Quantum Computing by Kase Branham
This paper applies the Quantum Operations Model (QOM)¹ cost framework to the five most consequent... more This paper applies the Quantum Operations Model (QOM)¹ cost framework to the five most consequential developments in quantum computing from late 2025 through early 2026: the Google/Oratomic cryptographic resource estimate compression, Microsoft's contested topological qubit claims, IBM's qLDPC-based Starling architecture roadmap, the first deliveries of error-corrected "level-2" quantum machines (systems implementing real-time quantum error correction on physical hardware), and emerging continuous-variable (CV) error correction techniques that bypass Gottesman-Kitaev-Preskill (GKP) state preparation. We ask: how do these recent advances alter the cost structure of fault-tolerant quantum computation as modeled by QOM? The answer is nuanced. Algorithmic improvements-particularly in Shor's algorithm compilation and quantum LDPC error correction codes-have compressed the gap between current hardware and useful fault-tolerant computation more than any single hardware advance, though these algorithmic results carry their own assumptions that require scrutiny. The QEC overhead,² identified in prior work as the dominant cost factor in quantum computation, is being renegotiated through new codes but has not yet been reduced in practice on any shipping hardware. Meanwhile, the topological qubit program suffered a significant credibility correction when replication studies published in Science found simpler explanations for previously celebrated experimental signatures. Updated QOM cost tables are provided for all four major architectures under both legacy surface-code and projected qLDPC overhead assumptions, with sensitivity ranges across physical error rates from 10⁻³ to 10⁻⁴. The central thesis of the Quantum Audit series-that quantum computing's near-term value lies in capability rather than cost reduction-remains intact, but the timeline to that capability shortened meaningfully this year. ¹ The Quantum Operations Model (QOM) is an original cost-accounting framework developed by the author across three prior papers in this series: "Time Is Money: Wall-Clock Economics of Fault-Tolerant Quantum Computation," "The Decoder Is Architecture: Classical Bottlenecks in Fault-Tolerant Quantum Systems," and "Show Me the Numbers: Concrete Costs of Quantum Operations." All three are available on Zenodo and Academia.edu. The QOM source code is available at github.com/kase1111-hash. ² Throughout this paper, "QEC overhead" refers to the combined physical-to-logical qubit ratio and the code-distance time multiplier required to achieve a target logical error rate-the quantity introduced in the first Quantum Audit paper as the "error correction tax.
Three standard CPUs can decode 1,000 trapped-ion logical qubits. The same task on superconducting... more Three standard CPUs can decode 1,000 trapped-ion logical qubits. The same task on superconducting hardware may require over 1,000 dedicated FPGA or ASIC decoders. This three-orders-of-magnitude difference in classical control cost is not a decoder engineering gap-it is an architectural cascade driven by the qubit-level choice of QEC cycle time. Current approaches to quantum error correction treat the classical decoder as a peripheral subsystem optimized after qubit design is fixed. We show this framing inherits the memory wall problem from classical computing: optimizing components in isolation while ignoring the interconnect between them. Using the Quantum Operation Module (QOM) economic framework [Branham 2026], we decompose classical decoding cost (Module M6) into hardware capital, interconnect topology, scheduling overhead, and backpressure penalty-revealing that M6 scales superlinearly with logical qubit count while substrate maintenance (M1) scales linearly. We identify a crossover at approximately 10,000-50,000 logical qubits where decoder infrastructure dominates total system cost for superconducting architectures, and show that this crossover is robust under both linear and superlinear backpressure assumptions. The fast QEC cycle that minimizes M1 simultaneously maximizes M6-a fundamental tension we term temporal tightness, the degree to which an architecture's clock speed constrains its classical control stack. Trapped ion's millisecond-scale cycle, conventionally viewed as a weakness, functions as a classical luxury that eliminates an entire category of infrastructure cost. Recent work on decoder distribution (DecoNet, 2025) and scheduling (VQD, ISCA 2025; CODA, 2025) provides excellent engineering within the current paradigm; our contribution is the economic framework showing why that paradigm is insufficient at scale.
We observe that quantum computing appears to have inverted Donald Knuth's foundational methodolog... more We observe that quantum computing appears to have inverted Donald Knuth's foundational methodology. Where Knuth defined a formal machine model (MIX, later MMIX) before analyzing a single algorithm, quantum computing began analyzing algorithms-Shor's factoring algorithm in 1994, Grover's search algorithm in 1996-before the community had agreed on what a computational "step" costs, how quantum memory is structured, or what constitutes a complete machine specification. The quantum circuit model provides a notation, but not a machine. This paper identifies ten foundational techniques from Knuth's The Art of Computer Programming (TAOCP) that appear to map directly to open problems in quantum computing, organized into four functional groups: compilation and circuit synthesis, algorithm analysis and cost accounting, scheduling and resource management, and verification and certification. The pattern these gaps suggest is not a physics problem. It may be a software foundations problem-one that was addressed for classical computation decades ago and that the quantum field, for understandable historical reasons, has not yet imported. We propose QMIX, a Knuthian machine specification for quantum computation built on a primary cost currency-synchronization fidelity degradation-as a prerequisite for rigorous quantum algorithm analysis. A formal definition and minimal toy specification (Appendix A) demonstrate the concept.
