research / vela / bibliography
Literature map (cross-thread)
Map of the field across the major Vela research threads.
Vela·Bibliography·source: people-analyst/vela/docs/research/literature-map.md
Literature Map — Vela Research Program
Status: ASN-575, first merged draft
Date: 2026-04-24
Scope: Every theoretical coordinate in docs/RESEARCH-PROGRAM.md §I; every scoring component in docs/engine-room/01-math-spec.md §§2–5; ≥1 supporting reference per desire dimension; instrument cross-walk (Section J); Reincarnation novelty analysis (Section K); downstream-artifacts index (Section L).
Method: Merged and reconciled from two parallel drafts (OpenAI ChatGPT and Claude.ai Opus 4.7 1M context, both run 2026-04-24). DOIs verified against publisher pages or CrossRef search results.
Citation style: APA 7 inline; BibTeX keys lastname_year_firstword in bibliography.bib.
DOI policy: DOIs and URLs stated only when verified to resolve. Unverified entries are flagged inline or omitted.
Column legend:
- Coordinate: the theoretical construct or Vela engineering component being positioned
- Vela measure: the concrete Vela table / column / scoring component
- Primary source: the canonical citation (APA 7)
- DOI / URL: verified publisher or stable identifier
- Relevance: why this source matters for this coordinate (≤40 words)
- Vela divergence or extension: what Vela does differently from the cited work
- Absorption candidate: specific variable, scale, method, or control we could port into Reincarnation — see Section K for the synthesis
- Meta-analyzable? whether the paper's effect sizes could be extracted into a formal meta-analysis if Vela attempted one — honest
nofor most theoretical reviews and neuroimaging papers
A. Aesthetic desire as a construct (RESEARCH-PROGRAM §1.1)
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Aesthetic desire as distinct construct — collative variables, novelty, complexity, ambiguity | SignalDiversity (§2.2); exploration budget | Berlyne, D. E. (1971). Aesthetics and psychobiology. Appleton-Century-Crofts. | WorldCat | Classical statement that novelty, complexity, and ambiguity drive aesthetic arousal and interest. | Vela treats heterogeneous response patterns as measurement value, not merely stimulus-side arousal. | Collative-variable taxonomy as a curator-side tag set for stimulus construction. | No (book, theoretical). |
| Aesthetic desire as distinct construct — appraisal beyond simple pleasure | Utility (§2.1); DesireScore | Silvia, P. J. (2005). Emotional responses to art: From collation and arousal to cognition and emotion. Review of General Psychology, 9(4), 342–357. | 10.1037/1089-2680.9.4.342 | Reframes art response as appraisal-driven emotion, especially interest under challenge and comprehension. | Vela operationalizes desire behaviorally through ratings, saves, and later fit, not only self-reported appraisal. | Interest vs. pleasure appraisal distinction as a within-rating split (confusion vs. understanding). | Maybe (review with some effect sizes). |
| Aesthetic desire as distinct construct — beauty as approach motivation | Utility; DesireScore; Resonates interaction | Armstrong, T., & Detweiler-Bedell, B. (2008). Beauty as an emotion: The exhilarating prospect of mastering a challenging world. Review of General Psychology, 12(4), 305–329. | 10.1037/a0012558 | Argues beauty is energizing, approach-oriented, and mastery-linked rather than calm liking. | Vela's resonance logic tracks this appetitive view and separates desire from flat preference. | "Exhilaration" and "prospect of mastery" as coded subjective tags to pair with rating. | No (theoretical). |
| Aesthetic desire as distinct construct — processing fluency | Utility; structure + classical dimensions | Reber, R., Schwarz, N., & Winkielman, P. (2004). Processing fluency and aesthetic pleasure: Is beauty in the perceiver's processing experience? Personality and Social Psychology Review, 8(4), 364–382. | 10.1207/s15327957pspr0804_3 | Explains why symmetry, prototypicality, contrast, and repetition often increase liking through ease of processing. | Vela partly inherits fluency but also rewards challenge, intensity, and unresolved pull. | Prototypicality and symmetry as decomposition features; millisecond-level response latency as fluency proxy. | Yes (authors report effect sizes across studies). |
| Aesthetic desire as distinct construct — perception–emotion–meaning triad | CandidateScore (§5); dimension_scores; axes | Chatterjee, A., & Vartanian, O. (2014). Neuroaesthetics. Trends in Cognitive Sciences, 18(7), 370–375. | 10.1016/j.tics.2014.03.003 | Canonical review framing aesthetic episodes through interacting perceptual, affective, and meaning-making systems. | Vela turns this triad into separate computational channels: population score, profile alignment, and curation priors. | Three-stage perception → affect → meaning architecture as a structural template for CandidateScore subcomponents. | No (review). |
| Aesthetic desire as distinct construct — self-relevance and intense aesthetic response | extreme-positive Utility; top-pool surfacing; (rating = 5 ∧ saved) | Vessel, E. A., Starr, G. G., & Rubin, N. (2012). The brain on art: Intense aesthetic experience activates the default mode network. Frontiers in Human Neuroscience, 6, 66. | 10.3389/fnhum.2012.00066 | Shows especially moving art engages self-referential/default-mode processing, not only sensory response. | Vela infers self-relevance from saves + strong positive responses rather than neural measures. | "Moving" as a separate binary signal alongside rating/save (one-click). | No (fMRI, N=16). |
| Aesthetic desire as distinct construct — aesthetic emotions are intense, sought, and mixed | intensity dimension; Sensory axis; responses.emotions[] | Menninghaus, W., Wagner, V., Wassiliwizky, E., Schindler, I., Hanich, J., Jacobsen, T., & Koelsch, S. (2019). What are aesthetic emotions? Psychological Review, 126(2), 171–195. | 10.1037/rev0000135 | Defines aesthetic emotions as intrinsically rewarding, intense, and motivational, spanning pleasure, awe, being-moved, and fascination. | Vela compresses this richer taxonomy into scalar desire, dimensions, and boundary signals. | Forced-choice emotion tags replacing free-text emotions[] — would enable emotion-specific regression. | Maybe (review; constructs useful for downstream meta). |
