The Diagnosis is Not Wrong.
The Map is.
How library science, phenomenology, and one Indian mathematician reveal why psychiatric classification keeps failing, and what to do instead.
Ranganathan · Merleau-Ponty
Heidegger · Levinas
HiTOP · RDoC · Network Theory
P : M : E : S : T
Something is wrong with psychiatric diagnosis. The field knows it.
Comorbidity rates that defy probability. Categories so heterogeneous that two people sharing a diagnosis share no symptoms. Edition after edition of the DSM reshuffling the same distress into new containers.
This is usually treated as an empirical problem: we have not found the right categories yet. Researchers propose new research frameworks, new dimensional axes, new statistical clusters (Insel et al., 2010). The DSM revises. The instability persists.
Diagnostic categories show high comorbidity that no categorical model would predict (Kessler et al., 1994), and heterogeneity within categories so severe that two individuals sharing a diagnosis may share no symptoms (Zachar, 2000). The p-factor (a statistical finding that a single general psychopathology dimension underlies most diagnostic variance) is the empirical shadow of this problem (Caspi et al., 2014).
What if the problem isn't empirical? What if it is structural: a mismatch between the kind of thing suffering is and the kind of container the DSM uses to hold it? Andreasen (2007) argued that the operationalized criteria of DSM-III effectively displaced phenomenological description from American psychiatric practice, producing an unintended narrowing of the clinical gaze.
"Two patients who both meet criteria for Major Depressive Disorder may share not a single symptom. The category is a point in a name-space, not a description of a thing." — After Zachar (2000), Psychiatric Disorders Are Not Natural Kinds
The argument here is that the DSM relies on an Aristotelian genus–species model of classification — what Bowker and Star call "infrastructural classification" whose categories become naturalized and invisible (Bowker & Star, 1999) — the same model that Shiyali Ramamrita Ranganathan spent his career demonstrating was inadequate for any domain of complex knowledge.
The DSM is a tree. Human suffering is not.
Enumerative classification (the genus-species model) places every entity at a single location on a branching hierarchy. It assumes that belonging to a category determines what other properties you have.
A tree works when categories are natural kinds: this is a dog, not a wolf; this is gold, not iron (Murphy, 2006). It fails when the domain is constitutively multidimensional: knowing one fact about an entity tells you almost nothing about the others. Wakefield's harmful dysfunction analysis exposed this: psychiatric categories conflate statistically co-occurring symptoms with natural-kind disease entities (Wakefield, 1992). Robins and Guze's original criteria for diagnostic validity (criteria the DSM has never fully met) required demonstration that categories carve nature at its joints (Robins & Guze, 1970).
⚠Heterogeneity within nodes: two MDD patients may share zero symptoms
⚠Edition-to-edition reshuffling: PTSD moved, BPD debated, ASD spectrum expanded
⚠Location in tree assumes a superordinate dimension. But which dimension is primary?
Forced Hierarchy
The tree demands a single "most important" dimension. Is PTSD a trauma disorder or an anxiety disorder? Both. But the tree forces a choice.
No Hospitality
New presentations require new branches. The system cannot accommodate novelty without restructuring. Hence edition after edition.
Combinatorial Blindness
A label tells you the branch, not the person. Two nodes in the same branch may be as different as two nodes in different branches.
S.R. Ranganathan and the Colon Classification
In 1933, Ranganathan published a classification scheme for library science unlike anything before it (Ranganathan, 1933). Instead of placing each book on a single branch of a tree, he described books as coordinates in multidimensional space.
His insight: complex subjects are not single entities to be located in hierarchies. They are combinations of values drawn from multiple independent dimensions: what he called facets. He elaborated the theoretical foundations of this system in the Prolegomena to Library Classification (Ranganathan, 1967).
Notice what this does: no question of whether the book "belongs" to economics or geography or history. It specifies all simultaneously. New combinations require no restructuring, only new coordinates. Ranganathan called this hospitality. Tennis (2002), analyzing how classification systems accommodate the emergence of new subjects over time, found that faceted systems handle this far more gracefully than enumerative ones, a finding directly relevant to DSM revision instability.
Ranganathan named five fundamental facets applicable to any domain of knowledge (the PMEST formula) and argued they were not arbitrary but reflected something deep about how any domain of complex knowledge is actually structured.