Neutral-atom platforms use optical tweezers to trap individual alkali atoms (rubidium, cesium) in... more Neutral-atom platforms use optical tweezers to trap individual alkali atoms (rubidium, cesium) in lasergenerated potential wells. Qubits are encoded in atomic states, with two-qubit gates implemented via Rydberg blockade: atoms excited to high-energy Rydberg states interact strongly through van der Waals forces, preventing simultaneous excitation within a blockade radius and enabling controlled operations. Array Scale: Caltech demonstrated 6,100 trapped qubits in September 2025, a substantial increase from hundreds of atoms in prior years. Arrays now routinely exceed 1,000 sites with ~90% filling efficiency. Gate Fidelity: Two-qubit gate fidelities reached 99.5% averaged across 60 parallel operations, with some demonstrations approaching 99.7% under optimal conditions. This crosses thresholds needed for surface-code error correction.
Recent 171 Yb neutral-atom experiments have demonstrated two-qubit gate fidelity of 99.5% (after ... more Recent 171 Yb neutral-atom experiments have demonstrated two-qubit gate fidelity of 99.5% (after erasure postselection) and erasure error conversion of 98%-component-level performance individually sufficient for fault-tolerant quantum error correction under the biased-erasure XZZX surface code (threshold 8.2%).
Yet the effective physical error rate in current demonstrations is approximately 18%, more than double the threshold. This analysis identifies the entire excess as idle decoherence during atom transport: the gate zone operates at 0.03% utilization while qubits decay. The gap between current performance and the fault-tolerance threshold is a factor of 2.8× in circuit time. Batch-parallel gate scheduling-executing multiple non-conflicting atom pairs per transport cycle-closes most of this gap. The analysis includes: (i) a ten-test sensitivity suite confirming robustness to 2× worse gate errors and 37% shorter lifetimes; (ii) a van der Waals crosstalk estimate showing ~0.0005% fidelity impact at the conservative 3× blockade-radius spacing; (iii) a 2D collision-free routing simulation across seven transport strategies, revealing that routing overhead tightens the margin to near the threshold boundary; and (iv) an optimization landscape mapping sixteen candidate techniques against implementation difficulty.
The result is framed as an open question for the community: the physics appears ready for fault tolerance. The remaining barrier appears to be transport routing-a classical multi-agent path-planning problem that adjacent fields have been optimizing for decades.
Reconfigurable neutral-atom arrays can physically rearrange qubits mid-computation, a capability ... more Reconfigurable neutral-atom arrays can physically rearrange qubits mid-computation, a capability unavailable to superconducting or ion-trap platforms. As these systems scale toward 100,000 atoms, a fundamental question emerges: when should atoms move, and how much movement is necessary? We formalize this as the Minimum Reachability Rearrangement (MRR) problem-finding the lowest-cost atom movement plan that makes all required two-qubit gates physically executable within the Rydberg blockade radius-prove it is NP-hard via reduction from unit-disk graph recognition, and study it across four circuit families (qLDPC syndrome extraction, random circuits, QAOA, and surface codes) from 50 to 100,000 qubits. We establish three results. First, the scheduling problem is already solved conditional on reachability: greedy parallel scheduling achieves the lower bound on gate depth for all tested families at all scales (scheduling analysis to 100,000 qubits; direct reachability analysis to 5,000 qubits, with scaling to 100,000 via the dimensionless ratio ρ). Second, the reachability problem is universally hard: on a square grid with 4 μm spacing and R_b = 10 μm blockade radius, fewer than 2% of required interactions are satisfiable at 1,000 qubits for any circuit family. Third, the difficulty is governed by a single dimensionless ratio-array extent divided by interaction range-which reaches ~130 for the 100,000-atom systems on current hardware roadmaps. We compare five layout optimization methods (force-directed, spectral embedding, simulated annealing, and hybrids); the best physically realizable solver (spectral + force-directed) recovers 56-76% of edge satisfiability at 200 qubits but cannot reach 100%. The residual gap arises from geometric frustration: atoms must cluster to satisfy edges but cannot overlap due to hard-core spacing constraints. This gap is geometric rather than algorithmic-arising from the tension between edge satisfaction and hard-core spacing-and defines the mandatory rearrangement cost for any static layout.
