The Representational Attraction Map

A cross-ontological, presentation-sensitive framework for what a mind can find attractive

Position note and research proposal
Target ontology, morphology, agency, representation, context, and multidimensional response

Andraž Đurič

Version 1 · August 2026

Abstract. Human attraction is usually measured inside a preselected target class: men and women, faces, fictional characters, anthropomorphic animals, robots, or objects. This makes each literature locally informative but leaves a basic question under-specified: what is the common domain over which an individual mind can direct attraction? This essay proposes a Representational Attraction Map, not as a claim that attraction is one scalar or that all targets lie on one simple continuum, but as a framework joining two things: a factorized ontology of represented targets and a person-, time-, context-, and presentation-sensitive response field over that ontology. The proposal separates entity kind, morphology, sex and gender cues, attributed mentality, referential status, presentation mode, target relation, and context. It also separates sexual attraction, felt arousal, romantic pull, aesthetic appreciation, affiliation, fascination, aversion, and behavior. The novelty claim is deliberately narrow. Lovemaps, multidimensional sexuality models, human desire maps, face-spaces, and studies of attraction to fictional characters, anthropomorphic animals, robots, and objects all supply important predecessors. The proposed contribution is their cross-class integration, an explicit representation layer, and a comparative empirical program. That contribution remains a hypothesis until it improves prediction over simpler, domain-specific alternatives.

1. The mapping problem

A map of attraction must first specify its domain: what can function as a target, which properties describe that target, and which responses are being mapped. Most research makes those decisions locally. It may compare attraction across human sex or gender categories, navigate a space of human faces, examine attachment to fictional characters, or study responses to robots, anthropomorphic animals, or inanimate objects. Those restrictions are often methodologically necessary, but they leave the regions without a shared coordinate system.

The missing comparison is not simply between named target classes. The same morphology can be represented as biological or artificial; the same identity can be presented through text, stylised imagery, photorealistic rendering, or direct interaction; similar forms can differ in attributed agency; and the same target can elicit sexual, romantic, aesthetic, affiliative, fascinating, and aversive responses in different combinations. A general map therefore requires explicit coordinates for both the represented target and the response, rather than treating familiar category labels as the domain's natural boundaries.

The central hypothesis is not “people are attracted to anything.” It is that attraction is evaluated over internally represented targets, and that the geometry of those representations need not respect the ordinary biological and ontological categories used to label them.

The phrase geometry is provisional. A map requires dimensions, neighbourhoods, or at least empirically useful similarity relations. It does not follow that the space is Euclidean, continuous, low-dimensional, or shared unchanged between people. Those are questions for measurement.

2. What is being claimed, and what is not

The term lovemap already has a history. John Money used it for a developmental template of the ideal lover and erotic programme. Multidimensional accounts of sexual orientation also long predate this essay. Storms represented homoerotic and heteroerotic responding as partly independent dimensions. The Klein Sexual Orientation Grid separated attraction, behaviour, fantasy, emotional and social preference, lifestyle, and identity across time. Diamond distinguished the systems of sexual desire and romantic love. Sexual Configurations Theory expanded the parameters by which partnered sexualities can be described. Zielona-Jenek proposed a particularly relevant extended description of sexual preferences that separates content, attractiveness, aversiveness, breadth, temporal change, awareness, and coherence among subjective, physiological, and behavioural response.

Nor is “map of desire” a new phrase. Bailey and colleagues published a study with that title in 2021, using multidimensional scaling to recover sex and maturity dimensions within a restricted set of human target categories. Face-space research has reconstructed individual preferred regions in high-dimensional facial morphology. Separate empirical literatures document attachment or attraction directed toward fictional characters, anthropomorphic animals, robots, and inanimate objects.

