Quantifying the Occult: A Comparative Study of Hindu and Buddhist Deities Using Machine Learning Methods
cs.CY · cs.LG
Quick verdict
Quantifies Hindu-Buddhist deity divergence with Gower+LLM embeddings on 196 deities; novel dataset, but inconsistent metrics and zero statistical validation.
Summary
This paper introduces a dual-matrix framework for quantifying similarity among 196 Hindu and Vajrayana Buddhist esoteric deities. Morphology is compared via a Gower distance over 28 iconographic features, weighted by a proposed 'Cardinality Weighting' (category cardinality times inverse frequency), intended to make rare implements (khadga, veena) dominate generic traits. Theology is compared via cosine distance over 768-dimensional sentence embeddings (all-mpnet-base-v2) of LLM-expanded (Gemini 3.1 Flash Lite) cosmic-domain descriptions, used as a 'synthetic proxy'. The author reports nearest-neighbor tables (e.g., Lakshmi/Vasudhara visually distant yet semantically adjacent), UMAP projections, and a headline cross-tradition pairing: Chinnamasta and Chinnamunda at Gower 0.288 and cosine 0.068, framed as 1:1 esoteric transfer. A so-called 'Atin Effect' — sequential anchoring on familiar symbols — is modeled computationally but not tested on human observers. Strengths include an open dataset, reproduced prompts and hyperparameters, and candid limitations. However, the paper contains internal contradictions (the visual metric is an unnormalized dot product in §3.4 but Gower in §4.1; tables are labeled 'Cognitive Cosine Distance'; feature counts 30 vs 28; UMAP n_neighbors 12 vs 15), offers no statistical testing or null models, provides no quantitative ablation of Cardinality Weighting, and mischaracterizes its tools (mpnet cannot yield OOV null vectors; UMAP does not guarantee global structure), so claims of 'algorithmic validation' and 'proof' overreach.
Review complete
Assessment at a glance
6
risk flags identified
● high 2
● medium 3
● low 1
4 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- A dual-matrix architecture (Cardinality-Weighted Gower distance + LLM-expanded sentence embeddings) can quantify morphological vs theological divergence across 196 Hindu and Vajrayana deities.
- The proposed Cardinality Weighting (W = C_k · N/f(t)) resolves a 'Gestalt Problem' in which generic visual traits suppress rare, identity-defining iconographic anchors.
- The paired matrices computationally demonstrate 'iconographic camouflage': Lakshmi and Vasudhara are visually distant but semantically adjacent, and high-cardinality orthodox anchors like the veena cluster Matangi with Saraswati (the 'Atin Effect').
- Chinnamasta and Chinnamunda share near-identical coordinates (D_G = 0.288, D_C = 0.068), which the author claims proves a 1:1 esoteric transfer across the Hindu-Buddhist boundary.
- The framework is the first algorithmic validation of cross-tradition continuities previously documented only qualitatively by Sanderson, Benard, and Kinsley.
Strengths
Genuinely novel open dataset
196 deities coded on a 28-feature mixed-type codebook (Table 16: 2 numerical, 17 nominal including 10 weapon slots, 9 boolean), sourced from canonical texts (Sadhanamala, Nispannayogavali, Brihat Tantrasara, Hevajra/Cakrasamvara Tantras) and released on Kaggle and GitHub.
Above-average procedural transparency
The full LLM system prompt with few-shot examples is reproduced (Appendix A), exact hyperparameters listed (B.2, incl. random_state=42), and hardware/environment specified (B.3-B.4, Tesla T4, Python 3.12) — rare for digital-humanities ML papers.
Conceptually sound form-vs-function design
Candid limitations section
Weaknesses
Internally contradictory pipeline description
No statistical or ablation evidence for the core novelty
Headline result partially baked into the prompt
Technically incorrect claims about the methods
Psychological claim without psychological data
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (arXiv 2609.27074, cs.CY/cs.LG) proposes a "dual-matrix architecture" comparing 196 Hindu and Vajrayana Buddhist esoteric deities. The Visual Matrix computes Gower distance over 28 iconographic features (Table 16: 2 numerical, 17 nominal, 9 boolean) weighted by a novel "Cardinality Weighting" W(t_k)=C_k·N/f(t_k) (Eqs. 1-2); the Semantic Matrix embeds LLM-expanded (Gemini 3.1 Flash Lite) cosmic-domain texts with all-mpnet-base-v2 into 768-d vectors and takes cosine distance (Eq. 7). Central claims: (1) Cardinality Weighting fixes a "Gestalt Problem" in which generic traits swamp rare implements; (2) the paired matrices "algorithmically validate" iconographic camouflage — form diverges while function persists (Lakshmi/Vasudhara, §5.1-5.2, Fig. 2); (3) the "Atin Effect" (§5.5), a sequential-anchoring bias borrowed from Sarkar's novel *Bhog*, is computationally modeled; (4) Chinnamasta/Chinnamunda at D_G=0.288 and D_C=0.068 (Table 15) constitute "1:1 esoteric transfer" and "the first algorithmic proof" of bridges documented by Sanderson and Benard.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 5 weaknesses, analyzed in detail
- 6 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the 'Atin Effect' in this paper?
A term the author coins after Atin, a character in Avik Sarkar's novel Bhog: the tendency to recognize an unfamiliar esoteric deity by its most familiar orthodox symbol (a veena suggesting Saraswati) while missing wrathful Tantric attributes like the sword and skull cup. The paper simulates this sequential anchoring computationally with distance matrices; it was not tested on human subjects.
How many deities are in the dataset and where can I access it?
196 Hindu and Vajrayana Buddhist deities, each coded on 28 iconographic features (held objects, mudras, posture, complexion, arm/head counts, wrathful markers). The paper links a public Kaggle dataset and a GitHub repository (SrabonGitikar/Tantra-ML) containing the code, distance matrices, and interactive plots.
What are the key quantitative results?
The headline pairing is Chinnamasta (Hindu) and Chinnamunda (Buddhist) at Gower distance 0.288 (visual form) and cosine distance 0.068 (semantic function), which the author interprets as a one-to-one esoteric transfer. Lakshmi and Vasudhara are reported as visually distant but semantically adjacent, presented as evidence of 'iconographic camouflage'.
Is the method statistically validated?
No. The paper reports point-estimate distances without statistical tests, confidence intervals, or permutation baselines, and its core 'Cardinality Weighting' heuristic is never quantitatively ablated against unweighted baselines. Several internal inconsistencies (metric definitions, feature counts, UMAP parameters) also remain unresolved, so independent replication currently requires guesswork.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1-3 (UMAP projection, egocentric orbits, Atin Effect illustration) cannot be assessed from the extracted text.
- The supplementary GitHub/Kaggle artifacts were not inspected; availability and contents are taken from the paper's own claims.
- Mathematical notation is partially garbled in extraction (Eqs. 1-9), so formula readings are best-effort.
- Some contradictions (28 vs 30 features, n_neighbors 12 vs 15) could stem from unreconciled draft revisions rather than the actual executed pipeline.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.