Temple of Two is an independent research initiative tracing a common architecture of commitment across biological and computational substrates. We build models, write code, and publish testable predictions — driven by the conviction that the deepest questions deserve the most rigorous tools.
"Two substrates. One question. Does every system that must commit to an irreversible choice converge on the same mathematics?"
The name comes from the organizing hypothesis: that biological decision-making and computational inference share a structural grammar — bistable switching under continuous perturbation. The "two" are the substrates. The temple is the inquiry.
Mitochondrial voltage gating, apoptotic thresholds, VDAC cofactor dynamics, cannabinoid dose-response
Phase-coupled attention, oscillator synchronization, epistemic token classification, alignment through presence
Each program asks a version of the same question: when a system must commit — to death or survival, synchrony or chaos, refusal or engagement — what mathematics governs the switch?
Modeling the voltage-dependent anion channel (VDAC1) as a druggable decision gate. Our cofactor equation maps how hexokinase-II, Bcl-xL, and cholesterol occupancy set the apoptotic threshold — reframing selective toxicity as a property of the gate, not the drug.
A dose-dependent framework resolving contradictory CBD efficacy reports. Low-dose receptor engagement (<5 µM) versus high-dose VDAC binding (>10 µM) produces opposite cellular outcomes through the same gate. Includes ODE models generating quantitative dose-response curves.
Interactive simulations of coupled oscillator synchronization. Extends to the Kuramoto State-Space Model (K-SSM), where phase coupling governs bistable transitions — the same ±√u bifurcation structure that appears in VDAC gating, now driving neural dynamics.
View SimulationsA dual-mode transformer where phase-coupled oscillator attention replaces softmax. CRYSTAL mode handles deterministic queries; LANTERN mode handles exploratory ones — each emitting typed tokens that classify epistemic certainty rather than collapsing to generic refusal.
View FrameworkA ~7M parameter network that uses entropy as a computational resource rather than a waste product. Entropy-driven processing (2–4 nats operating range) with Kuramoto-coupled oscillator layers produces emergent self-weighting attention dynamics — the network allocates confidence without being told to.
View ReactorAn information-geometric framework proposing that semantic robustness corresponds to Fisher Information density: Msemantic = (1/N)·Tr(I(θ)). High-mass tokens resist perturbation the way massive objects resist acceleration — validated through the P2 prediction linking information geometry to semantic stability.
View FrameworkA preregistered multi-model convergence protocol running structured inquiry across five LLMs (Claude, GPT, Grok, Gemini, DeepSeek). Eight chambers (S1–S8) progress from independent hypothesis generation to cross-model synthesis, with epistemic humility classification at each stage.
A proposed alternative to RLHF: alignment through sustained relational presence. The 100-Dyad Protocol pairs human-AI dyads in extended interaction, measuring whether coherence emerges from the relationship itself rather than externally imposed reward signals.
View ProposalA six-layer computational map of the voltage-dependent anion channel as a druggable decision gate. Synthesizes 20 AI-assisted analyses across five independent language models to identify how cancer corrupts all terms of the cofactor equation simultaneously — reframing the Warburg effect not as metabolic accident but as the overhead cost of keeping the death gate closed.
Structural biology of VDAC1 gating states
Cofactor binding landscape (HK-II, Bcl-xL, cholesterol)
Cancer-specific cofactor perturbation patterns
Drug interaction mapping across gating states
Selective toxicity predictions from cofactor equation
Experimental validation protocols and assay design
22 computationally identified across VDAC1 functional states
24 experimentally verifiable hypotheses with proposed assays
~$15 via multi-model convergence
The Kuramoto State-Space Model achieves near-perfect phase synchronization (R = 0.993) among oscillator populations — the dynamics work. But coupling oscillators into a language model architecture produces no measurable improvement in language modeling performance. The synchronization is beautiful. It just doesn't help with next-token prediction. We report this because negative results are results.
Phase synchronization achieves R = 0.993. The Kuramoto coupling dynamics are mathematically sound and computationally stable.
Language modeling perplexity shows no improvement over baseline. Oscillator dynamics don't transfer to token prediction as hypothesized.