Novelty firewall. This essay does not claim to discover nonhuman, fictional, artificial, or object-directed attraction; coin the idea of a lovemap or desire map; invent multidimensional sexuality; or provide the first model of individual preference in a stimulus space. Its proposed contribution is narrower: a representation-first, cross-class target ontology coupled to a multidimensional individual response field, with explicit bookkeeping for dimensions that neighbouring literatures usually study separately.

A targeted literature audit found no prior framework that clearly combines all of the following: entity kind, morphology, attributed agency and sentience, actual and perceived referential status, presentation mode, target relation, context, and multiple response types across human and nonhuman target classes. A negative search result cannot prove priority. The correct standing is therefore provisional integration claim, not established first discovery.

3. The first repair: stop forcing unlike distinctions into one hierarchy

An early version of the idea used a chain resembling “human ⊂ humanoid ⊂ fictional ⊂ total.” That chain mixes different axes. A fictional human and an actual human may share morphology but differ in referential status. A humanoid robot and a humanoid organism may share a body plan but differ in substrate. A text description and a rendered image can present the same fictional character in different modes. A car is not a larger class containing humanoids; it is a different entity and morphology region.

One legitimate inclusion can be stated on the morphology axis alone:

Human morphology ⊂ humanoid morphology ⊂ embodied morphology.

“Actual,” “fictional,” “simulated,” “biological,” “artificial,” “agentic,” and “inanimate” must not be inserted into that same chain. They answer different questions.

3.1 A factorized target ledger

Axis	Question answered	Illustrative values	Why it must stay separate
Entity kind or substrate	What sort of represented thing is this?	Biological organism, artificial agent, artifact, place or structure, event, abstract or indeterminate target	A humanoid organism and humanoid robot can match in shape while differing in constitution.
Morphology	What body plan and features are represented?	Human, near-human, humanoid, anthropomorphic, radically nonhuman, non-bodied; plus continuous feature values	Taxonomic labels hide graded shape similarity and feature interactions.
Sex, gender, and maturity cues	Which socially and biologically interpreted cues are present?	Masculine, feminine, mixed, absent, uncertain; adult-coded maturity for the proposed experiments	These cues can vary while body plan and entity kind remain fixed.
Attributed mentality	What mind is the participant treating the target as having?	Agency, sentience, personality, intention, reciprocity, moral patiency	Human-like appearance does not guarantee perceived mind, and nonhuman appearance does not preclude it.
Referential status	Is there an external referent, and what does the participant believe?	Actual referent, no actual referent, uncertain; believed actual, fictional, simulated, hypothetical, or ambiguous	Actual status and perceived status can diverge. The mind responds using the information available to it.
Presentation mode	How is the target encountered?	Direct perception, photograph, video, photorealistic render, stylized image, text, voice, imagination, virtual reality	The same represented identity can elicit different responses across media.
Target relation and location	Where is the desired target located in the scenario?	External other, self-as-target, partner-like other, object within a scenario, distributed or uncertain target	Attraction to being a target is not identical to attraction toward an external target.
Context and history	What changes the current interpretation?	Narrative, behaviour, familiarity, culture, learning, relationship, novelty, mood, current physiological state	A context-free target is an experimental abstraction, not ordinary cognition.

The ledger is not asserted to be exhaustive. Its purpose is to block avoidable category errors and to make omissions visible. Activities, situations, power relations, and interaction patterns can be represented in the context term or developed as adjacent modules. This essay remains centred on target-directed attraction rather than pretending to be a complete theory of sexuality.

4. The deepest domain is represented targets

Attraction is not a property read directly from an external object. A person receives a presentation, interprets it, fills gaps, attributes a mind and identity, and responds to the resulting target representation. An external entity may causally constrain that process, but it is neither necessary nor sufficient. Fictional characters have no external organism corresponding to them. A real person can be misperceived. A few lines of text can evoke a rich target. A highly realistic image can evoke less coherent personhood than a stylized drawing.

Let s be the available stimulus or presentation, c the immediate context, and h_p,t the history and current state of person p at time t. The constructed representation is:

r_p,t = G_p,t(s, c, h_p,t).