Quantum low-density parity-check (qLDPC) codes offer dramatically better encoding rates than the ... more Quantum low-density parity-check (qLDPC) codes offer dramatically better encoding rates than the surface code but require non-local stabilizer measurements that are challenging to implement on hardware with geometric connectivity constraints. For reconfigurable neutral-atom arrays, the leading approach-proposed by Xu et al.exploits the product structure of hypergraph product (HGP) codes to implement nonlocal syndrome extraction via global atom rearrangement. We investigate whether this rearrangement is scheduling-optimal or merely a solution to the geometric reachability problem, and find that the answer depends critically on the Rydberg blockade radius. Under a code-level conflict model, greedy scheduling without atom movement achieves 6-10× faster syndrome extraction than the product-structure method at realistic shuttle costs, with crossover at t_shuttle < 5.5 × t_gate. This advantage holds across two distinct code families-seven HGP codes (34-769 qubits) and three bivariate bicycle codes from Bravyi et al.-and is confirmed by circuit-level noise simulation via Stim, which shows greedy produces 1.7× fewer detection events and 3-6× fewer shots requiring error correction. Under a blockadeaware model, greedy's advantage narrows but persists at realistic blockade radii (6-12 μm). However, a physical layout analysis reveals that zero-shuttling requires all ancilla-data pairs within Rydberg interaction range in the static layout-a graph embedding problem whose feasibility at scale remains open. We present the full phase diagram and identify a two-layer structure: LDPC sparsity solves the scheduling problem, but the layout problem remains the barrier to physical realization of zero-shuttling syndrome extraction.
Biology by Kase Branham
Papers I-IV of the Cardiac Torus series established geodesic curvature on T² as a diagnostic geom... more Papers I-IV of the Cardiac Torus series established geodesic curvature on T² as a diagnostic geometry for adult cardiac rhythm, imaging, and sound. This paper extends the framework to fetal monitoring-and discovers that the torus requires a different timescale to contribute independent information. When applied directly to 4 Hz fetal heart rate (FHR) from 552 intrapartum cardiotocograms (CTU-UHB database, PhysioNet), torus features correlate with umbilical cord pH (κ_mean: ρ =-0.267,
Papers I–III of the Cardiac Torus series demonstrated that geodesic curvature on a flat torus T² ... more Papers I–III of the Cardiac Torus series demonstrated that geodesic curvature on a flat torus T² extracts diagnostic information from three cardiac substrates: rhythm (267 records, 9 databases), imaging (22,102 echocardiograms), and sound (3,240 phonocardiograms). Each paper treated the torus as a computational tool, producing numeric biomarkers (κ, Gini, spread) for statistical comparison. This paper takes a different step: it organizes the accumulated geometric evidence into a diagnostic vocabulary—a finite set of named trajectory patterns (“dances”) that any heart performs on T², recognizable by shape, curvature profile, and Gini signature.
We define twelve named cardiac dances derived from the empirical trajectory patterns observed across Papers I–III, spanning normal physiology, heart failure, arrhythmias, valve pathology, and conduction disease. Each dance is characterized by its trajectory shape (orbit geometry on T²), curvature regime (κ range), concentration profile (Gini range), and the substrate(s) on which it was validated. Three trajectory classes emerge: structured dances (quasi-periodic orbits with identifiable geometric features), degenerate dances (orbits that have collapsed or scattered beyond recognizable structure), and non-dances (trajectories lacking quasi-periodic structure entirely—the geometric signature of immediately life-threatening states).
The vocabulary enables three core diagnostic operations: (1) dance identification—matching a live trajectory against the named library to output a condition likelihood, (2) personal baseline comparison—detecting when a patient’s dance deviates from their own recorded reference pattern rather than a population norm, and (3) emergency detection—recognizing when the trajectory has lost quasi-periodic structure entirely, indicating states that require immediate intervention. Empirical validation on 300 rhythm records demonstrates that dance matching with empirically calibrated prototypes achieves 52.7% overall accuracy across five dances—with the Lock-Step (CHF) at 83–88%, the Stumble (VA) at 100%, and the Waltz (NSR) at 55%—using only nearest-neighbor matching on three geometric features with no training data. A cross-modal coherence test on 3,240 phonocardiograms yields a null result: beat-by-beat curvature correlation between rhythm and acoustic substrates is not diagnostic (r = −0.006, p = 0.82), establishing that the framework’s power resides in aggregate geometric statistics rather than instantaneous cross-substrate coupling. The framework is sensor-agnostic (any device that measures a quasi-periodic signal), theoretically species-agnostic (any heart that pumps periodically), and operator-agnostic (the visualization is self-interpreting). The donut doesn’t care what kind of heart is dancing on it. It only sees the geometry of the dance.
