By Nelson Ray • ncray.com
From Galileo's optic tube and Einstein's gedankenexperiments to the data floods of modern genomics: Freeman Dyson's great duality, why theorists overvalue concepts, and whether there are other ways science transforms.
Leverage and influence, collinearity, colliders and bad controls, sandwich standard errors, multiple imputation, the winner's curse, AIC versus BIC, calibration, covariate shift, partial pooling, autocorrelation and sloppy models. Each comes with a lab that shows where the method breaks.
Twelve ideas behind turning data into models you can trust, from plotting first, robust summaries and missing data to cross-validation, regularization, base rates, clustering and spurious trends. Each one has a simulation you can run in the browser.
The power of two choices, competitive paging, the random 3-SAT threshold, HyperLogLog, channel capacity, zero-knowledge proofs, FLP and Raft elections, quorums, tail latency, double descent, differential privacy and timing side channels. Each comes with a lab that shows where the guarantee stops.
Twelve ideas behind computing, from bits, Turing machines and compression to algorithms, caches, error-correcting codes, public-key cryptography and machine learning. Each one has a simulation you can run in the browser.
Kaplan–Meier and censoring, ROC curves, regression to the mean, immortal time bias, interim analyses, publication bias, saturable kinetics, receptor theory, superspreading, Guyton's circulation, radiotherapy fractionation, and log-kill. Each comes with a lab that shows where the method breaks.
Twelve ideas that turned medicine into a science, from germ theory and vaccines to fair trials, screening and dosing. Each one has a simulation you can run in the browser.
Phase space and symplectic integrators, least action, normal modes, chaos, dispersion, uncertainty, the Ising transition, Brownian motion, tunnelling, Bell's theorem, Lorentz boosts, and Mercury's perihelion. Each comes with a numerical lab.
Twelve ideas that explain how the universe works, from Newton's laws and energy to relativity and quantum mechanics. Each one has a simulation you can play with.
The welfare theorems, Arrow, Cournot, replicator dynamics, revenue equivalence, deferred acceptance, information cascades, dynamic programming, the Taylor principle, bank runs, staggered difference-in-differences, and Pareto tails. Each comes with a lab that pushes on the key assumption.
Twelve ideas that explain prices, trade, cooperation and growth, from opportunity cost to the multiplier. Each one has a simulation you can run in the browser.
Reference counting and the cycle collector, the C3 MRO, dict probing, bytecode, the specializing interpreter, circular imports, float cancellation, list over-allocation, NumPy strides, GIL races, the asyncio event loop, and variance. Each lab reproduces the CPython mechanism.
Twelve ideas behind programming in Python, from names and objects to concurrency. Each one has a hands-on lab you can run in the browser.
Gibbs, Baire and Weierstrass, Cantor sets, monodromy and the quintic, Burnside, zeta zeros and the primes, winding numbers, Gauss–Bonnet, Feigenbaum universality, random-graph thresholds, concentration of measure, and undecidability. Each comes with a lab that probes the theorem's edges.
Twelve ideas that built mathematics, from zero and the primes to calculus and the sizes of infinity. Each one has a live experiment you can play with.
Fisher information, Wilks at the boundary, empirical processes, Stein shrinkage, MCMC scaling, FDR, post-selection inference, bootstrap failure, minimax smoothing, the lasso, Marchenko–Pastur, and double robustness. Each comes with a live lab that probes the edge of the theorem.
Twelve ideas that turned uncertainty into knowledge, from the law of large numbers to the bootstrap. Each one has a live simulation you can run in the browser.
From Vovk’s algorithmic learning to Waymo’s motion tubes, Amazon’s inventory supply chains, and LLM hallucination gating: how modern machine learning abandoned the illusion of point estimates for finite-sample mathematical certainty.
From a North Carolina trucker waiting in Hoboken to Harrison’s marine watch, Kaiser’s Liberty ships, and Pasteur’s fermenting vats: the invariant mechanics of how adjacent schemas shatter incumbent blind spots, and why true breakthroughs are never born inside the guild.
From Boltzmann’s suicide and the martyrdom of Semmelweis to Azoulay’s empirical proof of star gatekeeping: why human brains refuse to unlearn false dogmas, how academic gerontocracy strangles discovery, and what happens when the gatekeepers stop dying.
From System R and Codd's betrayed algebra to NULL traps, the object-relational impedance mismatch, dialect balkanization, and modern engines: why the world's most criticized query language conquered computing.
From Codd's combinatorial hypercubes and the Volcano iterator trap to ClickHouse, Druid, and DuckDB: how cache lines, SIMD vector registers, and the memory wall dictate the absolute physical speed limits of analytical engines.
Why the Cox partial likelihood resists standard machine learning, fifty years of mathematical evasions and approximations, and how to build a production-grade survival package in Python from first principles.
Why feedback loops blow up without persistent excitation, how unmodeled dynamics destroyed the 1980s adaptive control revolution, and why any system that reliably improves its own substrate must be damped, bounded, and slow.