The function G includes perception, interpretation, memory, imagination, category learning, and uncertainty. If an external referent exists, it influences response through the evidence and history available to this construction. The same stimulus can therefore produce different representations in different people, or in the same person after learning that a supposedly actual target is fictional or artificially generated.

The domain is not “all physical objects,” because a fictional target need not be physical. It is also not “all possible and impossible concepts.” A literally incoherent object cannot be fully represented. What can be represented is a phrase, a partial model, or a shifting approximation. The practical domain is the set of target representations a particular cognitive system can construct with enough stability to support a response.

5. The response is a vector, not one attraction number

Everyday language compresses several responses into the word attraction. That compression is useful in conversation and dangerous in a model. Sexual attraction can occur without romantic attachment. Aesthetic appreciation can occur without either. Felt arousal, physiological response, approach motivation, fantasy, affiliation, nurturance, fascination, and aversion can dissociate. Attraction and aversion can even coexist rather than occupying opposite ends of a single scale.

The response field can be represented provisionally as:

Y_p,t(r,c) = (sexual attraction, felt arousal, romantic pull, aesthetic appreciation, affiliation, fascination, aversion, approach intention).

Physiological measures and behaviour can be added as separate observations rather than treated as transparent readouts of desire. The model is therefore a field from represented targets and contexts to a response vector:

Y_p,t = F_p,t(r_p,t, c) + ε.

The error term ε includes measurement noise and unmodelled state. The subscripts matter. There may be no timeless “true map” stored intact inside a person. A more realistic object of study may be a distribution over maps conditional on state and history.

Any claim about a “most attractive target” must select a response coordinate or declare a weighting over coordinates. Otherwise two targets can be incomparable: one higher in sexual attraction, another higher in romantic pull, a third higher in aesthetic appreciation. A scalar maximum is not licensed by a multidimensional response alone.

6. Peaks, basins, cliffs, islands, and valleys

If a useful similarity relation can be learned, the map may exhibit:

• peaks or attractors, local regions producing unusually strong response;
• basins, families of variations that remain attractive despite feature changes;
• cliffs, small representational changes that produce large response changes;
• valleys, including uncanny or incoherent regions between otherwise acceptable forms;
• disconnected islands, attractive target classes with no smooth attractive path between them;
• context warping, where narrative, behaviour, or presentation changes the effective neighbourhoods;
• temporal drift, where learning, repeated exposure, relationships, or state alter the surface.

These terms should not be mistaken for discoveries merely because they are mathematically suggestive. A basin exists only relative to a metric or transition procedure. If two different encodings produce incompatible neighbourhoods and neither predicts held-out responses, the geometric language is decorative. Learning a useful representation and distance function is one of the proposal's load-bearing problems.

7. What the closest predecessors already establish

Predecessor	What it contributes	Boundary relative to this proposal
Money's lovemap	An internal template concerning an ideal lover and erotic programme.	Historically foundational, but developmental and clinical rather than a formal cross-class ontology with explicit presentation and response axes.
Storms and the Klein Grid	Sexual orientation is not adequately represented by one bipolar line; response types and time periods can be separated.	The principal target contrasts remain human sex or gender categories.
Diamond and Sexual Configurations Theory	Romantic love and sexual desire should be distinguished; partnered sexualities can be described through multiple parameters.	They broaden response and partner description without developing the present cross-ontological target ledger.
Zielona-Jenek's extended model	Separates who or what is preferred, attractiveness, aversiveness, breadth, change, awareness, and coherence across response systems.	This is the closest descriptive bookkeeping precedent found. Its clinical purpose does not factor morphology, agency, referential status, and presentation into a shared target space.
Human target desire maps	Global morphological similarity and multidimensional scaling can organize responses across human sex and maturity categories.	Restricted target axes, measures, and mostly clinical or forensic samples limit cross-class inference.
Individual face-space studies	People can navigate high-dimensional face spaces toward individually preferred regions.	Facial attractiveness is not identical to sexual attraction, and the tested spaces remain within human-like facial geometry.
Objectum sexuality, mechanophilia, and fictophilia studies	Some people report emotional, romantic, or sexual relations to objects or fictional characters; fictional status and attributed personality can matter. A recent forensic case also documents car-directed mechanophilia.	Small, self-selected, qualitative, or single-case evidence establishes possible phenomena and meanings, not prevalence, a general mechanism, or a common metric across target kinds.
Furry and target-location research	Anthropomorphic targets and fantasies of the self as target complicate a simple “which external partner?” model.	Sample-specific results and contested theoretical interpretations should not be generalized wholesale.
Robot, animacy, and artificial-partner research	Human-like appearance, attributed agency, interactivity, realism, and illusion can be manipulated separately.	The literature is focused on artificial agents and does not supply a domain-general target ontology.
Sexual-prior theories	Stimuli may be evaluated against internal learned or innate representations rather than only surface labels.	A representational mechanism is proposed, but not the cross-class descriptive and experimental architecture developed here.