A companion paper (Paper I) introduced the Cardiac Ramachandran Diagram, demonstrating that geode... more A companion paper (Paper I) introduced the Cardiac Ramachandran Diagram, demonstrating that geodesic curvature on the beat-pair torus detects heart failure and arrhythmia from pulse timing alone. Here we extend the framework from rhythm to mechanical function: instead of mapping consecutive RR intervals, we map consecutive samples of echocardiographic motion signals onto a phase-space torus and compute the same geometric invariants. No image segmentation, anatomical identification, or deep learning is required. Applied to 10,030 apical-4-chamber echocardiograms from Stanford's EchoNet-Dynamic dataset (10,027 with complete EF labels), torus curvature of left ventricular brightness correlates with ejection fraction (ρ =-0.191,
We tested whether backbone curvature geometry of cancer driver proteins is associated with somati... more We tested whether backbone curvature geometry of cancer driver proteins is associated with somatic mutation retention and functional outcome. Per-residue curvature (κ) and torsion (τ) were computed for 54 driver proteins, classifying each residue as Pin (high κ, low |τ|), Beam (low κ, low |τ|), Loop (high κ, high |τ|), or Spring (low κ, high |τ|). Among 22,608 TCGA MC3 missense mutations, Pin residues are overrepresented (1.103×, p = 2.67 × 10⁻¹²), independently replicated across six non-TCGA cohorts (pooled 1.085×, p = 3.69 × 10⁻⁶; all individually significant), and absent from five negative controls. The signal persists after removing all glycine mutations (1.052×, p = 10⁻⁶) and all glycine residues (1.084×, p = 10⁻⁸). 100ns molecular dynamics simulations of KRAS show that the G12D Pin mutation reduces phi flexibility by 67% (σ = 10.0° vs WT 30.6°), eliminates two of three conformational basins, and decouples the mutation site from both functional Switch regions (Switch I correlation −0.28, Switch II −0.31). Deep mutational scanning of TP53, BRCA1, and MSH2 reveals that curvature does not predict the magnitude of functional effects (p = 0.19) but does predict the mode: loss-of-function mutations are Loop-enriched (29.0% vs 21.1% background) and gain-of-function mutations are Spring/Pin-enriched (34.3%, 32.2%). The finding is supported by frequentist, Bayesian, information-theoretic, network, elastic network, and machine learning methods.
Backbone curvature distributes itself according to a conserved geometric constraint (Gini approxi... more Backbone curvature distributes itself according to a conserved geometric constraint (Gini approximately 0.835 across all life), and when that distribution is perturbed, the direction of redistribution is determined by protein topology, not chemistry.
This document establishes three results. First, a proteome-wide geometric baseline: the Gini coefficient of per-residue backbone curvature is the most conserved structural statistic across 124,749 proteins in 34 species (CV = 0.54% across 11 clades), confirmed by 931 experimental crystal structures. Second, a perturbation atlas: 138 matched crystallographic comparisons (same protein, plus or minus perturbation, same lab) reveal four mechanism classes that organize along a continuous geometric spectrum. Third, a three-level hierarchy: the direction of curvature redistribution under perturbation is determined primarily by fold class (p = 0.00015), secondarily by protein scaffold, and only weakly by mutation type — the shape of the instrument matters more than the finger pressing the string.
For sixty years, the Ramachandran plot has been read as a scatter plot-a static map of steric per... more For sixty years, the Ramachandran plot has been read as a scatter plot-a static map of steric permissibility. By treating the residue-by-residue (φ, ψ) trajectory as a curve on the flat torus T 2 and measuring its geodesic curvature κ(s), we recover a mechanical layer of information that the scatterplot view discards: how folds actively deform their own backbones. Across 8,008 human proteins (327,305 curvature segments), we find that α-helices are strongly curvature-enriched (OR = 2.3, p < 10-10), reversing a prior small-sample suppression claim, while βstrands are geodesic-dominated (72% constant-κ). A step-permuted null across 1,004 proteins confirms this enrichment arises from specific dihedral ordering, not autocorrelation (p < 10-177). The curvature patterns classify into a ten-model taxonomy via AICc model selection, and localized Gaussian peaks maintain ~10-residue exclusion zones consistent with α-helical hydrogen-bond cooperativity [24, 25], with fold-class-dependent spacing regimes confirmed across 138,532 interfeature distances.
