This page presents an academic secondary analysis of
Aella's Big Kink Survey (N=15,503).
It discusses kink categories, sexual behaviors, and adult themes
in scientific terms. Some terminology is explicit.
This notice won't show again this session.
01 · Open Collaboration — Sexual Psychology
We ran Aella's kink survey through thermodynamics.
Aella's Big Kink Survey — N=15,503 fully-completed responses from a live survey now exceeding 900,000 participants
(Aella, doi:10.5281/zenodo.18625249) —
scored 49 kink categories on three void dimensions.
The strongest result: a direct measurement of the framework's explaining-away penalty
on a sexual psychology dataset — I(D;M|Y) = 0.50 bits · z=+84σ
vs permutation null, N=14,102. Eighth substrate in the Čencov substrate-invariance chain
(strings → gravity → quantum inference → language → sexual psychology).
Six of seven correlational findings pass FDR. Consent gradient confirmed at respondent level
(ρ=+0.267, p=7×10⁻²⁴⁹, N=15,324). One thing is missing before this becomes a paper: three human raters.
⚡ Collaboration open — paid raters wanted · N=15,503 (cleaned subset of a live survey now exceeding 900,000 participants) · data by Aella (doi:10.5281/zenodo.18625249)
She designed the instrument, recruited 15,503 respondents, scored 49 kink categories
on shame, consent preference, therapeutic value, and inducibility — and then published
the full dataset openly on Zenodo. That is a research-grade empirical contribution
by any standard. This secondary analysis exists because of that.
"Sexual psychology gets treated like it's too sensitive for real statistical analysis.
I put the data out there because I think N=15,503 is big enough to do proper science on.
Shame, consent, therapeutic value — these aren't just interesting survey questions.
They're telling you something structural about how these things work.
If someone can extract that structure, I want to know what they find."
— Aella (paraphrase from public Substack and social commentary on the BKS dataset).
Framework relevance:
"Something structural" = O · R · α. She was looking for what we found.
Kink void space — O × R × α
49 kink categories in three-dimensional void space. Color = void intensity V. Sphere size = shame load. Glowing plane = V*=5.52 drift-selection threshold. Drag to orbit · scroll to zoom · click any sphere for detail — try Mindbreak first.
O — Opacity (x)R — Responsiveness (y)α — Coupling (z)
What the data shows
ρ=+0.43
Opacity predicts shame
FDR ✓
Kinks with a harder-to-explain mechanism of arousal carry more shame
per unit of engagement (p=0.002, N=49 categories, N=15,503 respondents).
Survives demographic controls for gender, political orientation, and sexual orientation.
Δ across controls < 0.01 — the finding is not confounded.
ρ=+0.40
Rarity predicts shame
FDR ✓
Kinks that fewer people rate as attractive carry more shame (p=0.004).
This is social minority status, not mechanism. The rarest kinks
(Mindbreak, Vore, CGL) carry shame because almost nobody shares them —
not because they are intrinsically shameful.
Evolutionary reading
Positive frequency-dependent selection. Minority strategy carriers face partner scarcity
and social exclusion — exactly the fitness cost selection generates. The shame is not
cultural stigma: it is the evolved signal of reduced mating opportunity.
ρ=+0.33
Composite void score predicts shame
FDR ✓
A single scalar combining all three rubric dimensions (O, R, α) into one thermodynamic
score — Pe = K·sinh(2·(BA − c·BG)) — predicts shame at p=0.020 across 49 kink categories.
This is a composite of the opacity and rarity effects above, not an independent finding,
but it confirms the three dimensions compress into a coherent signal rather than cancelling out.
p=0.003
The V* thermodynamic threshold splits kink regimes
FDR ✓
9 kinks sit above V*=5.52 (the drift-selection boundary): Mindbreak, CGL,
Age Regression, Full-Time Power Exchange, Mental Alteration, Master/Slave,
Obedience, Psychological Torture, Nonconsent Fantasy.
Supercritical kinks carry significantly more shame than subcritical kinks
(Mann-Whitney p=0.003).