The proposal should be judged as a synthesis that makes new comparisons possible. If its only achievement is to rename the union of these literatures, it has failed. It must compress observations, expose missing variables, or improve prediction.

8. A precise version of the post-human optimum hypothesis

Suppose sexual attraction is the selected response coordinate and all other relevant conditions are fixed. Let H be the set of coherently represented human morphologies and U a broader set of coherently represented humanoid morphologies, with H ⊂ U. Then, for person p:

sup_{u ∈ U} S_p(u,c) ≥ sup_{h ∈ H} S_p(h,c).

This is true by set inclusion, assuming the same response function and context. The use of supremum avoids assuming that a maximum exists. The scientifically interesting claim is the strict inequality:

sup_{u ∈ U} S_p(u,c) > sup_{h ∈ H} S_p(h,c).

That statement does not follow from the framework. It is the post-human optimum hypothesis: for at least some people and conditions, the highest attainable response in a broader, coherent humanoid morphospace lies outside the biologically human region.

Several forces could make the inequality strict: exaggerated or recombined preferred features, removal of weakly preferred constraints imposed by human development, or forms made meaningful by art and narrative. Several forces could make equality hold: attraction may be strongly calibrated to natural human configurations; departures may reduce identity, animacy, or coherence; or the best nonhuman design may merely approximate an already attainable human target. An uncanny valley can also make the path from human to nonhuman non-monotonic.

Set inclusion establishes only the weak inequality. Evidence for strict inequality would require an empirical search across coherent human and nonhuman stimuli under matched conditions. Until such evidence exists, the post-human optimum remains a low-confidence research hypothesis, not an inference licensed by the map itself.

9. A comparative empirical programme

9.1 Start within morphology, then cross domains

The cleanest first study would use adult-coded targets only and manipulate a coherent latent character rather than collecting unrelated images. A generative three-dimensional model could vary eye scale and placement, nasal structure, ear form, cranial proportions, skin properties, facial symmetry, body plan, movement, and other features while preserving identity as far as possible. Participants would make adaptive pairwise choices and separate absolute ratings. Bayesian optimisation or an evolutionary search procedure could locate high-response regions efficiently, while periodic repeated trials would measure reliability.

The experiment should not force every target onto one human-to-alien line. After estimating a near-human and humanoid basin, it should branch into additional manifolds: anthropomorphic animal forms, artificial agents, non-agentic artifacts, places or structures, and non-bodied targets. Disconnected regions are a possible result, not a defect to be smoothed away.

9.2 Cross presentation without pretending the stimulus is unchanged

The same latent target should be presented through stylized art, photorealistic rendering, short animation, text, voice, and, where feasible, immersive simulation. The purpose is not to assume these are equivalent stimuli. It is to estimate how presentation changes the constructed representation and response. A stylized face with enormous eyes may be coherent in illustration but anatomically alarming when translated literally into a realistic render. A better photorealistic translation may preserve the relational or expressive features without preserving every pixel-level proportion.