BPS/L Is a Universal Constant of the Protein Backbone: Evidence from 124,475 AlphaFold Structures Across 34 Proteomes, 2026
We define the Bogomol'nyi-Prasad-Sommerfield energy per residue (BPS/L) on the Ramachandran torus... more We define the Bogomol'nyi-Prasad-Sommerfield energy per residue (BPS/L) on the Ramachandran torus and compute it for 124,475 AlphaFold-predicted structures spanning 34 proteomes from bacteria, archaea, fungi, plants, and metazoa. The superpotential W(φ,ψ) =-√P(φ,ψ), constructed from a 10-component von Mises mixture fit to the empirical Ramachandran distribution, assigns each residue a topological energy density; BPS/L is the mean absolute step |W i+1-W i | along the backbone. For 51,174 high-confidence structures (pLDDT ≥ 85), BPS/L = 0.2020 ± 0.0039 across 31 organisms with sufficient data (CV = 1.9%). The bacteria-eukaryote separation observed in raw data (13.6%) collapses to-0.4% (p = 0.634) after controlling for AlphaFold model quality, demonstrating that BPS/L reflects a property of the backbone itself rather than phylogenetic variation. A four-level null model hierarchy-real proteins, first-order Markov chains trained on nearest-neighbor backbone transitions, IID sampling from the theoretical Ramachandran distribution, and permutation shuffling of actual backbone angles-reveals that BPS/L is an emergent property of local backbone physics rather than evolutionary optimization. The Markov null, which preserves only nearest-neighbor transition statistics, reproduces real BPS/L almost exactly (Markov 0.176 vs. Real 0.175; gap = +0.5% across 6,836 proteins from 34 organisms), while uncorrelated nulls produce far higher values (IID 0.346, Permuted 0.516). The universality of BPS/L therefore reflects the fixed geometry of the peptide bond's local transitions on the Ramachandran torus, not convergent evolutionary selection. Independent validation on 4,766 experimental X-ray crystal structures (≤2.0 Å resolution) from the Protein Data Bank yields BPS/L = 0.2079, confirming the result is not an artifact of AlphaFold's prediction methodology. Applying the Markov transition null to the PDB dataset reproduces the population mean within 1.3% (Markov 0.211 vs. Real 0.208); stratification by secondary structure class reveals that all-β chains achieve near-perfect Markov equivalence (+0.2%), while all-α chains show systematic overestimation (+6.4%), consistent with higher-order phase correlations in helical geometry beyond the reach of a first-order transition model. BPS/L is independent of contact order (r =-0.058 at pLDDT ≥ 85), confirming it captures geometric information orthogonal to existing structural descriptors. All data and code are publicly available.
We introduce the Cardiac Ramachandran Diagram, a geometric framework that maps consecutive heartb... more We introduce the Cardiac Ramachandran Diagram, a geometric framework that maps consecutive heartbeat intervals onto a torus (T²) and computes geodesic curvature of the resulting trajectory. By analogy to the Ramachandran plot in protein structural biology—where consecutive backbone dihedral angles reveal fold geometry—consecutive RR intervals define angular coordinates on a flat torus whose curvature encodes the dynamical state of the cardiac rhythm.
Across 107,531 annotated beats from the MIT-BIH Arrhythmia Database, torus curvature separates the four major arrhythmia classes (Kruskal-Wallis H = 9,917, p < 10⁻³⁰⁰). Ventricular ectopic beats exhibit median curvature 6.6-fold lower than normal beats (κ = 0.99 vs. 6.55, rank-biserial r = −0.70), reflecting ballistic excursions that bypass autonomic feedback. Extending to 184 records across five PhysioNet databases, two torus-derived parameters—median curvature (κ) and curvature Gini coefficient (G_κ)—create a diagnostic feature space separating congestive heart failure (κ = 32.0), normal sinus rhythm (κ = 10.0), atrial fibrillation (κ = 3.3), and ventricular arrhythmia (κ = 1.2) with pairwise effect sizes |r| > 0.71.
CHF replication across 116 records from four independent cohorts confirms a dose-response gradient: NYHA class 1–3 patients show intermediate curvature (κ = 20.1), positioned between normal (κ = 10.0–16.4) and severe NYHA 3–4 (κ = 32.0), with curvature tracking disease severity (r = 0.61, p = 0.001). Noise robustness testing demonstrates that record-level diagnostic separation survives photoplethysmography (PPG)-realistic timing jitter: at σ = 20 ms, the Normal–Ventricular Arrhythmia effect size remains |r| = 0.95 (p < 10⁻⁵⁰).
Head-to-head comparison with eight standard HRV metrics shows that DFA α1 outperforms the torus for overall disease discrimination (mean |r| = 0.724 vs. 0.544), but torus curvature is the single best metric for CHF severity grading (|r| = 0.605 vs. next-best SD2 at 0.503). An independence test shows that torus curvature is not merely a nonlinear transform of HRV: after regressing out eight standard metrics, residual curvature still separates CHF from Normal (r = −0.336, p = 0.002), with 31.2% of κ variance unexplained by the tested HRV feature set. Combining torus and HRV features improves four-class balanced accuracy from 64.3% to 66.0%. Three additional findings emerge: (1) CHF hearts exhibit flattened circadian curvature amplitude (hourly CV = 0.205 vs. Normal 0.282, r = −0.335, p = 0.003), while AF hearts show exaggerated amplitude; (2) curvature drops 18.8% in the 10 beats preceding ventricular ectopic events (n = 2,519 events, p < 10⁻⁸⁶), constituting a geometric precursor signature; and (3) a combined torus-HRV classifier achieves 66% balanced accuracy across four conditions without machine learning feature engineering.