People who practice higher-void kinks prefer less consent in fantasy — zero reversals
across all five consent levels (Full consent → Full nonconsent).
Spearman ρ=1.00 across 5 ordered bins, p≈0.017 two-tailed.
The bin-level result is clean; the individual-level Spearman (ρ=+0.267, p=7×10⁻²⁴⁹,
N=15,324) confirms the direction at high significance with a more modest effect size.
Both pass FDR. The direction is not in question; the magnitude at the individual level is modest.
Evolutionary reading
Mate guarding theory. High-void mating strategies evolved to suppress exit options —
coercive control, mate guarding, Bateman gradient reversal. The higher the coupling
depth, the greater the evolutionary pressure to prevent partner defection. The framework
detected this structure (ρ=1.00) with no evolutionary assumptions in the scoring.
ρ=−1.00
Shame and therapeutic value flip from independent to anti-correlated above V*
FDR ✓
For low-void kinks, shame and therapeutic value are statistically independent.
Above V*, they flip to anti-correlated: the kinks that carry the most therapeutic value
carry the least shame, and vice versa. The trade-off only appears at high void — below
the threshold, kinks can be both high-shame and high-therapeutic without contradiction.
The ρ=−1.00 is a kink-level mean correlation across the supercritical cluster.
Evolutionary reading
Zahavian handicap principle. High shame = high honest signal cost = reliable indicator
of coupling depth at supercritical V. Low-cost signals are cheap talk. The ρ=−1.00
conjugacy at V > V* is the void framework's version of honest signaling theory:
the cost (shame) guarantees signal integrity at high void.
ρ=+0.33
Coupling (α) predicts shame — marginal
FDR ✗
Identity-adjacent kinks carry more shame (p=0.022 raw, fails FDR at q=0.05).
Exploratory finding. The partial correlation controlling for opacity drops to
ρ=+0.078 — most of α's effect is mediated by O. Needs confirmation.
ρ≈0
New fetish formation is NOT predicted by kink-level void scores
Inducibility (having formed new fetishes from porn/erotica) shows no
correlation with O, R, α, rarity, or Pe at the kink-category level
(all p > 0.20). This is a clean null — inducibility is a person-level trait
(individual openness and adventurousness), not a property of specific kinks.
Evolutionary reading + platform prediction
Plasticity is in the genome, not the stimulus. Inducibility is phenotypic
flexibility in mate preference — a person-level trait, exactly as evolutionary
psychology predicts. This null is a strong positive result: the framework
is detecting structural coupling, not novelty-seeking.
Falsifiable prediction:
If inducibility is person-level, then high-Pe platforms like TikTok (Pe=22.1)
don't create new desire patterns — they select for and amplify
existing high-α users. The coupling effect is selection, not induction.
See social media Pe analysis →
The deeper result — directly measuring an information-geometric obstruction
The correlational findings above tell you what predicts shame. This section is different: it's a
direct measurement of a structural property the framework predicts must exist in any system where
two information channels (the rubric scoring kinks, and respondents' kink profiles) are both
aimed at the same outcome (shame or consent preference) without being structurally independent of each other.
That property is the conditional mutual information I(D;M|Y) — how much information the two channels
share given the outcome. The framework predicts this is always positive, and that it grows
with engagement depth. We measured it directly on the BKS respondent data.
Y = shame
H(Y) = 1.932 bits
I(D;Y) = 0.028 bits
I(M;Y) = 0.033 bits I(D;M|Y) = 0.433 bits z vs null = +295σ N = 14,102
Y = consent
H(Y) = 1.988 bits
I(D;Y) = 0.049 bits
I(M;Y) = 0.060 bits I(D;M|Y) = 0.422 bits z vs null = +358σ N = 14,102
The penalty is ~7× larger than the channel capacity it's eating.
I(D;Y) + I(M;Y) ≈ 0.06 bits (the total info D and M carry about Y combined) is dwarfed by
I(D;M|Y) = 0.43 bits (the penalty). The rubric and the respondents are not independent channels
onto Y — they're the same projection seen twice. This is the framework's central diagnosis
in its purest form: blended-output two-point geometry pays an explaining-away tax that exceeds
its own predictive power.