| Aesthetic desire as distinct construct — negative aesthetic emotions | boundary_flag; BoundaryTags | Silvia, P. J., & Brown, E. M. (2007). Anger, disgust, and the negative aesthetic emotions: Expanding an appraisal model of aesthetic experience. Psychology of Aesthetics, Creativity, and the Arts, 1(2), 100–106. | 10.1037/1931-3896.1.2.100 | Establishes anger and disgust as real aesthetic responses, not non-aesthetic failures. | Vela formalizes these as durable boundary evidence and tag production. | Typology distinguishing anger, disgust, confusion, boredom — finer-grained boundary_flag categories. | Maybe (small-N empirical + review). |
B. The figurative body as a special case (RESEARCH-PROGRAM §1.2)
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Embodied simulation — viewing depicted poses recruits motor/somatosensory cortex | visual_decompositions.pose_type; gesture; figurative-body decomposition | Freedberg, D., & Gallese, V. (2007). Motion, emotion and empathy in esthetic experience. Trends in Cognitive Sciences, 11(5), 197–203. | 10.1016/j.tics.2007.02.003 | Core statement that depicted bodily action and expression can recruit embodied resonance. | Vela uses pose/light/rendering metadata as weak computational proxies for embodiment rather than measuring it. | Implied-motion vs. static-pose as a binary decomposition field; target-specific embodiment tag. | No (theoretical review). |
| Body-selective visual coding — cortical area specific to bodies | primary_dimension; figurative filter | Downing, P. E., Jiang, Y., Shuman, M., & Kanwisher, N. (2001). A cortical area selective for visual processing of the human body. Science, 293(5539), 2470–2473. | 10.1126/science.1063414 | Demonstrates specialized cortical processing (extrastriate body area) for human bodies as a distinct visual class. | Supports Vela's assumption that figurative-bodily stimuli deserve dedicated measurement features. | None directly; grounds the claim that figurative is a privileged research domain. | No (fMRI). |
| Portrait gaze and social address | gaze-related priors; narrative dimension | Kesner, L., Grygarová, D., Honsnejmanová, I., & Fus, P. (2018). Perception of direct vs. averted gaze in portrait paintings: An fMRI and eye-tracking study. Brain and Cognition, 125, 88–99. | 10.1016/j.bandc.2018.06.004 | Direct gaze in painted portraits changes viewing behavior and social-cognitive response. | Vela can encode gaze as metadata and test its effect behaviorally at scale. | Explicit gaze_direction ∈ {direct, averted, downcast, occluded} as a first-class decomposition field. | Maybe (fMRI + eye-tracking). |
| Portrait aesthetics — artwork appeal vs. depicted-person appeal | profile alignment for portraits | Leder, H., Ring, A., & Dressler-Stross, S. (2013). See me, feel me! Aesthetic evaluations of art portraits. Psychology of Aesthetics, Creativity, and the Arts, 7(4), 358–369. | 10.1037/a0033311 | Portrait liking can split perceived person-likability from artwork evaluation. | Vela currently collapses these into one utility channel. | Two-channel response: "I am drawn to the work" vs. "I am drawn to the person depicted." | Yes (controlled ratings). |
| Theory of mind increases aesthetic appreciation | narrative dimension; implied-intention decomposition | Iosifyan, M. (2021). Theory of mind increases aesthetic appreciation in visual arts. Art & Perception, 9(2), 113–133. | 10.1163/22134913-bja10011 | Better grasp of intentions and mental states increases appreciation of visual art. | Vela's narrative dimension likely blends ToM legibility, story density, and symbolic readability. | Short ToM prime prompt before session (reading the figure's intention) as a manipulation arm. | Yes (four experiments, effect sizes reported). |
C. Adaptive recommendation as experimental instrument (RESEARCH-PROGRAM §1.3)
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Item Response Theory — adaptive measurement | eligibility gate; Confidence (§4.3); ResponseCount | van der Linden, W. J. (Ed.). (2018). Handbook of item response theory (Vols. 1–3). Chapman and Hall/CRC. | Publisher page | Standard reference for treating trait estimates as improving with informative items and sufficient data. | Vela is heuristic rather than full latent-trait estimation, but its sample and confidence logic is psychometric in spirit. | Fisher information as a second-generation confidence denominator; 2PL/graded-response scoring. | No (handbook). |
| Experience beyond accuracy | CandidateScore composite; pool/profile/curator/population blend | Knijnenburg, B. P., Willemsen, M. C., Gantner, Z., Soncu, H., & Newell, C. (2012). Explaining the user experience of recommender systems. User Modeling and User-Adapted Interaction, 22(4–5), 441–504. | 10.1007/s11257-011-9118-4 | Recommender evaluation must include felt experience, perceived quality, and satisfaction — not accuracy alone. | Vela's runtime score is explicitly a curation-quality function, not a pure prediction score. | The Knijnenburg ResQue framework as a periodic user-survey for Vela sessions. | Yes (meta of UX metrics). |
| Intent, context, diversity, beyond-accuracy | CandidateScore; exploration budget; session momentum | Schedl, M., Zamani, H., Chen, C.-W., Deldjoo, Y., & Elahi, M. (2018). Current challenges and visions in music recommender systems research. International Journal of Multimedia Information Retrieval, 7(2), 95–116. | 10.1007/s13735-018-0154-2 | Synthesizes why recommendation should model intent, diversity, context, and curation quality. | Vela ports that agenda from music to figurative-image recommendation. | Context-dependent bandit formalism; intent classification as a session-level latent variable. | No (review). |
| Implicit feedback with unequal evidential strength | per-response utility weight $w(r)$ (§2.1) | Hu, Y., Koren, Y., & Volinsky, C. (2008). Collaborative filtering for implicit feedback datasets. In Proc. 8th IEEE Int'l Conf. on Data Mining (pp. 263–272). | 10.1109/ICDM.2008.22 | Foundational paper treating behavior as preference evidence with varying confidence. | Vela hand-weights saves, low ratings, and boundaries in a psychometrically legible way. | Formal confidence weighting of each implicit signal against an explicit preference anchor. | Yes (quantitative RS). |