In school, master equations are unveiled like pristine marble statues. In reality, every iconic formula is a patched battleground of desperate kludges, pruned scaffolding, and unhealed singularities waiting for their next historical revision.
Deconstructing four years of real-world immunology and population registries: why a saponin-protein nanoparticle quieted the mRNA orthodoxy, how Berkson’s paradox invented an apocalyptic reinfection spiral, and what the biophysics of immune memory reveal.
Why hiring a nanny costs a California professional three dollars for every dollar the carer takes home, how the tax code treats raising children as an illicit luxury rather than human capital, and the fiscal architecture that drove American fertility off a cliff.
Deconstructing the Marc Tessier-Lavigne resignation, the Sylvain Lesné forensic scandal, the true genetic bedrock of the Amyloid Cascade Hypothesis, and why the brutal biophysics of clinical translation—not doctored Western blots—stalled the search for an Alzheimer's cure.
Why conditioning on common effects invents false associations out of thin air, how Berkson’s fallacy tricks medicine, AI, and social science, and the mathematical blueprints for surviving selected data.
Why the most uncompromising attempts to create "useless" knowledge inevitably become the bedrock of global technology—from prime-number cryptography and Boolean circuits to Riemannian GPS navigation, medical CT scans, and oncology chemotherapy.
Why ninety-year-old monetary history is the wrong threat model for Bitcoin—and why the true danger is neither a midnight raid nor confined to states with a pedigree of expropriation. Deconstructing the 1933 gold reserve bypass, analyzing modern interventions in Cyprus, Canada, India, and the G7, and mapping the real vector of institutional capture.
From smoothed analysis and industrial SAT to the Continuum Hypothesis, quantum spectral gaps, and spin glasses: why worst-case bounds dissolve in natural data—and the exact physical, mathematical, and strategic boundaries where nature becomes adversarial.
Why ordinary structural steel delivers only 0.1% of its theoretical atomic shear strength, why severe cold-drawn pearlitic wire reaches within a factor of three of Frenkel's cohesive limit, and how single-crystal turbine blades survive hundreds of degrees above their melting point against a hard thermodynamic ceiling.
From democratic ladder to credential treadmill: the economic history of the college wage premium, the Bennett-Baumol inflation engine, the Dale-Krueger causation puzzle, California's test-blind experiment, and the resurgence of the skilled trades.
Settler mortality, weak instruments, the Albouy critique, and why the Nobel Committee crowned an econometric landmark it conceded was flawed.
Why a random walker is doomed to wander into infinity in three dimensions, why the sample mean ceases to be optimal in the exact same dimension, and how a single integral over the boundary of Euclidean space unifies mathematical physics and decision theory.
Why the popular gospel of "third- and fourth-order thinking" is an epistemic illusion in open, adaptive worlds—and the strict physical, informational, and structural conditions under which higher-order attribution actually holds.
From $O(1)$ dictionaries and cryptographic truth to HyperLogLog, Count-Min sketches, and locality-sensitive geometry: why the deliberate, non-injective destruction of information is computing's most generative architectural idea.
From rate caps, Clean Air mandates, and feed-in tariffs to shadow banking, intermodal freight, and rare earth bottlenecks: how complex systems route around 2nd-order failure to achieve intent (+X) through decentralized bypass—only to crystallize new systemic fragilities (−X) at scale.
From Bradford Hill to collider stratification, immortal time bias, vibration of effects, Western blot splice lines, and the Carlisle method: an adversarial field guide for scrutinizing empirical papers, detecting causal illusions, and catching outright scientific fraud.
Why extreme outcomes are almost never what they seem. Unifying Michael Mauboussin's success equation with Maurice Tweedie's empirical Bayes de-noiser across sports, gambling, and the hierarchy of investment titans (Jim Simons, Warren Buffett, Ken Griffin, Peter Lynch, Bill Miller) to the 20-year SPIVA decimation of active mutual funds and the power-law geometry of venture capital.
From the green ball with hair to quantum nonlocality, Banach-Tarski, Stein's shrinkage, Braess's paradox, and the PCP theorem: how physics, mathematics, statistics, economics, and computer science construct, shatter, and rebuild human intuition.
Why treat statistical significance as an unalterable act of God when it is simply a scalar loss function over your analytical pipeline? Drawing on Gelman's Garden of Forking Paths, Ioannidis's epistemic production functions, and the vibration of effects: a tongue-in-cheek engineering manual for guaranteeing $p < 0.05$ through rigorous, data-contingent coordinate descent.
From the nuclear primers of Los Alamos, Maxwell's radio waves, and Dirac's antimatter to the steam engine, smallpox vaccines, high-temperature superconductors, and the modern non-convex miracle of deep learning: an epistemic anatomy of why civilization's greatest technologies bifurcate between deductive theory and empirical engineering—and how high-dimensional geometry explains why deep networks actually work.