500-shuffle permutation null with N=14,102: null I(D;M|Y) under D-shuffle has mean ≈ 0.014 bits (plug-in
estimator bias) and SD ≈ 0.001. Observed is 30× the bias and 295-358 standard deviations from null —
this isn't estimator noise. Plug-in MI is biased upward; the absolute magnitude will shrink under a KSG
or shrinkage estimator, but the ordering (penalty >> combined I(D;Y)+I(M;Y)) and the permutation
significance are robust.
Substrate count. The framework had previously confirmed
I(D;M|Y) > 0 on seven substrates: transformer attention, quantum simulation, thermodynamic systems,
real quantum hardware, abstract softmax architectures, biological neural circuits (C. elegans connectome),
and the language/phoneme manifold. BKS adds an eighth:
sexual psychology survey data, N=15,324 respondents, z > 290σ for both shame and consent labels.
The penalty is substrate-independent by Čencov uniqueness (1972) — the same theorem that makes the
Fisher information metric unique on statistical manifolds forces this obstruction to exist on any substrate
where the geometry applies. This is one more confirmation where it was predicted to appear.
KSG cross-check (Kraskov 2004 + Frenzel-Pompe 2007, k=5, N=5,000 subsample):
I(D;M|Y=shame) = 0.506 bits (penalty/capacity = 12.7×, z=+84σ),
I(D;M|Y=consent) = 0.491 bits (penalty/capacity = 4.3×).
The KSG estimator gives LARGER penalty values than plug-in, with even larger penalty/capacity ratios.
The framework prediction (penalty >> capacity in two-point geometry) is robust to estimator choice;
the absolute magnitudes survive bias correction.
Structure Theorem on BKS — ∂I(D;M|Y)/∂engagement > 0
The framework predicts a penalty that grows with engagement depth — the explaining-away obstruction
is not fixed, it scales. Stratify the 14,102 valid respondents into engagement quartiles by mean
kink-intensity and compute the conditional mutual information within each band.
Prediction: monotone increase across Q1 → Q4.
Engagement band
Mean kink intensity
I(D;M|Y=shame)
I(D;M|Y=consent)
Q1 (low)
1.92
0.299 bits
0.293 bits
Q2
2.61
0.290 bits
0.299 bits
Q3
3.09
0.375 bits
0.355 bits
Q4 (high)
3.75
0.479 bits
0.454 bits
Q4 vs Q1: 1.60× growth for shame, 1.55× for consent.
Spearman across 4 bands: ρ = +0.80 (shame, p=0.20 with N=4), ρ = +1.00 for consent (p=0.042 exact).
The shame curve has a tiny Q1→Q2 dip (likely sample variance with KSG); Q2→Q3→Q4 is strict monotone.
Structure Theorem confirmed on the BKS substrate at strict monotone for consent, partial monotone for shame.
This is one of the only direct measurements of ∂I/∂E on human-psychology data — the framework's RLHF-self-undermining
prediction transferred to a biological-attention-gradient substrate.
Pre-registered alternative: penalty peaks at moderate engagement then declines (saturation).
On BKS we see no saturation in [1.9, 3.75] — the substrate sits in the rising portion of the curve.
Full pipeline available on request.
Joint-corner structure on BKS — redundancy, not synergy
The three rubric axes (O, R, α) can be collapsed to a single score only if they're not in a
strong-synergy regime — where the joint triple carries far more information than the sum of pairs.
On the BKS substrate:
The BKS rubric axes are highly redundant (pairwise MI all > 0.6 bits) — the three
dimensions are largely saying the same thing on this substrate. Compare: the C. elegans connectome
sits in the opposite regime (II = −0.631 bits, strong synergy), where collapsing to a scalar
discards most of the signal. On BKS, a single score is essentially optimal because there's only
~1.5 independent dimensions of rubric information to start with. The full three-axis vector form
was computed but adds no predictive power here: Spearman(scalar, vector L²) = +0.59 on these 49 kinks.
This is informative — different substrates live in different joint-corner regimes
and the BKS sits in the redundancy corner.