| Temporal drift in taste | decay factor $T_{1/2} = 60$ d (§4.2) | Koren, Y. (2009). Collaborative filtering with temporal dynamics. In Proc. 15th ACM SIGKDD (pp. 447–456). | 10.1145/1557019.1557072 | Preference models improve when time-varying taste is modeled explicitly. | Vela uses a transparent half-life rather than latent temporal factorization. | Per-dimension decay constants instead of one global $T_{1/2}$ — fashion drifts faster than pose preference. | Yes (RS quant). |
| Diversification vs. accuracy | RedundancyPenalty (§2.3, §5.4) | Ziegler, C.-N., McNee, S. M., Konstan, J. A., & Lausen, G. (2005). Improving recommendation lists through topic diversification. In Proc. 14th Int'l Conf. on World Wide Web (pp. 22–32). | 10.1145/1060745.1060754 | Classic argument that diversified lists can improve user satisfaction despite accuracy tradeoffs. | Vela penalizes similarity via relation graphs + CLIP near-duplicates. | Topic-diversification α as an exposed dial; intra-list similarity metric for session scoring. | Yes (RS quant). |
| Exploration/exploitation | exploration budget; pool bonus schedule (§5.1) | Li, L., Chu, W., Langford, J., & Schapire, R. E. (2010). A contextual-bandit approach to personalized news article recommendation. In Proc. 19th Int'l Conf. on World Wide Web (pp. 661–670). | 10.1145/1772690.1772758 | Formalizes why systems must explore even while serving personalized content. | Vela hard-codes exploration quotas rather than learning an online policy. | LinUCB or Thompson sampling as a second-stage exploration policy over the hand-coded budget. | Yes (RS quant). |
| Short-term session intent | session momentum (§5.3) | Jannach, D., Ludewig, M., & Lerche, L. (2017). Session-based item recommendation in e-commerce: On short-term intents, reminders, trends and discounts. User Modeling and User-Adapted Interaction, 27(3–5), 351–392. | 10.1007/s11257-017-9194-1 | Ongoing-session behavior reveals immediate goals that long-term profiles miss. | Vela uses a bounded additive momentum term rather than a full next-item session model. | GRU4Rec-style short-term embedding as a candidate augmentation of the current additive momentum. | Yes (RS quant). |
| Content-based cold start | cold Utility prior (§2.1 cold path); cold SignalDiversity prior; decomposition-seeded score (ASN-567) | Lops, P., de Gemmis, M., & Semeraro, G. (2011). Content-based recommender systems: State of the art and trends. In Recommender systems handbook (pp. 73–105). Springer. | 10.1007/978-0-387-85820-3_3 | Canonical source on using item features when interaction data are sparse. | Vela's decomposition priors are a hand-crafted, art-specific variant of content-based cold start. | TF-IDF or item-embedding baseline against which decomposition-seeded priors should be benchmarked. | Maybe (review with pointers to quantitative sub-studies). |
| Multimodal embedding similarity | runtime CLIP redundancy; relation graph | Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., … Sutskever, I. (2021). Learning transferable visual models from natural language supervision. In Proc. 38th Int'l Conf. on Machine Learning (pp. 8748–8763). PMLR. | Proceedings | Establishes a robust multimodal similarity space. | Vela uses CLIP narrowly for redundancy and dedup, not as the whole preference model. | Caption-conditioned CLIP retrieval to find absent aesthetic categories in the corpus. | No (ML methods). |
| Matrix factorization baseline | absence of — CandidateScore is interpretable by design | Koren, Y., Bell, R., & Volinsky, C. (2009). Matrix factorization techniques for recommender systems. Computer, 42(8), 30–37. | 10.1109/MC.2009.263 | Establishes latent-factor decomposition as the RS standard. Contextualizes where Vela's explicit 8-dimension model fits relative to implicit latent methods. | Vela's dimensions are curator-labeled and interpretable; MF would recover latent dimensions without semantic labels. Vela trades flexibility for interpretability by design. | MF as a baseline to benchmark the interpretable 8-dim model against (RQ5 or RQ6 supplement). | Yes. |
D. The 8 desire dimensions
Each row gives at least one supporting reference from aesthetic or perception research. Where Vela's dimension compresses multiple literature constructs, that is noted.
| Dimension | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Softness — preference for curved, smooth, rounded over angular | experience_units.primary_dimension = 'softness'; rendering (painterly, impressionistic); light_quality (soft, diffuse) | Bar, M., & Neta, M. (2006). Humans prefer curved visual objects. Psychological Science, 17(8), 645–648. | 10.1111/j.1467-9280.2006.01759.x | Robust preference for curved over angular visual forms, linked to threat appraisal. | Vela broadens curvature into a construct that also absorbs light, pose, and rendering. | Curvature index as an independent decomposition field distinct from softness dimension. | Yes (classical effect). |
| Intensity — high-arousal, high-contrast aesthetic response | primary_dimension = 'intensity'; light_quality (hard, chiaroscuro); contrast | Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39(6), 1161–1178. | 10.1037/h0077714 | Locates intensity on the arousal axis of the affective circumplex; provides the framework within which Vela's intensity dimension sits (high arousal, variable valence). | Russell's circumplex is a general affect model; Vela narrows intensity to the aesthetic domain and operationalizes via decomposition features rather than self-report. | Post-response valence + arousal self-report as circumplex coordinates to validate intensity from the outside. | Yes (foundational). |
| Narrative — image as story, implied before/after, readable sequence | primary_dimension = 'narrative'; pose_type (action, transitional); gaze_direction | Leder, H., Belke, B., Oeberst, A., & Augustin, D. (2004). A model of aesthetic appreciation and aesthetic judgments. British Journal of Psychology, 95(4), 489–508. | 10.1348/0007126042369811 | The Leder model's "cognitive mastering" stage captures narrative interpretation as a processing phase in aesthetic appreciation. | Vela operationalizes narrative through pose and gaze features rather than verbal self-report. | Self-reported "Is there a story here?" 0–1 field next to rating; direct test of the model's cognitive stage. | Yes (model citations). |