Reports about an imagined presentation and ratings of an externally generated presentation must therefore be recorded as different observations. A change in response may reflect a translation failure, genuine presentation sensitivity, or a change in the represented target; the design should not decide among those explanations in advance.

9.3 Manipulate mind while holding appearance as constant as possible

A character can be described as scripted, autonomous, sentient, reciprocal, indifferent, hostile, or incapable of experience. Interactive behaviour and narrative history can be varied while the visual model remains fixed. Conversely, appearance can vary while behaviour and biography remain fixed. Robot and animacy research makes this separation especially important: human-like appearance and attributed agency are correlated in ordinary experience but conceptually and experimentally distinct.

9.4 Measure the response vector

At minimum, each target should receive separate measures of sexual attraction, felt arousal, romantic pull, aesthetic appreciation, affiliative interest, fascination, aversion, and uncertainty. Behavioural choice and physiological measures can be collected where ethical and useful, but disagreement among measures should be treated as data. It should not be “resolved” by declaring one channel the real preference.

9.5 Repeat, predict, and compare models

A map earns its name if it predicts. Sessions should be repeated to estimate within-person stability. A model fitted to some targets and presentations should predict held-out responses to new combinations. The factorized common-space model should be compared against:

• flat category labels such as human, robot, furry, fictional, or object;
• separate domain-specific models with no shared representation;
• a simple human-likeness scalar;
• presentation-only and familiarity-only baselines;
• high-capacity black-box predictors that test whether interpretability costs too much accuracy.

The decisive test is out-of-sample performance and explanatory economy, not whether the visualisation looks map-like.

9.6 Control the obvious confounds

Image quality, identity drift, pose, lighting, sexualisation, movement quality, familiarity, novelty, genre association, cultural learning, demand characteristics, scale use, and the participant's current state can all masquerade as target-space effects. Generated targets may contain artifacts that selectively penalise nonhuman conditions. Narrative conditions may change attention rather than target representation. Photorealism may introduce uncanniness. The study needs matched stimuli, manipulation checks, preregistered exclusions, repeated measurements, and open reporting of failed translations.

10. Predictions that distinguish the framework from a catalogue

1. Feature sharing should sometimes beat category labels. A nonhuman target sharing a person's preferred morphology and mind cues should be more predictable from those features than from the label “nonhuman.”
2. Presentation changes should be person-specific. Translation from stylized to realistic form will preserve attraction for some participants and collapse it for others, partly according to how each person reconstructs identity, animacy, and coherence.
3. Agency will interact with morphology rather than simply add to it. Attributed agency may increase romantic or affiliative response while leaving sexual attraction unchanged, or may rescue one morphology and damage another.
4. Attraction and aversion will not behave as exact opposites. Some uncanny targets will produce simultaneous interest and aversion; a one-dimensional valence scale will lose predictive information.
5. Some maps will contain disconnected islands. Object-directed or highly stylised target regions may not connect smoothly to the person's human-face optimum.
6. The apparent optimum will depend on the selected response. A target optimising sexual attraction need not optimise romantic, aesthetic, or affiliative response.
7. Stable traits and changing states will both matter. Repeated sessions should reveal more within-person similarity than random reassignment if a personal map is useful, while still showing systematic state and learning effects.