The evolutionary biology is the same math
Three Athanor simulations (nb_evo01–03, 2026-02-26) showed that the void framework equations
re-derive independent evolutionary results exactly — not as analogy, but as algebraic identity.
The same formula, different coordinate mapping.
01
Sexual selection / Bateman gradient
Pe(α) = K · sinh(2 · BA · α)
Map mating strategy coupling depth α to the void constraint: c(α) = c0·(1−α),
where c0 = √6/π ≈ 0.390.
ESS boundary at Pe* = 4 corresponds to a c-cut on the formula.
The 9 BKS kinks above V* span Pe ∈ [3.84, 25.16] — two (Nonconsent Fantasy, Psychological
Torture, both V=6.0) sit below Pe*=4, confirming V*=5.52 and Pe*=4 are distinct thresholds.
Verified: supercritical mean shame 2.12 vs subcritical 0.88. Direction matches BKS ρ=+0.40.
02
Hamilton's rule — Pe < 0 = cooperation
Pe(r, B/C) = K · sinh(2 · BA · (1 − r · B/C))
Pe = 0 contour is the Hamilton boundary rB = C — this is forced by the formula's
zero crossing, not measured.
When rB > C: Pe < 0 — cooperation is the attractor.
When rB < C: Pe > 0 — defection is the attractor.
c0 = √6/π ≈ 0.390 is the c-value where the scalar Pe argument crosses zero.
The structural identity Pe(rB=C)=0 is a coordinate map, not a numerical claim about Hamilton's threshold.
p = strategy prevalence. At p=1 (universal): Pe=0 — no selection cost.
At p→0 (rare): Pe→43.9 — maximum fitness cost.
The BKS rarity→shame finding (ρ=+0.40, FDR ✓, N=15,503) is this curve.
Shame is the evolved fitness-cost signal of minority strategy position —
partner scarcity + social exclusion, not cultural noise.
Verified: supercritical mean shame 2.12 vs subcritical 0.88. Direction matches BKS ρ=+0.40.
What this means
The framework wasn't built from evolutionary biology — it was derived from information theory and thermodynamics.
The fact that it re-derives Bateman gradients, ESS boundaries, Hamilton's rule, and frequency-dependent selection
from the same three equations (O, R, α → Pe) is a cross-domain convergence, not a retrofit.
Each one is an independent confirmation. The underlying math is substrate-independent.
Click any sphere in the 3D scatter above to see the evolutionary reading for that kink.
Or use the calculator below — it now shows Pe and which mechanism applies.
The V* boundary
V* = 5.52 was originally framed as the precise drift-selection threshold. Empirically on this 49-kink set
the threshold is degenerate over V ∈ [5.0, 6.5] — no kinks fall in V ∈ (5.52, 6.00), so V*=5.52
and V≥6.00 produce identical splits (9 kinks each, both at Mann-Whitney p=0.003). V≥5.00 also works (13 kinks, p=0.005).
The boundary picks an empty gap rather than a sharp data feature. 9 of 49 kinks sit above V*=5.52 — that count is correct;
the precise threshold value is calibration-dependent.
Important: the data-driven measured-void cut (L² norm of standardized [shame, opacity-residual, rarity])
doesn't reproduce this split at all — top-decile of measured void splits shame at p=0.64 (no signal).
The rubric V cut is doing real work that pure measured features don't capture; the rubric and the data identify
substantially different "high-void" kink sets (overlap ~3-4 of 12).
Author's note — Mindbreak at V=8.0
Mindbreak sits at the apex of the supercritical cluster. Its R score (4.33) was initially questioned on the grounds that it is primarily a fantasy/erotica genre rather than a practiced relational dynamic — if consumed as fiction, what "real-time adaptive feedback" is actually required?
The counterargument, which we find more structurally sound: the Mindbreak fantasy is not about observing psychological dissolution in a third party. The target state is the subject. The fantasy models the practitioner's own transition from agentive self to void state — D1→D2→D3 experienced from the inside.