| Structure — geometric order, compositional balance, architecture | primary_dimension = 'structure'; composition (rule of thirds, centered, diagonal) | Jacobsen, T., & Höfel, L. (2002). Aesthetic judgments of novel graphic patterns: Analyses of individual judgments. Perceptual and Motor Skills, 95(3 Pt 1), 755–766. | 10.2466/pms.2002.95.3.755 | Symmetry and complexity strongly shape beauty judgments in formal patterns. | Vela extends structure beyond symmetry to compositional scaffolding more generally (also Arnheim 1974, infra). | Symmetry index (vertical, radial) as a decomposition subfield, not merged into composition. | Yes (quantitative). |
| Structure — compositional tension, balance, visual order (theoretical foundation) | same | Arnheim, R. (1974). Art and visual perception: A psychology of the creative eye (new version). University of California Press. | WorldCat | Foundational analysis of compositional tension, balance, and order in visual art — the perceptual vocabulary for what Vela's structure dimension captures. | Arnheim is theoretical; Vela provides empirical validation through behavioral response data. | Arnheim's compositional vocabulary (weight, vector, center) as additional decomposition tags. | No (book). |
| Texture — surface quality, material specificity, tactile-visual | primary_dimension = 'texture'; medium; rendering (impasto, graphic, photographic) | Lederman, S. J., & Abbott, S. G. (1981). Texture perception: Studies of intersensory organization using a discrepancy paradigm, and visual versus tactual psychophysics. Journal of Experimental Psychology: Human Perception and Performance, 7(4), 902–915. | 10.1037/0096-1523.7.4.902 | Texture is a robust perceptual dimension with cross-modal organization. | Vela turns this perceptual primitive into a desire dimension without splitting visual from haptic texture. | Within-texture taxonomy: smooth, rough, granular, fibrous — split a blunt field into axes. | Yes (classical psychophysics). |
| Texture — recovery of material properties from image | same | Fleming, R. W. (2014). Visual perception of materials and their properties. Vision Research, 94, 62–75. | 10.1016/j.visres.2014.06.004 | Reviews how the visual system recovers material properties (gloss, roughness, translucency) from image cues. | Vela measures desire response to texture rather than perceptual accuracy. | Gloss and translucency estimators from the vision literature as candidate Vela decomposition fields. | No (review). |
| Abstraction — figurative-to-abstract continuum; representational specificity | primary_dimension = 'abstraction'; rendering (abstract, stylized, representational) | Pihko, E., Virtanen, A., Saarinen, V.-M., Pannasch, S., Hirvenkari, L., Tossavainen, T., Haapala, A., & Hari, R. (2011). Experiencing art: The influence of expertise and painting abstraction level. Frontiers in Human Neuroscience, 5, 94. | 10.3389/fnhum.2011.00094 | Abstraction level changes gaze, affect, and evaluation, moderated by expertise. | Vela compresses a graded representational continuum into one dimension and one coarse axis. | Gaze-pattern time-series per abstraction level as a signal for the narrative vs. abstraction disambiguation. | Yes (N=40, quantitative). |
| Abstraction — neural response to painting category | same | Kawabata, H., & Zeki, S. (2004). Neural correlates of beauty. Journal of Neurophysiology, 91(4), 1699–1705. | 10.1152/jn.01225.2003 | Compared brain responses across abstract, landscape, portrait, and still-life. Provides empirical support for abstraction as a distinct class. | Vela's abstraction dimension is measured continuously via curator tagging and tested behaviorally. | Category-conditioned baselines (abstract vs. representational) as standard RQ2 covariates. | Yes (fMRI). |
| Classical — historical, canonical, art-historically legible registers | primary_dimension = 'classical'; style_tier; medium (oil, fresco, marble) | Di Dio, C., Macaluso, E., & Rizzolatti, G. (2007). The golden beauty: Brain response to classical and renaissance sculptures. PLoS ONE, 2(11), e1201. | 10.1371/journal.pone.0001201 | Classical proportions in human-form sculpture alter aesthetic and neural response. | Vela broadens classical from sculptural canon to a wider style prior. | Golden-ratio / canonical-proportion score as a decomposition subfield for figurative works. | Yes (small-N fMRI). |
| Contemporary — novel, unfamiliar, insight-dependent aesthetic registers | primary_dimension = 'contemporary'; rendering (photographic, graphic, digital) | Muth, C., & Carbon, C.-C. (2013). The aesthetic aha: On the pleasure of having insights into Gestalt. Acta Psychologica, 144(1), 25–30. | 10.1016/j.actpsy.2013.02.004 | Documents the "aesthetic aha" — insight-driven pleasure specific to initially ambiguous or unfamiliar works. Dominant response mode for contemporary art. | Vela measures desire for contemporary work across sessions, not in single-exposure insight paradigms. Dwell time is its proxy for the resolution of aesthetic ambiguity. | Per-exposure dwell trajectory as an aha-detection metric; second-exposure effect as contemporary-specific signal. | Yes (quantitative). |
| Contemporary — expertise-dependent conceptual spaces | same | Augustin, M. D., & Leder, H. (2006). Art expertise: A study of concepts and conceptual spaces. Psychology Science, 48(2), 135–156. | PsycNET | Contemporary-art perception depends on conceptual spaces and style knowledge. | Vela operationalizes contemporary as a taste dimension for non-experts as well. | VAIAK art-knowledge score as a covariate for contemporary dimension regressions. | Yes (quantitative, N≈100). |
| Realistic vs. abstract preference across cultures (dimension meta-control) | Aesthetic axis (realistic ↔ abstract) | Darda, K. M., & Cross, E. S. (2022). The role of expertise and culture in visual art appreciation. Scientific Reports, 12, 10666. | 10.1038/s41598-022-14128-7 | Representational preference and expertise effects remain measurable across cultural settings. | Vela's single aesthetic axis is narrower than culture-sensitive art appreciation. | Culture-of-origin prior as a per-user covariate; ingroup bias as a testable confound for RQ4. | Yes (two preregistered experiments, effect sizes). |
