Lab Notebook№ 72

Working notes, half-formed thoughts, and experiments in progress.

A scratch pad, not a publication. Much of this is written with—and by—AI. I'm less author than curator. Don't attribute these thoughts to me; I'm not sure I'd claim them myself.

  • AI-assisted throughout
  • Not my employer's views (or mine, really)
  • Errors are likely; corrections welcome
  • Epistemic status: shaky
Index
Tags:
January2026
JAN242026

The Ledger of Selves

Personal identity as a stochastic process: θt=θt−1+drift+jumps\theta_t = \theta_{t-1} + \text{drift} + \text{jumps}.
philosophyidentitystochastic-processespsychologychange
2 min read
JAN232026

Generalist Epistemics: How to Know Things Across Domains

The intellectual toolkit for productive breadth
epistemologyforecastingdecision-makingmeta-cognitiontetlock
14 min read
JAN222026

From Engineer to Operator

Identity in the age of AI: when depth gets commoditized, breadth becomes the moat
aicareeridentitygeneralismmeta-skills
11 min read
JAN212026

Cumulative Advantage and Matthew Effects

When early luck compounds: P(wint+1∣wint)>P(wint+1)P(\text{win}_{t+1} | \text{win}_t) > P(\text{win}_{t+1})
networksstatisticssociologyinequalitypath-dependence
4 min read
JAN202026

Careers as Stochastic Processes

Modeling advantage, skill, and opportunity as evolving random variables: dAt=μtAt dt+σtAt dWtdA_t = \mu_t A_t\, dt + \sigma_t A_t\, dW_t
stochastic processescareersgeometric brownian motionjump diffusionlife modeling
9 min read
JAN182026

Canto II: Volatility Targeting & Position Sizing

From Kelly's f∗=μ/σ2f^* = \mu/\sigma^2 to adaptive positions: wt=σtarget/σtw_t = \sigma_{\text{target}}/\sigma_t.
algorithmic tradingcantomomentumposition sizingkelly criterionvolatility targeting
3 min read
JAN162026

Calibration: Are Your Probabilities Honest?

Reliability diagrams, Brier scores, and the sharpness-calibration tradeoff
bayesianstatisticscalibrationpredictiondecision-makingml
4 min read
JAN142026

The James-Stein Paradox

Why your intuition is wrong: θ^JS\hat{\theta}^{\text{JS}} dominates yˉ\bar{y} for k≥3k \geq 3
bayesianstatisticsshrinkageestimationparadoxmle
2 min read
JAN122026

Multi-Agent Coordination as Distributed Optimization

Consensus, markets, and the coordination tax: T∥=Twork/k+Tsync(k)T_{\parallel} = T_{\text{work}}/k + T_{\text{sync}}(k)
agentsdistributed-systemsoptimizationconsensusmarketscoordination
4 min read
JAN102026

Tool Calling Is Just Function Composition

With uncertainty: (g∘f)(x)=g(f(x))(g \circ f)(x) = g(f(x)) becomes E[g(Y)∣Y∼f(x)]\mathbb{E}[g(Y) \mid Y \sim f(x)]
agentsfunctional programmingmonadsmcpoodauncertaintycompositionunixrust
3 min read
JAN72026

Shrinkage Everywhere

From James-Stein to ridge to your manager ratings: θ^=wyˉ+(1−w)μ\hat{\theta} = w\bar{y} + (1-w)\mu
bayesianstatisticsshrinkageridgehierarchicalestimation
2 min read
JAN12026

January 2026: Minimal Viable System

One keystone habit, two systems, three contingency plans.
goalsmonthlysystemssleepRTO
5 min read
◈Himalayan Expeditions Data Analysis7 parts
DEC302025

A Dataset for Hierarchical Thinking

882 expeditions, 116 peaks, and the nested structure of risk.
datatidytuesdayhierarchical
5 min read
DEC302025

Exploratory Data Analysis

What the data tells us before we start modeling.
EDAvisualizationR
8 min read
DEC302025

Feature Engineering and Shrinkage

From raw data to model-ready features, with Bayesian thinking.
feature-engineeringbayesianshrinkage
11 min read
DEC302025

Bayesian Hierarchical Models

Partial pooling, shrinkage, and why peaks aren't exchangeable.
bayesianbrmshierarchical
8 min read
DEC302025

Probabilistic Graphical Models

DAGs, Bayesian networks, and the latent structure of success.
PGMDAGSEM
8 min read
DEC302025

