharp here: accommodations themselves presume enough social and economic standing to request them, whilst some environments are injurious at the level of their basic organisation.

So yes—build the better study.

Separate access from capacity. Test the cue. Test the clock. Test the field. Put autistic and gestalt-processing researchers inside the design team before the questions harden into instruments.

But know what the research is for.

We do not need increasingly precise maps of the places where autistic people drown.

We need someone to build the fucking bridge.

Go back to the beginning.

A researcher shows an autistic youngling a picture, and the expected difference disappears.

Sit with that long enough and it stops looking like a footnote about methodological variation. It becomes a much more difficult question about what we have been calling absence when the thing being measured may instead be access—available along one axis, unavailable along another, and recorded by an instrument with no variable capable of telling the difference.

The capacity may have been there all along.

I should probably acknowledge, before closing, that this piece contains rather more colourful language than my writing here usually does.

I began writing it at about one o’clock in the morning because I could not sleep.

There is no shortage of advice about how to calm an autistic person sufficiently for sleep, although a remarkable amount of it appears to have been written for parents trying to settle children rather than wee middle-aged autistic women lying awake because something in the architecture of an argument will not leave them alone.

This is how I calm myself.

I follow the thing.

I read the paper again. I find the contradiction. I put one study beside another. I swear at them. I write until the pieces stop arriving as pieces and the whole becomes traversable enough that I can finally set it down.

Which is, I suppose, an inconveniently appropriate way for this essay to have been written.

It began as a note about scene construction. It became an argument about something this project has been circling for a long time: psychology repeatedly mistakes punctual availability for underlying capacity—across memory, across language, across every domain in which a clock quietly becomes a theory because nobody thought to ask what the clock was assuming before it started.

Perhaps that is why I needed to write this morning.

The whole was already there.

I needed somewhere for it to arrive.

Now, perhaps, I can sleep.

Agron, A. M., Martin, A., & Gilmore, A. W. (2024). Scene construction and autobiographical memory retrieval in autism spectrum disorder. Autism Research, 17(2), 204–214. https://doi.org/10.1002/aur.3066

Anger, M., Wantzen, P., Le Vaillant, J., Malvy, J., Bon, L., Guénolé, F., ... & Guillery-Girard, B. (2019). Positive effect of visual cuing in episodic memory and episodic future thinking in adolescents with autism spectrum disorder. Frontiers in Psychology, 10, 1513.

Ciaramelli, E., Spoglianti, S., Bertossi, E., Generali, N., Telarucci, F., Tancredi, R., ... & Igliozzi, R. (2018). Construction of past and future events in children and adolescents with ASD: Role of self-relatedness and relevance to decision-making. Journal of Autism and Developmental Disorders, 48(9), 2995-3009.

Griffin, S. (2026). The Autism Gap. The Cognitive Ecologist . https://thecognitiveecologist.substack.com

Hoerricks, J. (2026a). The geometry of meaning: Five dimensions of coherence. The AutSide. https://autside.substack.com

Hoerricks, J. (2026b). The alcove knew before I did. Sensual Residue. https://sensualresidue.substack.com

Hoerricks, J. (2026c). Quarter-turns: Enough surface area to believe itself. Sensual Residue. https://sensualresidue.substack.com

Hoerricks, J. (2026d). Hidden in the criteria: re-reading autism measures for evidence of Gestalt Language Processing. In Towcester Abbey Research Brief / Occasional Paper Series (Version 1.0, pp. 1–3). Towcester Abbey. https://doi.org/10.5281/zenodo.21281975

Lind, S. E., Williams, D. M., Bowler, D. M., & Peel, A. (2014). Episodic memory and episodic future thinking impairments in high-functioning autism spectrum disorder: An underlying difficulty with scene construction or self-projection?. Neuropsychology, 28(1), 55.

Rubin, D. C., Deffler, S. A., & Umanath, S. (2019). Scenes enable a sense of reliving: Implications for autobiographical memory. Cognition, 183, 44-56.