| Boundary modulation — art context softens negative content | boundary_flag; boundary rate; BoundaryTags | Gerger, G., Leder, H., & Kremer, A. (2014). Context effects on emotional and aesthetic evaluations of artworks and IAPS pictures. Acta Psychologica, 151, 174–183. | 10.1016/j.actpsy.2014.06.008 | Negative stimuli can be liked more when framed as art, even if negative emotion remains. | Vela is more conservative: strong boundary evidence heavily suppresses future exposure. | "Art context" vs. "unframed" as a between-participants condition in instrument-validation studies. | Yes (quantitative, N≈30). |
E. Batch scorer components (01-math-spec §2)
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Weighted aggregation of explicit + implicit feedback | $\mathbf{Utility}(R_u)$ (§2.1); weight map $w(r)$ on rating, saved, boundary_flag | Hu, Koren, & Volinsky (2008). See Section C. | 10.1109/ICDM.2008.22 | Establishes case for differential weighting of implicit vs. explicit signals. | Vela adds a negative boundary_flag signal ($w = -1.0$) with no standard CF analogue; operates on a curated corpus, not transactional data. | Confidence-weighted CF as a baseline scorer to benchmark against current $w(r)$. | Yes. |
| Rating variance as information-bearing signal | $\mathbf{SignalDiversity}(R_u)$ (§2.2) = $\operatorname{Var}(\rho)/V_\text{max}$ | Silvia (2005) appraisal theory implies heterogeneous response = informative. No primary source for variance-as-first-class-signal in this exact form. | — | Theoretical anchor exists, but the specific operationalization is novel to Vela. | Novel engineering move — flagged as overclaim candidate (Section K, RESEARCH-PROGRAM §I.1 qualification needed). | Construct validity study: does SignalDiversity predict divergence in expert judgments? | No (Vela-specific). |
| Semantic similarity for redundancy | $\mathbf{RedundancyPenalty}$ (§2.3); CLIP σ_th = 0.85, σ_hard = 0.92 | Radford et al. (2021). See Section C. | Proceedings | CLIP provides the cosine similarity space; thresholds are design choices. | Vela applies CLIP in aesthetic sequencing, not retrieval; penalty targets experienced redundancy, not corpus dedup. | Tuning the thresholds against a held-out human-rated redundancy set. | Maybe. |
| Composite desire score — utility × diversity − redundancy | $\mathbf{DesireScore}(u) = \text{clamp}(\mathbf{Utility} \cdot \mathbf{SignalDiversity} - \mathbf{RedundancyPenalty})$ (§2.4) | No strong primary source. | — | Multiplicative form is Vela's primary scoring innovation. | Novel. Penalizes both universally dull content (low utility) and universally loved content (low diversity/low information). | Benchmark against additive, weighted-sum, and Bayesian alternatives in an ablation study. | No (Vela-specific; demands construct validation before formal meta). |
F. Pool dynamics (01-math-spec §3)
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Adaptive item routing under accumulated evidence | Pool ladder $P$ [INFINITY → D → C → B → A → PURGATORY]; thresholds $\tau_\text{promo} = 0.7$, $\tau_\text{demo} = 0.3$ (§3.1) | van der Linden (2018). See Section C. | — | CAT routing selects items based on accumulated evidence; Vela's pool ladder is the analogue. | Vela's items are routed by collective cross-user desire, not by individual-user ability; the pool is a collective reputation system. | Item-information-function-style routing as an alternative to fixed thresholds. | No (Vela-specific design). |
| Specific threshold values (0.7 / 0.3) | promotion/demotion cutoffs | No strong primary source. | — | Hysteresis is reasonable engineering; the exact constants are not literature-derived. | Overclaim candidate — flagged in Section K. | Threshold sensitivity analysis against pool-movement rate and user-reported quality. | No. |
G. User desire profiles (01-math-spec §4)
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Person-parameter estimation from response patterns | user_desire_profiles.dimension_scores $\theta_v[d]$ (§4.1) | van der Linden (2018). See Section C. | — | IRT person-parameter logic parallels profile estimation from sparse responses. | Vela estimates 8 simultaneous dimension scores with simple weighted aggregation, not MLE. | Marginal MLE or Bayesian estimator as a second-generation profile computation. | No. |
| Temporal decay of preferences | $\exp(-\Delta t / T_{1/2})$ with $T_{1/2} = 60$ d (§4.2) | Koren (2009). See Section C. | 10.1145/1557019.1557072 | Preference models improve when temporal dynamics are modeled. | Vela uses transparent exponential decay rather than latent temporal factorization. | Per-dimension decay constants (fashion fast, pose slow); change-point detection. | Yes. |
| Sample-size-to-confidence mapping | confidence = $\min( | R_{v,d}^* | /20, 1)$ (§4.3); CONFIDENCE_DENOMINATOR = 20 | Maas, C. J. M., & Hox, J. J. (2005). Sufficient sample sizes for multilevel modeling. Methodology, 1(3), 86–92. | 10.1027/1614-2241.1.3.86 | Provides empirical guidance on minimum sample sizes; Vela's denominator is consistent in magnitude with Maas & Hox's guidance. | $N = 20$ is an engineering approximation of a credible-interval concept; not motivated by a specific psychometric paper. |
H. Candidate scoring / runtime engine (01-math-spec §5)
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Pool-level quality bonuses | POOL_BONUS $B_A = 0.5$, $B_B = 0.3$, ..., $B_\text{PURG} = -999$ (§5.1) | Schedl et al. (2018). See Section C. | 10.1007/s13735-018-0154-2 | Pool bonuses implement a structured exploration–exploitation tradeoff. | Hard PURGATORY exclusion via $B = -999$ is a curation-integrity mechanism with no standard RS parallel. | Explicit ablation study: with/without pool bonuses; with/without PURGATORY exclusion. | No (Vela-specific). |