The framework requires no ECG morphology, no frequency-domain analysis, and no machine learning. It operates on pulse timing alone, making it deployable on any device that measures the interval between heartbeats—including consumer PPG wearables. An interactive real-time visualizer demonstrates that the torus trajectory is visually self-interpreting: an untrained observer distinguished 10 of 13 cardiac conditions by trajectory shape alone, while correctly identifying the method’s blind spots (conditions that alter waveform morphology without changing rhythm). Two numbers from a donut could complement existing wearable algorithms to extend cardiac screening capabilities beyond those currently available in consumer devices.
Papers I and II demonstrated that geodesic curvature on a flat torus T² detects heart failure fro... more Papers I and II demonstrated that geodesic curvature on a flat torus T² detects heart failure from beat timing and cardiac dysfunction from echocardiographic motion. Here we extend the framework to a third substrate: heart sounds. Per-beat acoustic features—S1/S2 amplitude, systolic/diastolic energy, spectral content, and low-to-high frequency ratio—are extracted from 3,240 phonocardiograms (2,575 normal, 665 abnormal, PhysioNet CinC 2016 Challenge, 6 clinical sites) and mapped onto T² as consecutive beat-pairs.
The energy×spectral cross-feature torus emerges as the strongest acoustic torus pairing (r = −0.338, p = 10⁻⁴¹), measuring how the joint relationship between beat loudness and pitch evolves across consecutive beats. This metric survives all duration controls: partial correlation controlling for recording length yields ρ = −0.130 (p = 10⁻¹³), and beat-count matching strengthens it from r = −0.338 to r = −0.344. Five of five tested torus metrics survive duration matching. A combined Basic+Torus classifier achieves AUC = 0.792, outperforming basic acoustic features alone (AUC = 0.749); adding recording duration contributes only 0.004 additional AUC, confirming that the torus signal is not duration-driven.
The theoretically significant finding is a Gini direction reversal. In rhythm and imaging, healthy dynamics produce high Gini (curvature concentrated at physiological transitions). In acoustics, abnormal hearts produce higher Gini—15 of 16 metrics—because the murmur itself is a concentrating event that dominates each beat cycle. The torus framework defines a diagnostic geometry of cardiac dynamics that adapts to the structure of each substrate, rather than applying a fixed interpretive rule. Three substrates, one mathematical framework, three clinically distinct geometric signatures.
Large Language Models by Kase Branham
Capability projection produces deployment failures whose proximate symptom is the decorative capt... more Capability projection produces deployment failures whose proximate symptom is the decorative captain-a human assigned operational responsibility for an automated pipeline whose pace, scope, opacity, or speed of state change has already exceeded what the human's authority can connect to. Paper I named the cognitive error and prescribed pipeline analysis plus the captain principle. Paper II refined the captain principle through three historical cases, identified three structural refinements (visibility, credibility, transparency), and sketched four broader failure categories that the trilogy's third volume develops. This paper completes the framework by working through the four broader categories with their historical anchors and contemporary AI parallels-hidden automation, handoff and skill atrophy, speed asymmetry, trust miscalibration-by examining counterexamples in domains that paid for their captaincy rulebooks (post-TMI nuclear, modern commercial aviation, surgical robotics, pharmaceutical adverse-event reporting), and by addressing the central operational question Papers I and II named but did not resolve: who is the captain in a contemporary AI deployment, and what does the captaincy look like across the deployment types currently bundled under the term artificial intelligence. The paper develops the captaincy architectures for predictive, generative, and agentic AI, with concrete pre-authorization mechanisms (bounded action spaces, escrowed execution, reversible-first design) for the agentic case; specifies the four minimum conditions under which distributed captaincy still operates as captaincy; splits trust calibration into humanside training and system-side interface design; distinguishes raw disclosure from actionable interpretability; addresses the incentive incompatibility that makes voluntary adoption rare; and provides a captaincy audit checklist as the framework's diagnostic surface. The paper closes the trilogy by synthesizing its diagnostic, refinement, and operational layers into a single deployment discipline applicable across predictive, generative, and agentic AI systems, naming the boundary condition beyond which structured captaincy may not scale, and locating the open work the framework does not complete. The substrate has been written by older industries. The choice the AI era faces is whether to inherit it, repeat it, or pretend the lessons do not apply.