11. Claim ledger

ID	Claim	Standing	Confidence
C1	Human, fictional, anthropomorphic, artificial, and inanimate represented targets can support attraction or attachment in at least some people.	Empirical existence claim; does not establish prevalence or one mechanism.	High
C2	Sexual attraction, arousal, romantic pull, aesthetic appreciation, affiliation, aversion, and behaviour should not be collapsed by default.	Strong conceptual and empirical precedent.	High
C3	Entity kind, morphology, agency, referential status, presentation, and target relation are distinct bookkeeping axes.	Conceptual claim that prevents category errors; operational independence remains empirical.	High
C4	No established prior framework integrates all proposed axes across the full target range.	Negative result from a targeted, non-systematic audit; load-bearing for novelty.	Moderate
C5	A factorized common representation will predict held-out responses better than flat labels or isolated domain models.	Core empirical hypothesis; load-bearing for usefulness.	Moderate to low
C6	Individual maps will show meaningful stability across sessions.	Testable hypothesis; stability may be conditional rather than global.	Moderate
C7	For some people, the optimal coherent humanoid target lies outside human morphology.	Post-human optimum hypothesis; coherent and indirectly motivated, but lacking direct optimisation evidence.	Low
C8	One canonical distance metric or scalar attraction score can organise the whole domain.	Not assumed; plausible only in restricted tasks.	Low

12. Load-bearing joints

The representation problem. The framework says response is directed at a constructed target, but the construction is latent. Self-report, similarity judgements, neural or behavioural measures, and model-based stimulus features are imperfect indicators. If the inferred representation merely redescribes the response, the explanation becomes circular.

The metric problem. “Near,” “far,” “basin,” and “cliff” require a distance or transition rule. Pixel distance is psychologically poor; semantic embeddings import unknown training biases; hand-selected axes risk confirming the theory. The metric must be validated by prediction and perturbation.

The coverage problem. A finite target ledger can omit decisive dimensions. The correct response to an unrepresented distinction is to revise the ontology, not force the observation into the nearest box.

The scalarisation problem. Optimisation requires an outcome. A result changes when sexual attraction, romantic pull, aesthetic appreciation, or a weighted combination is selected. There may be a Pareto frontier rather than one best target.

The stability problem. If within-person responses vary almost as much as between-person responses, the trait-like map metaphor will be misleading. A conditional, dynamic field may still work, but it is a different claim.

The coherence problem. Manipulating one feature can unintentionally alter identity, realism, animacy, genre, or plausibility. A “nonhuman morphology penalty” may really be a poor-rendering penalty.

13. What would falsify or materially revise the proposal?

• Discovery of an earlier framework that already integrates the stated axes across human and nonhuman targets would defeat the novelty claim and require this essay to be reframed as exposition or extension.
• If factorized cross-class models do not outperform flat labels, human-likeness scales, or separate domain models on held-out data, the proposed common ontology is not earning its complexity.
• If within-person response structure is not reproducible beyond measurement noise, the stable personal-map interpretation should be rejected.
• If shared features never support cross-domain generalisation, the assumption of a useful common space should yield to plural domain-specific spaces.
• If inferred topology changes arbitrarily with reasonable feature encodings, claims about basins, cliffs, and optima should be withdrawn.
• If crucial target distinctions repeatedly fall outside the ledger, the present axes should be treated as an inadequate first draft, not defended as exhaustive.
• If presentation changes response without detectable changes in target representation, the simple two-stage representation-first model will need additional direct pathways from medium and context to response.

A weaker outcome is also possible: the framework may be useful only within a restricted family, such as humanoid faces across visual styles. That would still be an empirical contribution, but not the domain-general map proposed here.

14. Scope and ethical firewall

A descriptive coordinate is not an endorsement. Attraction, fantasy, intention, and behaviour are distinct. Consent, capacity, legality, and harm constrain action independently of whether a response can be represented or measured. The proposed empirical programme should use consenting adult participants and adult-coded target stimuli.

The framework should also resist unnecessary pathologisation. Rare target patterns are not disorders merely because they are rare. Clinical significance depends on distress, impairment, harm, and applicable diagnostic criteria, not category distance from the population mode. Conversely, neutral description does not erase real risks in behaviour or relationships. The map's task is bookkeeping and prediction, not moral adjudication.