| Profile alignment — content features → user trait profile | $\mathbf{ProfileAlignment} = \alpha_\text{score} \cdot \theta_v[d]\text{score} \cdot \theta_v[d]\text{confidence}$ (§5.2) | Knijnenburg et al. (2012). See Section C. | 10.1007/s11257-011-9118-4 | User–item alignment via latent preference matching is the core of CF and CBF. | Vela's alignment is interpretable by design (8 named dimensions), supporting construct validity. | MF baseline vs. 8-dim interpretable model head-to-head — framing for an RS paper. | Yes. |
| Session momentum — recency and sequential priming | $\mu^{\pm}$ capped at $\pm 0.15$ (§5.3) | Murdock, B. B. (1962). The serial position effect of free recall. Journal of Experimental Psychology, 64(5), 482–488. | 10.1037/h0045106 | Documents serial position effects in memory and sequential processing; grounds within-session momentum. | Vela's momentum is aesthetic (desire intensifying or fatiguing), not mnemonic; bidirectional and capped, which serial-position literature does not model. | Position-in-session as a first-class covariate in RQ3 models; U-shape prior on session desire. | Yes (classical). |
| Cosine similarity deduplication | σ_hard = 0.92; σ_th = 0.85 (§5.4) | Radford et al. (2021). See Section C. | Proceedings | CLIP cosine similarity defines Vela's runtime redundancy metric. | Applied within a single session's candidate set, not across corpus. | Human-rated redundancy calibration set to validate thresholds. | Maybe. |
I. Statistical methods supporting RQs
| Coordinate | Vela measure | Primary source (APA) | DOI / URL | Relevance | Vela divergence | Absorption candidate | Meta-analyzable? |
|---|---|---|---|---|---|---|---|
| Latent profile analysis — subgroups in multivariate response space | RQ1: LPA on {rating, saved, dwell_ms, boundary_flag, emotions, intensity} | Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14(4), 535–569. | 10.1080/10705510701575396 | Gold-standard guidance on class enumeration via BIC, LMR-LRT, and entropy. Required for RQ1. | Vela's feature space is behavioral rather than self-report; richer but noisier than typical LPA inputs. | Direct adoption — the BIC/LMR-LRT/entropy triad for RQ1 class-count selection. | Yes (methodological). |
| Multilevel regression for nested data | RQ2: mixed-effects regression predicting response from decomposition features | Maas & Hox (2005). See Section G. | 10.1027/1614-2241.1.3.86 | Establishes 50 level-2 units as minimum for stable random-effects estimation — cited in RESEARCH-PROGRAM §IX as the Phase 1 constraint. | Vela's crossed random effects (responses in units in users) are more complex than Maas & Hox's 2-level simulation; guidance is conservative. | Direct adoption — minimum 50 users per arm as the power floor for RQ2 and RQ4. | Yes (methodological). |
| Intraclass correlation — profile stability | RQ3: ICC on dimension_scores across profile_version snapshots (requires versioned snapshots — gap flagged in 02-instrument-validation §4) | Shrout, P. E., & Fleiss, J. L. (1979). Intraclass correlations: Uses in assessing rater reliability. Psychological Bulletin, 86(2), 420–428. | 10.1037/0033-2909.86.2.420 | Defines ICC(2,1) and ICC(2,k) for consistency across occasions. Method for RQ3 between-session profile stability. | Vela's "raters" are session-level profile estimates; infrastructure gap is an immediate engineering ticket. | Direct adoption; also: implement user_desire_profile_versions before RQ3 data collection. | Yes (methodological). |
J. Validated instruments for cross-walk
Closest existing validated instruments we might compare Vela's profile against. Each row: construct measured, validation N, public availability, primary citation, feasibility of cross-walking with Vela data. Note: PLOAS (mentioned in Mike's scoping prompt) was not locatable as a real instrument under that acronym in the psychometric literature; it is not carried forward. BFAS-R was also not confirmed — the closest published instruments are BFAS (DeYoung, Quilty, & Peterson 2007) and its short form BFAS-40 (2022), neither of which is a dedicated aesthetic instrument. Dropped from the cross-walk; Openness-Aesthetics subfacet would be a weak substitute only.
| Instrument | Construct measured | Primary citation | DOI / URL | Validation N | Public availability | Vela cross-walk feasibility | Recommended use |
|---|---|---|---|---|---|---|---|
| AReA — Aesthetic Responsiveness Assessment | 3 first-order factors: aesthetic appreciation, intense aesthetic experience, creative behavior; 2nd-order general aesthetic responsiveness | Schlotz, W., Wallot, S., Omigie, D., Masucci, M. D., Hoelzmann, S. C., & Vessel, E. A. (2021). The Aesthetic Responsiveness Assessment (AReA): A screening tool to assess individual differences in responsiveness to art in English and German. Psychology of Aesthetics, Creativity, and the Arts, 15(4), 682–696. | 10.1037/aca0000348 | 781 (US + Germany, measurement-invariant) | Scale items published in supplementary materials; freely reproducible | High. The "intense aesthetic experience" factor maps directly onto Vela's joint high-desire signal (rating=5 ∧ saved=true ∧ long dwell). AReA run as a pre-study questionnaire would validate whether Vela's in-session behavior converges with trait-level aesthetic responsiveness. | Collect AReA at study onboarding for RQ1 + RQ4; correlate AReA subscales with user_desire_profiles dimension scores. |
| VAIAK — Vienna Art Interest and Art Knowledge Questionnaire | Two separable scales: art interest (11 items, α = .94); art knowledge (26 items, α = .89) | Specker, E., Forster, M., Brinkmann, H., Boddy, J., Pelowski, M., Rosenberg, R., & Leder, H. (2020). The Vienna Art Interest and Art Knowledge Questionnaire (VAIAK): A unified and validated measure of art interest and art knowledge. Psychology of Aesthetics, Creativity, and the Arts, 14(2), 172–185. | 10.1037/aca0000205 | 600 (lay + art-history students + professionals) | Items on OSF and in supplementary materials | High. Expertise is orthogonal to desire; VAIAK gives a principled control variable for RQ4. Knowledge subscale is especially useful for interpreting Vela's classical and contemporary dimensions. | Optional onboarding questionnaire; include as covariate in RQ2 and RQ4 regressions. |