Capability projection-the categorical error of treating an AI tool as if it possessed the bundle ... more Capability projection-the categorical error of treating an AI tool as if it possessed the bundle of capabilities associated with a human worker-produces deployment failures whose structural pattern is older than AI. Industrial automation has produced the same shape of disaster for nearly a century, in domains as different as nuclear power, medical radiation therapy, marine navigation, commercial aviation, and financial trading. Each disaster has been followed by a regulatory revision written in the blood of the people the automation harmed. This paper extends the framework developed in Capability Projection (Branham, 2026a) by walking through three historical cases-the 1986 explosion of Chernobyl Unit 4, the 1985-1987 Therac-25 radiation overdoses, and the 1995 grounding of the cruise ship Royal Majesty-and identifying three structural refinements the captain principle requires before it can hold under deployment conditions. Authority without visibility into what the authority does in the pipeline's current state is decorative, as Chernobyl's operators learned. Authority and visibility together, without credibility upward, is also decorative, as the Therac-25 operators learned. Authority, visibility, and credibility together, without transparency about state changes the automation makes silently, is also decorative, as the Royal Majesty's officers learned across thirty-four hours of dead-reckoning drift toward a shoal. The framework treats capability projection as a crosscutting failure layer that interacts with the causes proper to each domain rather than as a totalizing explanation, and admits subtypes-design-side, operator-side, and institutional projection-that the cases below illustrate. The paper closes by sketching a broader taxonomy of captain-displacement failures across industries and by mapping the AI deployment moment onto the recognized failure modes of older automated systems. The rulebook for AI deployment is being written. The substrate is the one all rulebooks have been written in.
Uploads
Quantum Computing by Kase Branham
Yet the effective physical error rate in current demonstrations is approximately 18%, more than double the threshold. This analysis identifies the entire excess as idle decoherence during atom transport: the gate zone operates at 0.03% utilization while qubits decay. The gap between current performance and the fault-tolerance threshold is a factor of 2.8× in circuit time. Batch-parallel gate scheduling-executing multiple non-conflicting atom pairs per transport cycle-closes most of this gap. The analysis includes: (i) a ten-test sensitivity suite confirming robustness to 2× worse gate errors and 37% shorter lifetimes; (ii) a van der Waals crosstalk estimate showing ~0.0005% fidelity impact at the conservative 3× blockade-radius spacing; (iii) a 2D collision-free routing simulation across seven transport strategies, revealing that routing overhead tightens the margin to near the threshold boundary; and (iv) an optimization landscape mapping sixteen candidate techniques against implementation difficulty.
The result is framed as an open question for the community: the physics appears ready for fault tolerance. The remaining barrier appears to be transport routing-a classical multi-agent path-planning problem that adjacent fields have been optimizing for decades.
Biology by Kase Branham
We define twelve named cardiac dances derived from the empirical trajectory patterns observed across Papers I–III, spanning normal physiology, heart failure, arrhythmias, valve pathology, and conduction disease. Each dance is characterized by its trajectory shape (orbit geometry on T²), curvature regime (κ range), concentration profile (Gini range), and the substrate(s) on which it was validated. Three trajectory classes emerge: structured dances (quasi-periodic orbits with identifiable geometric features), degenerate dances (orbits that have collapsed or scattered beyond recognizable structure), and non-dances (trajectories lacking quasi-periodic structure entirely—the geometric signature of immediately life-threatening states).
The vocabulary enables three core diagnostic operations: (1) dance identification—matching a live trajectory against the named library to output a condition likelihood, (2) personal baseline comparison—detecting when a patient’s dance deviates from their own recorded reference pattern rather than a population norm, and (3) emergency detection—recognizing when the trajectory has lost quasi-periodic structure entirely, indicating states that require immediate intervention. Empirical validation on 300 rhythm records demonstrates that dance matching with empirically calibrated prototypes achieves 52.7% overall accuracy across five dances—with the Lock-Step (CHF) at 83–88%, the Stumble (VA) at 100%, and the Waltz (NSR) at 55%—using only nearest-neighbor matching on three geometric features with no training data. A cross-modal coherence test on 3,240 phonocardiograms yields a null result: beat-by-beat curvature correlation between rhythm and acoustic substrates is not diagnostic (r = −0.006, p = 0.82), establishing that the framework’s power resides in aggregate geometric statistics rather than instantaneous cross-substrate coupling. The framework is sensor-agnostic (any device that measures a quasi-periodic signal), theoretically species-agnostic (any heart that pumps periodically), and operator-agnostic (the visualization is self-interpreting). The donut doesn’t care what kind of heart is dancing on it. It only sees the geometry of the dance.