PMEST meets Phenomenology
Ranganathan's five facets were developed as an epistemological tool for organizing knowledge. Phenomenology, independently, arrived at a strikingly parallel set of dimensions for describing the structure of lived experience itself. Merleau-Ponty's analysis of the body-subject (Merleau-Ponty, 1945/2012), Heidegger's existential analytic (Heidegger, 1927/1962), and the phenomenological psychopathology tradition inaugurated by Jaspers (Jaspers, 1913/1963) and continued by Sass and Parnas (Sass & Parnas, 2003) all converge on the same structural insight: lived experience is irreducibly multidimensional. The convergence with Ranganathan is not coincidental.
Both schemas are finite in their dimensions but infinite in their combinations. Both are anti-reductive: you need all facets to locate anything. Both are synthetic rather than analytic. The epistemology recapitulates the ontology: the library was built by beings with a lifeworld, and its deep grammar reflects that. Ratcliffe's analyses of existential feelings confirm that alterations in basic temporal and bodily experience constitute rather than merely accompany psychiatric disturbance (Ratcliffe, 2008).
"Trauma disrupts not one thing but the synthetic unity of a life: temporality collapses, embodiment becomes unsafe, spatiality contracts, relational field ruptures, selfhood fragments. Healing is re-coordination across facets: not fixing a category but restoring a location." — Sorg (2026, unpublished manuscript), drawing on van der Hart et al. (2006) and Merleau-Ponty (1945/2012)
Recent transdiagnostic models such as HiTOP (Kotov et al., 2017), RDoC (Insel et al., 2010), and network theory (Borsboom & Cramer, 2013) represent movement toward multidimensional modeling, but lack a coherent classificatory logic. The most philosophically developed response to this gap has been homeostatic property cluster (HPC) theory, which proposes that mental disorders are held together not by essential definitions but by mutually reinforcing causal mechanisms (Fried, 2025). HPC theory identifies what is wrong with enumeration; it does not specify how to build a better system. Ranganathan's faceted framework supplies that operational logic: dimensions are universal, combinations are infinite, coordinates are always revisable.
Meet A.: the same person, two maps
The following composite vignette illustrates what changes when we shift from categorical to faceted description. All identifying details are fictional. The clinical logic is real.
A. is a 34-year-old presenting with chronic sleep disruption, emotional volatility, difficulty maintaining employment, and what she describes as "not feeling like a real person." She reports a childhood marked by unpredictable caregiving and a recent relationship ending in abuse.
She endorses depressed mood most days, difficulty concentrating, recurrent intrusive memories, hypervigilance in public spaces, and periods of feeling "outside herself." She has previously been diagnosed with Major Depressive Disorder and Generalized Anxiety Disorder. Her current clinician is considering adding PTSD and querying Borderline Personality Disorder.
A. herself says: "I just want to feel like I'm actually here."
Click each label to see what symptoms justify it and what it misses.
Four labels. Each places A. on a different branch of the tree, branches that were designed independently and do not speak to each other. The comorbidity is not a finding about A.; it's a symptom of the classificatory model's inability to hold her as a unified person. The "r/o BPD" signals that the system is straining at its seams. Notice that none of these labels answers her question: "Why don't I feel real?"
Click each facet to reveal A.'s coordinates one dimension at a time.
Build your own faceted profile
Select values across all five dimensions. Any combination of three or more will generate a profile and a DSM comparison.
What follows: clinically, ethically, philosophically
If diagnosis is fundamentally a classificatory act, and the classificatory model is structurally inadequate, then improving diagnosis requires not better measurement but a different ontology.
Haslam (2000) has documented empirically that reification of diagnostic categories (treating them as natural kinds) produces measurable clinical and research harms. Foucault's genealogical analyses of psychiatric power show how classification systems become self-reinforcing infrastructures that shape what clinicians are even able to see (Foucault, 1961/2006). The ethical stakes are not only epistemic.
Assessment as location
The clinical intake becomes an act of coordinate-finding, not category-assignment. Where is this person along dimensions of temporality, embodiment, relational field, self-continuity? Different questions produce different seeing.
Hospitality to novelty
A faceted system accommodates new presentations without requiring revision of the whole structure. No new branch needed. Only new coordinates. The system is epistemically humble by design.
Ethical demand
Levinas: the face of the other exceeds any concept we impose on it. A diagnostic label that forecloses re-description commits an ethical violence. Faceted profiles are provisional by structure; they invite revision.
Beyond HiTOP and RDoC
Dimensional models represent progress, but they still fix the axes in advance. Ranganathan's framework supplies the missing classificatory logic: the dimensions are universal, the combinations are infinite, and the coordinates are always revisable.