15. Open questions

1. Are target representations best modelled as continuous vectors, structured graphs, generative programmes, or mixtures of discrete and continuous variables?
2. Which axes are psychologically primitive, and which are labels summarising lower-level features?
3. Can one person's metric be aligned with another's without erasing genuinely idiosyncratic structure?
4. How much of a map transfers across image, text, voice, imagination, and interactive embodiment?
5. When does attributed agency expand attraction, and when does it introduce expectations that suppress it?
6. Do disconnected target islands reflect separate learning histories, separate motivational systems, or only poor sampling between them?
7. Can adaptive search distinguish a genuine post-human optimum from novelty, exaggeration, or generator artifacts?
8. How does the map change through familiarity, attachment, identity development, culture, and deliberate imagination?
9. Which response dimensions require their own spaces rather than different values on a shared target representation?

16. Standing of this essay

Document type: Conceptual synthesis and research proposal.

Evidence base: Targeted, non-systematic audit of adjacent scientific literatures, conceptual analysis, and explicitly marked empirical hypotheses.

Strongest contribution: Separation of commonly mixed target axes and their connection to an explicit representation layer and multidimensional response field.

Strongest precedent: Zielona-Jenek's extended descriptive model, followed by multidimensional human desire mapping and individual face-space optimisation.

Primary unresolved burden: Demonstrating that a shared factorized space predicts new responses better than simpler or domain-specific models.

Novelty confidence: Moderate and provisional. The audit supports a gap; it does not prove absolute priority.

Framework confidence: High for the need to separate the listed concepts; moderate to low for the proposed common geometry; low for any universal post-human optimum.

17. Conclusion

The Representational Attraction Map begins from a modest correction: minds do not respond to ontology labels alone. They respond to constructed targets whose morphology, attributed mind, identity, status, medium, and context can cross the boundaries of human, fictional, artificial, and inanimate categories. Existing research has already mapped important pieces of this terrain. The task is to connect them without claiming their discoveries as new.

The resulting framework is deliberately more demanding than a list of unusual attractions. It requires a target ledger that does not mix axes, a response vector that does not collapse distinct experiences, an explicit model of presentation and interpretation, and tests against simpler alternatives. It also converts the intuition of a “post-human optimum” into a clean but low-confidence hypothesis: expanding a coherent morphospace cannot lower its supremum, but whether the best point actually lies outside the human region is an empirical question.

If the framework succeeds, it will reveal common structure across target classes while preserving meaningful differences. If it fails, its falsifiers show where: the common axes will not generalise, the maps will not stabilise, or the geometry will depend on arbitrary encodings. Either result would be more informative than continuing to mistake familiar categories for the boundaries of attraction itself.