This document establishes three results. First, a proteome-wide geometric baseline: the Gini coefficient of per-residue backbone curvature is the most conserved structural statistic across 124,749 proteins in 34 species (CV = 0.54% across 11 clades), confirmed by 931 experimental crystal structures. Second, a perturbation atlas: 138 matched crystallographic comparisons (same protein, plus or minus perturbation, same lab) reveal four mechanism classes that organize along a continuous geometric spectrum. Third, a three-level hierarchy: the direction of curvature redistribution under perturbation is determined primarily by fold class (p = 0.00015), secondarily by protein scaffold, and only weakly by mutation type — the shape of the instrument matters more than the finger pressing the string.
Across 107,531 annotated beats from the MIT-BIH Arrhythmia Database, torus curvature separates the four major arrhythmia classes (Kruskal-Wallis H = 9,917, p < 10⁻³⁰⁰). Ventricular ectopic beats exhibit median curvature 6.6-fold lower than normal beats (κ = 0.99 vs. 6.55, rank-biserial r = −0.70), reflecting ballistic excursions that bypass autonomic feedback. Extending to 184 records across five PhysioNet databases, two torus-derived parameters—median curvature (κ) and curvature Gini coefficient (G_κ)—create a diagnostic feature space separating congestive heart failure (κ = 32.0), normal sinus rhythm (κ = 10.0), atrial fibrillation (κ = 3.3), and ventricular arrhythmia (κ = 1.2) with pairwise effect sizes |r| > 0.71.
CHF replication across 116 records from four independent cohorts confirms a dose-response gradient: NYHA class 1–3 patients show intermediate curvature (κ = 20.1), positioned between normal (κ = 10.0–16.4) and severe NYHA 3–4 (κ = 32.0), with curvature tracking disease severity (r = 0.61, p = 0.001). Noise robustness testing demonstrates that record-level diagnostic separation survives photoplethysmography (PPG)-realistic timing jitter: at σ = 20 ms, the Normal–Ventricular Arrhythmia effect size remains |r| = 0.95 (p < 10⁻⁵⁰).
Head-to-head comparison with eight standard HRV metrics shows that DFA α1 outperforms the torus for overall disease discrimination (mean |r| = 0.724 vs. 0.544), but torus curvature is the single best metric for CHF severity grading (|r| = 0.605 vs. next-best SD2 at 0.503). An independence test shows that torus curvature is not merely a nonlinear transform of HRV: after regressing out eight standard metrics, residual curvature still separates CHF from Normal (r = −0.336, p = 0.002), with 31.2% of κ variance unexplained by the tested HRV feature set. Combining torus and HRV features improves four-class balanced accuracy from 64.3% to 66.0%. Three additional findings emerge: (1) CHF hearts exhibit flattened circadian curvature amplitude (hourly CV = 0.205 vs. Normal 0.282, r = −0.335, p = 0.003), while AF hearts show exaggerated amplitude; (2) curvature drops 18.8% in the 10 beats preceding ventricular ectopic events (n = 2,519 events, p < 10⁻⁸⁶), constituting a geometric precursor signature; and (3) a combined torus-HRV classifier achieves 66% balanced accuracy across four conditions without machine learning feature engineering.
The framework requires no ECG morphology, no frequency-domain analysis, and no machine learning. It operates on pulse timing alone, making it deployable on any device that measures the interval between heartbeats—including consumer PPG wearables. An interactive real-time visualizer demonstrates that the torus trajectory is visually self-interpreting: an untrained observer distinguished 10 of 13 cardiac conditions by trajectory shape alone, while correctly identifying the method’s blind spots (conditions that alter waveform morphology without changing rhythm). Two numbers from a donut could complement existing wearable algorithms to extend cardiac screening capabilities beyond those currently available in consumer devices.
The energy×spectral cross-feature torus emerges as the strongest acoustic torus pairing (r = −0.338, p = 10⁻⁴¹), measuring how the joint relationship between beat loudness and pitch evolves across consecutive beats. This metric survives all duration controls: partial correlation controlling for recording length yields ρ = −0.130 (p = 10⁻¹³), and beat-count matching strengthens it from r = −0.338 to r = −0.344. Five of five tested torus metrics survive duration matching. A combined Basic+Torus classifier achieves AUC = 0.792, outperforming basic acoustic features alone (AUC = 0.749); adding recording duration contributes only 0.004 additional AUC, confirming that the torus signal is not duration-driven.
The theoretically significant finding is a Gini direction reversal. In rhythm and imaging, healthy dynamics produce high Gini (curvature concentrated at physiological transitions). In acoustics, abnormal hearts produce higher Gini—15 of 16 metrics—because the murmur itself is a concentrating event that dominates each beat cycle. The torus framework defines a diagnostic geometry of cardiac dynamics that adapts to the structure of each substrate, rather than applying a fixed interpretive rule. Three substrates, one mathematical framework, three clinically distinct geometric signatures.
Large Language Models by Kase Branham