| Aesthetic Fluency Scale (revised) | Knowledge-based assessment of expertise in the arts (10 art-history items, self-rated) | Smith, J. K., & Smith, L. F. (2006). The nature and growth of aesthetic fluency. In P. Locher, C. Martindale, & L. Dorfman (Eds.), New directions in aesthetics, creativity, and the arts (pp. 47–58). Baywood Publishing. Revised: Rodriguez-Boerwinkle, R. M., Silvia, P. J., Cotter, K. N., Christensen, A. P., & Wanzer, D. L. (2023). Updating the Aesthetic Fluency Scale. PLOS ONE, 18(2), e0281547. | Revised: 10.1371/journal.pone.0281547 | Revised: ~1,500 across studies | Revised items open access in PLOS ONE supplementary | Medium. Overlaps with VAIAK's knowledge scale but is lighter and more accessible. Either one is sufficient — pick based on session-length budget. | Alternative to VAIAK knowledge subscale if participant fatigue is a concern. |
| AEQ — Aesthetic Experience Questionnaire | 6 dimensions: emotional, cultural, perceptual, understanding, flow conditions, flow experience | Wanzer, D. L., Finley, K. P., Zarian, S., & Cortez, N. (2020). Experiencing flow while viewing art: Development of the Aesthetic Experience Questionnaire. Psychology of Aesthetics, Creativity, and the Arts, 14(1), 113–124. | 10.1037/aca0000203 | 3 independent studies, N > 800 (Polish replication, Gładyszewska-Cylulko et al. 2023) | Items in supplementary materials | High for state-level measurement. The 6-dimension schema is closest analog to Vela's dimensional profile; flow dimensions offer a direct external validator for Vela's session momentum. | Post-session debrief instrument for RQ3 (session-level aesthetic experience) and RQ5 (adaptive vs. editorial engagement). |
| APPS — Aesthetic Processing Preference Scale | Willingness to engage in effortful/controlled cognitive processing of art | Christensen, A. P., Cotter, K. N., Silvia, P. J., & Rodriguez-Boerwinkle, R. M. (2022). Development and validation of the Aesthetic Processing Preference Scale. Psychology of Aesthetics, Creativity, and the Arts [advance online]. | 10.1037/aca0000449 | Multi-sample, N ≈ 1,000 | Items in supplementary materials | Medium. Would discriminate Vela's "learning" vs. "calibrated" mode users independently of session count. | Onboarding or end-of-session; useful stratifier for RQ5 (does adaptive benefit high-APPS or low-APPS users more?). |
| AESTHEMOS — Aesthetic Emotions Scale (added 2026-04-24 during ASN-594 merge) | 9 families of aesthetic emotion (pleasure, awe, being-moved, interest, vitality, fascination, tenderness, nostalgia, sublime) across visual, musical, and text-aesthetic stimuli — 21 subscales, 2 items each | Schindler, I., Hosoya, G., Menninghaus, W., Beermann, U., Wagner, V., Eid, M., & Scherer, K. R. (2017). Measuring aesthetic emotions: A review of the literature and a new assessment tool. PLOS ONE, 12(6), e0178899. | 10.1371/journal.pone.0178899 | 4 studies, N ≈ 200–500 per study, online + lab, English + German | Items in supplementary materials | Very high. The only validated cross-modal aesthetic-emotion scale — visual, music, text. Directly maps onto the Menninghaus 2019 taxonomy. Should be the primary post-session aesthetic-emotion instrument for RQ3 and for any future Vela text-side platform. Missed in ASN-575 §J; added during ASN-594 merge. | Post-session instrument for all modalities; primary dependent variable in any cross-modal Vela study. |
Recommended default bundle for Phase 1 instrument-validation study: AReA (trait responsiveness) + VAIAK or Aesthetic Fluency (expertise covariate) + AEQ (state-level post-session) — roughly 50 items total, ~10 minutes. Adds construct-validity evidence to every RQ at minimal participant cost.
K. Reincarnation — novelty, superiority, gaps
Structured comparison feeding the "Reincarnation vs. the field" paper (future assignment). Each row: the specific mechanism, what existing systems do, what Vela does, whether the difference is genuinely novel, whether it's superior, what problem it solves, and what the literature does better (if anything).
K.1 Genuinely novel (no close prior art identified)
- DesireScore multiplicative form ($\mathbf{Utility} \cdot \mathbf{SignalDiversity} - \mathbf{RedundancyPenalty}$). No recommender or empirical-aesthetics paper we located combines user-response magnitude with inter-user response variance as a first-class scoring signal. The intuition — "a unit must be both liked and opinion-splitting to rank highly" — treats heterogeneity as evidence that the stimulus is doing aesthetic work, not as noise to be averaged out. Construct-validity study is a required next step before claiming superiority; see
docs/engine-room/02-instrument-validation.md§4. - 45-field visual decomposition schema applied to a curated figurative corpus. No prior aesthetic-behavior study has structured pose, gaze, light, composition, and mood metadata at this granularity across ~400 curated works. The dataset itself is a contribution that enables RQ2.
- Collective pool ladder with PURGATORY exclusion as a curation-integrity mechanism orthogonal to user-level personalization. Standard RS literature treats item retirement as a business-rule concern; Vela treats it as a batch-state transition with audit history.
- Boundary evidence as a first-class negative signal with permanent tag production. Standard implicit-feedback CF assumes missing-or-positive; Vela explicitly models disliking with durable user_boundary_rules. Silvia & Brown (2007) established that negative aesthetic emotions are real, but no RS system we found operationalizes them this directly.
K.2 Derivative but competently executed
- Temporal decay ($T_{1/2} = 60$ d) — a transparent exponential is simpler than Koren (2009) latent temporal factorization, sacrificing fit quality for interpretability. Superior for Vela's auditable-engine goal; inferior for pure prediction accuracy.
- Exploration budget (25% default) — simpler than LinUCB / Thompson sampling (Li et al. 2010). Superior for transparency; likely inferior for long-horizon reward.
- IRT-flavored confidence metric ($\min(N/20, 1)$) — engineering approximation of psychometric confidence; lacks Fisher information or credible-interval grounding. Fine for Phase 1; should upgrade by RQ3.
- CLIP-based redundancy — direct adaptation of Radford et al. (2021) with Vela-specific thresholds. No novelty in the method; the novelty is the aesthetic-sequencing application.