References

1. Money, J. (1986). Lovemaps: Clinical Concepts of Sexual/Erotic Health and Pathology, Paraphilia, and Gender Transposition in Childhood, Adolescence, and Maturity. Irvington. VERIFIED · AUGUST 2026
2. Storms, M. D. (1980). Theories of sexual orientation. Journal of Personality and Social Psychology, 38(5), 783–792. VERIFIED · AUGUST 2026
3. Klein, F., Sepekoff, B., & Wolf, T. J. (1985). Sexual orientation: A multi-variable dynamic process. Journal of Homosexuality, 11(1–2), 35–49. VERIFIED · AUGUST 2026
4. Diamond, L. M. (2003). What does sexual orientation orient? A biobehavioral model distinguishing romantic love and sexual desire. Psychological Review, 110(1), 173–192. VERIFIED · AUGUST 2026
5. van Anders, S. M. (2015). Beyond sexual orientation: Integrating gender/sex and diverse sexualities via Sexual Configurations Theory. Archives of Sexual Behavior, 44, 1177–1213. VERIFIED · AUGUST 2026
6. Zielona-Jenek, M. (2018). Sexual preferences and associated disorders: Toward an extended model for description. Psychiatria Polska, 52(6), 1063–1073. VERIFIED · AUGUST 2026
7. Blanchard, R., Kuban, M. E., Blak, T., Klassen, P. E., Dickey, R., & Cantor, J. M. (2012). Sexual attraction to others: A comparison of two models of alloerotic responding in men. Archives of Sexual Behavior, 41, 13–29. VERIFIED · AUGUST 2026
8. Bailey, J. M., Blanchard, R., Hsu, K. J., & Revelle, W. (2021). A map of desire: Multidimensional scaling of men's sexual interest in male and female children and adults. Psychological Medicine, 51(15), 2714–2720. VERIFIED · AUGUST 2026
9. Ibáñez-Berganza, M., Amico, A., & Loreto, V. (2019). Subjectivity and complexity of facial attractiveness. Scientific Reports, 9, 8364. VERIFIED · AUGUST 2026
10. Simner, J., Hughes, J. E. A., & Sagiv, N. (2019). Objectum sexuality: A sexual orientation linked with autism and synaesthesia. Scientific Reports, 9, 19874. VERIFIED · AUGUST 2026
11. Osvath, P., Könczöl, F., Fekete, S., & Tényi, T. (2026). Mechanophilia: A case report of a specific fetishistic disorder. Frontiers in Psychiatry, 17, 1749802. VERIFIED · AUGUST 2026
12. Karhulahti, V.-M., & Välisalo, T. (2021). Fictosexuality, fictoromance, and fictophilia: A qualitative study of love and desire for fictional characters. Frontiers in Psychology, 11, 575427. VERIFIED · AUGUST 2026
13. Hsu, K. J., & Bailey, J. M. (2019). The “furry” phenomenon: Characterizing sexual orientation, sexual motivation, and erotic target identity inversions in male furries. Archives of Sexual Behavior, 48, 1349–1369. VERIFIED · AUGUST 2026
14. Szczuka, J. M., Hartmann, T., & Krämer, N. C. (2019). Negative and positive influences on the sensations evoked by artificial sex partners: A review of relevant theories, recent findings, and introduction of the Sexual Interaction Illusion Model. In AI Love You, 3–19. VERIFIED · AUGUST 2026
15. Szczuka, J. M., & Krämer, N. C. (2019). There's more to humanity than meets the eye: Differences in gaze behavior toward women and gynoid robots. Frontiers in Psychology, 10, 693. VERIFIED · AUGUST 2026
16. Crowell, C. R., Deska, J. C., Villano, M., Zenk, J., & Roddy, J. T., Jr. (2019). Anthropomorphism of robots: Study of appearance and agency. JMIR Human Factors, 6(2), e12629. VERIFIED · AUGUST 2026
17. Looser, C. E., & Wheatley, T. (2010). The tipping point of animacy: How, when, and where we perceive life in a face. Psychological Science, 21(12), 1854–1862. VERIFIED · AUGUST 2026
18. Abé, C. (2025). The concept of innate sexual priors in the brain: A theory on why we are attracted to what we are attracted to. Sexuality & Culture, 29, 636–666. VERIFIED · AUGUST 2026

Verification key. “Verified” means that the bibliographic identity and the specific use made of the source were checked against the linked publisher, journal, database, or full-text record in August 2026. It does not certify every claim in the source or imply endorsement of its theoretical framing.

Internal lineage and version history

This essay develops an Epistemic Forge concern with separating world, representation, observation, and inference. It has no prerequisite essay: the framework, its uncertainty, and its correction conditions are stated here.

Version 1, August 2026. Initial public formulation. Added the factorized target ledger, formal representation and response functions, the post-human optimum hypothesis, predecessor audit, empirical programme, claim ledger, load-bearing problems, and explicit falsifiers.

Author: Andraž Đurič, Slovenia.

Developed through the Epistemic Forge corpus and dialogue with ChatGPT (OpenAI). The conceptual commitments, scope decisions, and final editorial responsibility remain the author's. Contributions are assessed by their content rather than their origin.

License: CC BY 4.0.