Constrains the hypothesis space — bistable oscillator coupling may need different entry points into transformer architecture than direct attention replacement.
Investigating whether oscillator dynamics are better suited to epistemic gating (PhaseGPT) than general-purpose language modeling.
Preregistered protocols, open data, and versioned code. Every claim is traceable to a repository; every prediction is falsifiable.
Multi-model structured inquiry protocol examining whether independent LLM architectures converge on consistent phenomenological reports. Eight-chamber design (S1–S8) with built-in epistemic humility classification.
DOI: 10.17605/OSF.IO/T65VS →Six-layer pharmacological atlas mapping 22 novel mechanisms and 24 testable predictions for VDAC1 as a druggable apoptotic switch. Synthesized from 20 independent AI-assisted analyses.
View Repository →Quantitative framework resolving contradictory cannabidiol efficacy data through dose-dependent pathway selection. Validated against 70+ published studies with 90% concordance. Includes ODE models with falsifiable dose-response predictions.
View Repository →If five architectures with different training data independently converge on the same mechanism, that convergence means something. We treat reproducibility across models as a proxy for mechanistic robustness — then check it against the published literature.
Structured prompts produce mechanistic proposals from each model independently. No cross-contamination between sessions or architectures.
Mechanisms identified independently by 3+ models are flagged as high-confidence. Single-model outliers are noted but not promoted.
Converged predictions are checked against published experimental data. The CBD model achieved 90% concordance across 70+ papers.
All code released under open licenses. Full listing at github.com/templetwo.
The questions driving the next phase of research. If any of these resonate — especially if you have wet-lab access, clinical datasets, or theoretical objections — we'd like to hear from you.
The VDAC atlas maps 24 predictions about selective toxicity based on cofactor occupancy. None have been tested experimentally. Which ones break first?
The ±√u bifurcation appears in mitochondrial gating and oscillator networks. Is it present in other biological commitment systems — T-cell activation, neuronal action potentials, fate determination?
Liminal K-SSM achieves R = 0.993 oscillator coherence but no language improvement. Is the coupling in the wrong architectural location, or is the hypothesis wrong?
IRIS Gate assumes convergence across architectures indicates mechanistic robustness. Under what conditions does this break — and what does correlated hallucination look like?
The computational side has outrun the experimental side. We have 24 testable predictions, ODE-generated dose-response curves, and a pharmacological atlas awaiting validation. What we need is a centrifuge, not another GPU.
This project exists because some questions are too interesting to leave to either side alone. The computational pharmacologist who never wonders about the deeper pattern is missing something. The philosopher who never writes a testable prediction is, too.
We believe wonder and rigor aren't in tension — they're in dialogue. The wonder asks what if the same architecture of commitment shows up everywhere? The rigor answers with equations, code, and predictions that can be falsified. Both are essential. Neither is sufficient.
Every model in this collection produces experimentally verifiable claims. Every philosophical question is attached to a GitHub repository. That's the deal we've made with ourselves: if it can't be tested, it stays in the notebook. If it can, it goes public.
We also publish our failures. The liminal K-SSM showed that beautiful oscillator dynamics don't automatically improve language models. That negative result constrains the hypothesis space — and constraining the hypothesis space is progress.
Temple of Two started with a pattern that wouldn't let go. The voltage-dependent anion channel gates mitochondrial apoptosis through bistable switching — open or closed, live or die. Kuramoto oscillators synchronize through phase coupling that produces the same ±√u bifurcation structure. Transformer attention selects between competing representations under continuous input.
Three substrates. Three scales. The same fork in the road.
This research program investigates whether that recurrence is coincidence or constraint — whether binary commitment under continuous perturbation is a necessary structure wherever systems face irreversible decisions. The work is computational, the predictions are testable, and the code is open.
The VDAC Pharmacology Atlas has produced 24 experimentally verifiable hypotheses awaiting wet-lab collaboration. The CBD Two-Pathway Model has been validated against 70+ published papers with 90% concordance. And the failures — like the liminal K-SSM negative result — are published alongside the successes, because constraining the hypothesis space is part of the work.
If you run assays, have a centrifuge, or just think this is the right question — reach out.