K.3 Problems Reincarnation uniquely solves
- Adaptation without dark-pattern optimization. Standard RS systems optimize for engagement-as-revenue, which produces doom-scrolling dynamics. Vela's utility weights explicitly include a negative boundary term and cap exploration — the system is designed to stop pulling when the user has seen enough, not to extend session length indefinitely. No published RS system we located has this as an explicit design axis.
- Instrumented measurement with auditable state. The pool ladder, profile snapshots, and response log together form a reproducible instrument state. Standard RS systems keep latent parameters that cannot be externalized; Vela's state is fully inspectable. Superior for scientific use; no direct competitor.
- Curator + algorithmic co-authorship.
curator_priority, chosen_one, and PURGATORY are human-in-the-loop controls that override the algorithmic score. No RS paper treats curator veto as first-class; it is usually a business override bolted on. Vela's integration is the contribution.
K.4 Where the literature does it better (gaps to close)
- No construct validity evidence yet. Vela claims 8 desire dimensions but has no external validation against AReA / VAIAK / AEQ. Until a cross-walk study ships (Section J), the dimensions are engineering-chosen, not empirically supported.
- No Item Response Theory / Fisher information. Standard CAT systems select items by expected information gain; Vela routes by scalar DesireScore. Literature (van der Linden 2018) has 40 years of technique Vela doesn't yet use.
- No formal causal identification. Vela's natural-experiment framing for RQ5 (editorial vs. calibrated mode within same user) has obvious confounds (mode eligibility is endogenous to experience). Standard causal-inference techniques (DoubleML, IV) would be a rigor upgrade.
- Effect-size reporting discipline. Research community standards (APA 7, Open Science) require effect-size + confidence interval for every inferential claim. Vela's pulse/reporting pipeline currently emits counts and means; add effect sizes (Cohen's d, η², partial-η²) before external publication.
K.5 Falsifiable novelty claims for the Reincarnation comparison paper
Each of these is a testable hypothesis that, if confirmed, constitutes novel contribution:
- Opinion-splitting = desire signal. On a held-out subset, units with higher
SignalDiversityreceive more saves (per rating) than units with high mean rating alone. Falsifiable via the current response corpus once N ≥ 200 per unit. - Interpretable beats latent. An 8-dim curator-labeled model predicts held-out response within 3–5% of a matrix-factorization baseline, while being auditable. Falsifiable via ablation.
- Auditable state enables construct validation. Running AReA / VAIAK / AEQ alongside Vela sessions yields partial correlations ≥ .3 between Vela's dimension_scores and theoretical constructs. Falsifiable via Section J study.
- Boundary signal = measurable health axis. Users for whom
user_boundary_rulesare active show fewer session terminations and longer return latency than matched controls without boundary evidence. Falsifiable via survival analysis.
L. Downstream-artifacts index
How this literature map feeds the writing pipeline Mike named (2026-04-24). Each downstream artifact is a future assignment — not part of ASN-575.
| Downstream artifact | Literature-map sections | Primary sources | Scope | Target venue or surface |
|---|---|---|---|---|
| Formal literature review — aesthetic desire, figurative body, adaptive instruments | A, B, C, D, J | Full bibliography | Narrative synthesis with Vela's 8-dim model as the integrating frame; positions Vela in the field | Psychology of Aesthetics, Creativity, and the Arts (target), Empirical Studies of the Arts (secondary) |
| Formal meta-analysis (feasibility: partial) | C, D, K.1 | Rows tagged "Yes" in Meta-analyzable column | Likely too heterogeneous for single meta. Feasible sub-meta: diversification-vs-accuracy in RS (Ziegler, Koren temporal, Jannach); figurative body-selective preferences (Bar, Di Dio, Pihko). | User Modeling and User-Adapted Interaction or Psychological Bulletin (sub-meta) |
| Reincarnation: novelty and instrument validation paper | K | All of the above + AReA / VAIAK / AEQ cross-walk data | Validates K.5 falsifiable claims; publishes the instrument design alongside first-pass evidence | ACM RecSys or Psychology of Aesthetics, Creativity, and the Arts |
| Absorption / engineering uplift — variables & designs to port | A–I Absorption column, K.2, K.4 | Rows where absorption candidate ≠ "None" | Reincarnation v2 engineering ticket queue: per-dimension decay, Fisher-info confidence, gaze-direction decomposition field, etc. | Internal (ASN ticket queue) |
| Research proposals across time horizons | A–K | Synthesized | Phase 0 (IRB + intake), Phase 1 (50–200 users, instrument validation), Phase 2 (cohort study, 500+), Phase 3 (longitudinal + cross-cultural) | Internal + grant submissions (NSF SBE, NEH, Templeton) |
| Public-ready accessible version — website | Synthesis of B, D, K.3 | Narrative pick of 10–15 top sources with named findings | /research or magazine companion article; link back to full lit map and peer-reviewed papers as they land | vela.study/research + magazine |
| Magazine pieces with stories/examples | docs/research/story-hooks.md (companion file) | Curated subset with named participants, anecdotes, surprising findings | Human-scale essays: "Why we prefer curved things" (Bar & Neta), "The aha you feel at contemporary art" (Muth & Carbon), "What the brain does with beauty" (Vessel, Di Dio) | Vela Magazine |
Open questions & limitations
- The eight Vela dimensions are productive composites, not one-to-one imports from orthodox empirical-aesthetics constructs.
- The strongest literature ancestry is for construct families (interest, embodiment, abstraction, diversification, exploration), not for exact numeric constants in the engine. Pool thresholds (0.7/0.3), confidence denominator (20), DesireScore multiplicative form, and pool-bonus schedule are engineering choices to validate empirically, not canonized truths.
- No construct validity evidence against AReA / VAIAK / AEQ yet — highest-priority next study.
- Meta-analysis across the domain is not feasible as a single paper (stimulus sets, outcomes, and samples are too heterogeneous). Sub-meta on diversification-vs-accuracy RS is feasible; sub-meta on body-selective preferences is feasible.
- Two citations (Arnheim 1974; van der Linden 2018 handbook) are books without per-volume DOIs — cited via WorldCat / publisher pages.
- PLOAS (mentioned in scoping prompt) and BFAS-R (ditto) could not be verified as real instruments; dropped. Aesthetic Fluency Scale and APPS substituted as verified alternatives.