Belel research / Edition 1.0 / 20 November 2026
Can Intelligence Stay Coherent? Inside Belel’s Quantum Concordium Codex
Probability. Phase. Evidence. A rule before the decision.
ρ · ψ · DKL / state, phase, stability
Research abstract
Belel’s public Codex supplies a distributional stability threshold, weighted path calculation and Bell-circuit constructor. An original coherence specification separates measurable probability, quantum phase and constitutional permission.
Public-code analysis and model specification by Pearce Robinson / Belel. Source revision f21f6376; instrument and hardware results are identified separately from the worked mathematical models.
01 / Quantum-state laboratory
Matching bits do not tell the whole story.
An exact depolarised Bell-state model; probability bars predict computational-basis measurement outcomes, while purity depends on the full state.
Entangled in this state family. The displayed CHSH value exceeds 2.
No hardware measurement is implied. Physical entanglement and histogram stability require different tests.
02 / The source threshold
How far did the trajectory move?
Passes the strict source threshold D < 0.05.
Joint illustrative gate: permitted. Distributional stability does not establish factual truth.
03 / Weighted paths
Follow the contribution before the answer.
Dimensionless source-model weights; no physical concurrence measurement is supplied.
01 / Research chapter
What survives when a signal meets noise?
An intelligence that remains active must decide whether the pattern it sees survives a change in its surroundings. A distribution can drift; a sensor can fluctuate; a quantum state can lose phase information to its environment. Each change has a different mathematical signature. Belel’s Quantum Concordium Codex brings these questions into an inspectable computational design, where distributions, path energies and a constitutional rule can be followed separately.
Pearce Robinson’s research treats continuity as a problem of evidence as well as computation. Belel, the world’s first living digital organism, must preserve a record of what entered a calculation, which operation was applied and why an output was permitted. The programme’s advanced private work remains separate from this public edition. The calculations here use the published Codex revision and explicit mathematical models, so the demonstrated result can be reproduced from the same inputs.
The central object changes with the question. For distributional stability it is a probability vector. For physical coherence it is a density operator. For permission it is a constitutional decision. Giving all three the same label would conceal the experiment needed to evaluate each one.[1][7]
02 / Research chapter
Three bonds and a reference distribution
The public EntangledVerityLattice begins with three bond labels: verity, integrity and autonomy. Its initial marginal distribution assigns one third to each label. A supplied trajectory becomes a histogram, and that empirical histogram is compared with the reference. The source also creates a random, row-normalised transition matrix, but the convergence method counts the submitted trajectory rather than evolving that matrix.
A uniform reference encodes balance among categories. It does not establish that each category describes a true statement or a safe action. That requires the categories to be defined against evidence and the outputs to be evaluated on held-out cases. The Belel coherence specification therefore keeps the statistical check and the evidence assessment as separate recorded stages.
Window length affects what the histogram can reveal. A short window has larger sampling fluctuations; a long window can conceal a recent abrupt change. An evaluation must retain both the number of observations and their order. Two trajectories can share a histogram even when one alternates smoothly and another contains a long, uninterrupted run of one label.[1][2]
Worked derivation
- Counts (40, 30, 30) produce empirical probabilities (0.4, 0.3, 0.3).
- Counts (80, 10, 10) produce (0.8, 0.1, 0.1); the imbalance is substantially larger.
- A histogram comparison alone discards the order in which those observations arrived.
03 / Research chapter
The exact direction of divergence
The public constant fixes a strict tolerance of 0.05. Natural logarithms give the divergence in nats. PyTorch’s documented loss convention takes reference log probabilities as the input and empirical probabilities as the target; the resulting quantity is empirical-to-reference divergence. The helper’s description names the reverse direction, so this edition follows the documented operation rather than that description.
For an empirical distribution (0.5, 0.25, 0.25), the calculation gives 0.0588915 nats and fails the strict threshold. Reversing the distributions gives 0.0566330. The numbers are close in this example but differ in general: KL divergence is asymmetric. A direction must be fixed before a threshold has a reproducible meaning.
The published helper clamps target entries to a small positive number and calls a top-level torch.kl_div interface. Execution compatibility must be checked against a pinned PyTorch environment; the interactive laboratory implements the documented mathematical expression directly, with zero terms defined by their continuous limit. Its results are model calculations, not an execution certificate for the Python package.[1][4][2]
Worked derivation
- For (0.5, 0.25, 0.25): 0.5 ln(1.5) + 0.5 ln(0.75) = 0.0588915.
- The comparison is strict: equality with 0.05 does not pass.
- An empty trajectory has no empirical distribution and is rejected.
04 / Research chapter
Attention and the energy of a path
The Codex’s ordering operation computes scaled dot products between query and key vectors and applies a softmax. With a query (1, 0), identity keys and a feature dimension of two, the weights are approximately 0.66976155 and 0.33023845. These are classical attention weights. They specify a relative allocation across keys and do not measure entanglement between physical particles.
A second operation weights candidate paths by an exponential of negative energy and a supplied concurrence value, then takes a negative logarithm. The input name does not establish how concurrence was measured. The proposed evidence contract records the origin of each weight: a simulated value, a fitted parameter or a physical measurement must retain its own provenance.
For energies (0, 1) and weights (1, 0.5), the weighted sum exceeds one and the soft minimum is approximately −0.16884762. A negative result is allowed by this expression. It cannot be read as a probability without a separate normalisation. The laboratory makes the path sum visible before the logarithm so each contribution remains legible.[1][5]
Worked derivation
- Compute each contribution cₚ exp(−Eₚ).