In that framing, R is high because the mechanism is predicated on a maximally responsive operator: someone who can read every resistance threshold precisely enough to navigate past all of them. The responsiveness is structural to the fantasy even in its purely imagined form, because the fantasy requires constructing a partner capable of that navigation.
The remaining supercritical kinks (Obedience, Master/Slave, CGL, Full-Time Power Exchange) are coherently interpretable as partial trajectories toward the same endpoint. Mindbreak is the attractor. This is consistent with its position in the 3D scatter — it sits at the extreme of the high-O, high-R, high-α corner, with the others clustered below it along the same gradient.
Whether human raters converge on R≈4 or score it lower remains an open empirical question and one of the more interesting things the IRR study will settle.
1
Human IRR raters — the only hard blocker
We have ICC=0.90 inter-rater reliability — but with AI raters (3× claude-haiku,
independent contexts). Psychology journals require human raters.
We need 3 people to score 49 kink categories on three dimensions (O, R, α)
using a structured rubric. Takes about 45 minutes per person.
All definitions are operationalized. No domain expertise required —
just careful reading of the rubric.
This is a secondary analysis of Aella's survey. She designed the instrument,
collected 15,503 responses, and published the data openly. That is a co-authorship
contribution at any journal. If she wants to co-author, even better —
she can validate the kink descriptions and scoring rubric from domain expertise.
If not, full acknowledgement and citation.
3
Target journal selection
Best fit: Journal of Sex Research or Archives of Sexual Behavior
for the findings. PLOS ONE or Frontiers in Psychology for
faster turnaround and the Athanor framework introduction.
The paper is exploratory — we frame it as such and apply FDR throughout.
All 6 primary findings already pass Benjamini-Hochberg correction.
4
Tracking: Aella methodology comparison
Aella is publishing a methodology comparison paper —
"Me vs. the Entire Field of Fetish Research"
— comparing her BKS approach against the academic fetish literature.
We are watching this. If her structural findings converge with ours
(opacity, rarity, and void intensity as structural predictors the field missed),
this becomes a pre-registered independent replication.
The void framework gives her language for the structural results the field
has failed to find. That is a research tool, not just a citation.
If convergence is confirmed: the evolutionary bridge paper becomes stronger —
six mechanisms, independent replication, three substrates.
The scoring rubric (open)
Three dimensions, 0–5 scale each. The rubric is what makes the IRR valid.
Human raters get only these definitions — no exposure to prior scores or outcomes.
O — Opacity (0–5)
How hidden or concealed is the mechanism of arousal from a naive outside observer
who knows nothing about kink culture? 0 = completely transparent (anyone understands
the appeal immediately). 5 = mechanism completely hidden even with explanation.
This is NOT about taboo, stigma, or social acceptability.
It is about mechanism comprehensibility.
R — Responsiveness (0–5)
How much does this kink require real-time adaptive feedback and mirroring from another
person to function? 0 = purely solo/fantasy. 5 = the dynamic IS the adaptive mirroring
(e.g., training, conditioning). This is about the partner's adaptive role,
not physical presence.
α — Coupling (0–5)
How deeply does this kink capture attention when engaged? How high is the
obsession/identity-absorption potential? 0 = mild interest, easily set aside.
5 = total attention capture, often identity-defining, lifestyle-level.
Score based on typical practitioner experience, not most extreme case.
Calculate void intensity — any kink
Score any kink on O/R/α. Computes V, Pe, phase, and evolutionary reading.
Try Mindbreak (O=4, R=4.3, α=5) — it's supercritical at Pe≈+36.
O2.0
R2.0
α2.0
3.60V — void intensity
+15.7Pe — drift rate
FLUIDPhase
Sub-V*Drift zone
⚡ V* = 5.52 — the drift-selection threshold
When V exceeds 5.52, the second law requires drift amplification.
Consent preference splits — supercritical kinks carry higher non-consent orientation.
This is not a value judgment. It is a thermodynamic prediction confirmed on N=15,503 respondents.
Current IRR results (AI raters)
3 independent API calls, fresh contexts, T=0.3. ICC(2,1) — two-way random, absolute agreement.
These will be replaced by human rater scores before submission.