From Nyquist-Shannon and Abbe's diffraction to NP-completeness, comparison sorting, and Heisenberg's limit: why mathematical impossibility proofs are conditional statements—and how engineers repeatedly shatter theoretical ceilings by attacking their unexamined axioms.
Why population-level guidelines systematically under-screen the wealthy while biohacker hype dangerously over-tests them—and how Bayesian risk stratification, advanced diagnostics, and physiological reserves rewrite the actuarial table. An actuarial and clinical guide to the largest preventable causes of death across life and the precise protocols to mitigate them.
Why classical Gang of Four patterns warp under Python's data model—and the structural paradigms that replaced them. From compile-time protocols, decorator registries, descriptors, and generator-based inversion of control in FastAPI, SQLAlchemy, and Celery, to the leak-proof Estimator contract in Scikit-Learn, PyTorch composite hook trees, Polars lazy expression DAGs, and hardware-aware strategy dispatch in vLLM.
Why the flattest geometry in mathematics dominates physical law, statistical inference, and deep learning. From Taylor expansions swallowed by ambient noise floors and quadratic basis expansions to the Universal Approximation Theorem as piecewise-affine origami, backpropagation as associative adjoints, the Neural Tangent Kernel, and the thermodynamic lock-in of systolic silicon.
From Bacon's Idols and the Vienna Circle to Popperian refutation, Kuhnian revolutions, and David Deutsch's hard-to-vary truth: the historical struggle to understand how human minds generate objective knowledge about an unobserved reality.
From paralyzed infants and the peanut allergy catastrophe to COURAGE, ORBITA, and the war over the open medical chart: why the clinical establishment conflates mechanistic plausibility with empirical reality—and why paternalism defends error to preserve its authority.
From the Shannon bound and compressed sensing to the stubborn thermodynamic reality of Carnot engines, materials physics, and galactic algorithms: an analytical anatomy of when theory and practice converge—and why they so often drift apart.
From Socrates’ war on the written alphabet and Victorian “Bicycle Face” to the Luddite gallows, the 1674 petition against coffee, and the automation of the mind: why every epoch-defining technology is first condemned as an assault on the body, the soul, and the social order.
A comprehensive primer on modern causal experimentation in online tech companies: from statistical heuristics, ratio metric linearization, and CUPED variance reduction to Wager and Athey's causal forests, doubly robust machine learning, two-sided marketplace cannibalization, city-scale crossover switchbacks, and synthetic difference-in-differences.
The comforting myth is that incompleteness is just an ouroboros—a linguistic trick of self-reference. The terrifying reality is the $\Pi^0_1$ asymmetry: if a checkable conjecture is unprovable, it is guaranteed to be true. From Hilbert's 10th problem and Matiyasevich's trap to Lagarias's arithmetic Riemann Hypothesis, 748-state Busy Beavers, and undecidable quantum phase transitions.
Why nature's meiotic lottery is an illusion without parental genomes. A statistician's forensic takedown of Hardy-Weinberg false comfort, genetic nurture, and polygenic confounding—from the clinical triumphs of targeted PCSK9 cis-MR to the structural collapse of educational attainment IVs.
Why stats students are taught that confidence intervals quantify uncertainty when they only measure sampling luck. From Angrist's charter school lotteries to Manski bounds and Pearl's transportability theorems: what happens when the baseline counterfactual drifts, institutions evolve, and the sample size approaches infinity.
Second-place catch-up dynamics, industrial statecraft, and the geopolitics of commoditizing your complement. Why a regime built on information control is aggressively bankrolling open models—and why the endgame of AI belongs to whoever owns the metal.
In a world where lightweight distillation handles ninety percent of everyday tasks, humanity already generates multiple full-length libraries per person each day. Why human reading speed is no longer the ceiling—and what computing history reveals about where the FLOPs will go.
Why finding a valid instrumental variable is the hardest detective work in empirical science. A taxonomy of causal identification from flawed weather shocks to gold-standard encouragement designs.
From SpaceX Starship's metallurgy pivots to the $50 billion semiconductor lithography bet: understanding path dependency, yield chasms, and physics bottlenecks.
Mapping evidential standards across human inquiry—from kernel-checked mathematics to quantum electrodynamics, GRADE clinical medicine, reflexive economics, and aesthetic coherence.
Things I run myself, served from subdomains.
Spaced-repetition (FSRS) flashcards on statistics, data science, Python, and math.
A browser space ARPG with Diablo II-style loot and skills.
Explore IMDb ratings for series and mini-series, with episode rating trends. Login required.
Small web apps hosted on Vercel.
Adaptive competence mapping with Bayesian prerequisite graphs.
An adversarial audit of empirical papers for collider bias, immortal time, and p-hacking.
Adaptive FAA private pilot written-exam prep and Cessna 172 ground school.
A gentle phonics and early-reading tutor.
A calm, picture-first app for daily routines and feelings.