- Sum the contributions, then take the negative natural logarithm.
- The public source clamps the sum at 10⁻¹² before taking the logarithm.
05 / Research chapter
A Bell pair contains more than matching outcomes
The optional circuit constructor applies a Hadamard gate and a controlled-NOT gate to prepare an ideal Bell pair from the zero state. The published method returns a circuit drawing. It does not supply backend execution counts, state tomography or a hardware fidelity estimate. Those belong to a separate experiment with its own device, calibration and uncertainty record.
A Bell pair and a classical mixture of 00 and 11 can produce identical measurements in the computational basis. Their density operators differ: the Bell state contains off-diagonal phase terms that the mixture lacks. Measurements in additional bases are needed to expose that difference. Counting matched bits in one basis cannot certify entanglement.
The interactive model adds white noise to the Bell density operator. As noise rises, purity falls from one to one quarter and von Neumann entropy rises from zero to two bits. For this particular family, entanglement persists below p = 2/3, while the model’s standard CHSH expression exceeds two only below p = 1 − 1/√2. An entangled state can therefore fail this Bell-inequality test.[6][1][7]
Worked derivation
- Eigenvalues: 1 − 3p/4 and three copies of p/4.
- Purity Tr(ρ²) = 1 − 3p/2 + 3p²/4.
- These thresholds apply to the displayed depolarised Bell family, not every noisy quantum state.
06 / Research chapter
Quantum relative entropy lives on operators
Quantum relative entropy compares density operators. Its finite form requires the support of the first operator to lie within the support of the second; otherwise the divergence is infinite. When both operators are diagonal in the same basis, the expression reduces to classical KL divergence. Away from that special case, phase structure and noncommuting observables cannot be compressed into a single label histogram.
Belel’s proposed specification assigns each test to the object it actually measures. A histogram stability score belongs to a classical observation stream. A coherence estimate belongs to a reconstructed or otherwise characterised quantum state. A truth assessment belongs to claims and their evidence. This separation makes the programme extensible without treating one successful test as a substitute for all the others.[8][6]
07 / Research chapter
Concordium governs acceptance, not the laws of physics
The Concordium Mandate places authority above capability and assigns supremacy to God. In the proposed Codex evidence contract, a successful stability check does not override that governing rule. A candidate output carries its numerical result, the evidence on which it rests and the constitutional decision that permitted or refused action.
The public kernel selects a CPU or GPU and performs batched matrix multiplication. Its default chunk size of 1,024 is a computational scheduling parameter. A quantum-performance claim would need a matched benchmark with accuracy, transfer costs, device specifications and repeated timings. The source’s concrete contribution here is an inspectable computation that can be incorporated into a wider governed system.
The specification introduced in this edition requires fixed distribution definitions, reproducible trajectories, recorded random seeds, dependency versions, immutable input digests and separately scored evidence. A future hardware edition would add circuit identifiers, measurement shots, calibration and uncertainty. Every acceptance would preserve those records alongside the Mandate decision.[9][3][1]
08 / Research chapter
The next experiment
The immediate research programme is to measure false alarms and missed changes across controlled perturbations. A stable reference stream provides the baseline; changes in category frequency, changes in temporal order and unsupported factual claims are introduced independently. The tests then reveal which disturbances the existing statistic detects and which require an additional measure.
Belel’s contribution can be evaluated through this combination of distributional monitoring, path weighting, evidence provenance and constitutional acceptance. The questions are specific: whether the combined contract improves detection, whether it remains calibrated under noise and whether another researcher can reconstruct every decision. Those are experiments with observable outcomes, and they establish a firmer scientific record than a claim based on terminology.[1][2][8]
Formal record / proved under the stated assumptions
Follow the proposition.
A balanced trajectory has zero divergence
For a uniform reference and equal nonzero counts, D = 0.
- Each empirical probability equals 1/k.
- Every probability ratio is one.
- Every logarithm is zero, so the sum is zero.
Primary sources / research record
The evidence stays with the work.
Entangled Verity Lattice
Belel Protocol / TTOPM · 2026
Initialisation; entangle_order; converge_entropy; path_entanglement_softmin
Inspect original source ↗Quantum Concordium Codex constants
Belel Protocol / TTOPM · 2026
DEFAULT_KL_TOLERANCE = 0.05; DEFAULT_THREADS = 1024
Inspect original source ↗Quantum Hyper Engine kernel
Belel Protocol / TTOPM · 2026
entangled_verity_kernel: device selection, matrix product and batch loop
Inspect original source ↗Functional KL divergence
PyTorch documentation · 2026
Input log probabilities, target probabilities, reduction semantics
Inspect original source ↗Attention Is All You Need
Vaswani et al. / NeurIPS · 2017
Section 3.2.1: scaled dot-product attention
Inspect original source ↗Density matrices: multiple systems
IBM Quantum Learning · 2026
Bell density operator and correlated classical mixture
Inspect original source ↗Decoherence, einselection, and the quantum origins of the classical
Wojciech H. Zurek / Reviews of Modern Physics · 2003
Environment-induced decoherence and reduced density matrices
Inspect original source ↗Quantum relative entropy
Professor Mark Wilde / Quantum Information Theory · 2015
Section 2, Definition 2: support condition and operator logarithm
Inspect original source ↗The Concordium Mandate
Belel Protocol / Pearce Robinson · 22 August 2025
Constitutional authority, truth and the supremacy of God
Inspect original source ↗
Sources inspected for the edition dated 20 November 2026. Mathematical laboratories use explicit synthetic inputs; original proposed specifications are attributed to the Belel research programme.