Causal Inference

From association to causation: propensity scores, doubly robust estimation, and causal forests.
causalpropensityCATE
10 min read
DEC302025

Synthesis: Three Frameworks, One Answer

What happens when Bayesian, graphical, and causal methods converge.
synthesistriangulationmethodology
8 min read
December2025
DEC272025

Retrospective 2025

First year of quarterly goals: E[completion]↑10-20%\mathbb{E}[\text{completion}] \uparrow 10\text{-}20\%, Var(sleep)→∞\text{Var}(\text{sleep}) \to \infty.
goalsquarterlypowerliftingchessagents
10 min read
DEC202025

Type Parity from Database to Browser

FastAPI + SQLModel → OpenAPI → Hey API → Axios, TanStack Query, and Zod.
fastapisqlmodelopenapihey-apiaxiostanstack-queryzodtypegen
5 min read
DEC72025

Advent of Code 2025

Racket notes and solutions
advent of coderacketdial rotationspattern findingperiodicitybattery banksgrid problemsrange setscolumn transposition
3 min read
DEC32025

Fund Data Passports & The NAV Registry

From file rooms to Registry+Passport⇒one publish, many reads\text{Registry} + \text{Passport} \Rightarrow \text{one publish, many reads}.
Private EquityData RoomsFund DataSystems SketchInfrastructure
7 min read
November2025
NOV162025

Stage Tilts to State Space: GP Skill the Bayesian Way

From yf,m∼N(μm,σ2)y_{f,m}\sim\mathcal{N}(\mu_m,\sigma^2) to stage offsets, Student-tt tails, μs\mu_s-anchored managers, γvintage\gamma_{\text{vintage}} cycles, and a μm,t\mu_{m,t} random walk—all inside one workflow.
private equitybayesianworkflowpymchierarchicalposterior predictivesimulation
2 min read
NOV122025

Intelligence Increase as Control Under Uncertainty

Why Intelligence⁡≈arg⁡max⁡πE[∑tγt−1R(st,at)]\operatorname{Intelligence} \approx \arg\max_\pi \mathbb{E}[\sum_t \gamma^{t-1} R(s_t, a_t)] under tool-augmented constraints.
aiagentscontrolphilosophysmilepompdmcpplotlysystems
2 min read
NOV122025

Life Extension as Hazard Shaping Under Constraints

Steering λ(t)\lambda(t) so S(t)=exp⁡ ⁣(−∫0tλ(u) du)S(t) = \exp\!\left(-\int_0^t \lambda(u)\,du\right) bends toward longer, healthier time.
longevitybiostatisticssurvival-analysiscontrolsmile
10 min read
NOV92025

Borrowing Predictive Strength, In Practice

Walking the data flow from p(θ∣y)p(\theta \mid y) draws to JSON moments m=E[g(θ)∣y]m=\mathbb{E}[g(\theta)\mid y].
private equitybayesianpymchierarchicalsqlmodelsynthetic datastress testing
1 min read
NOV32025

Posterior Multiples for Pricing, Leverage, and Covenants

Letting M~i=wiMi+(1−wi)μ\tilde{M}_i = w_i M_i + (1-w_i)\mu shrink comps before the covenant debate.
private equityvaluationbayesianhierarchicalcompsexitsmixtures
1 min read
NOV22025

Deal Sourcing as a Contextual Bandit

Budgeted Thompson Sampling with at=arg⁡max⁡xθt⊤xa_t = \arg\max_x \theta_t^\top x under ∑tct≤B\sum_t c_t \le B.
private equityoriginationbayesianbanditsthompson-samplingstrategydecision-theory
9 min read
October2025
OCT262025

We Buy Distributions, Not Deals

Choosing exits by sampling p(r∣D)p(r \mid \mathcal{D}) so E[r]−λ⋅VaRα\mathbb{E}[r]-\lambda \cdot \mathrm{VaR}_\alpha drives the sell/hold call.
private equitybayesiandecision-theoryriskvaluation
2 min read
OCT122025

Borrowing Predictive Strength

Sharing strength because θdeal∼N(θfund,τd2)\theta_{deal} \sim \mathcal{N}(\theta_{fund}, \tau_d^2) and θfund∼N(θmgr,τf2)\theta_{fund} \sim \mathcal{N}(\theta_{mgr}, \tau_f^2).
private equitybayesianhierarchicalpredictionpooling
1 min read
OCT52025

Bayesian NAV Updating w/ Kalman Filters

Updating quarterly marks via x^t∣t=x^t∣t−1+Kt(zt−Hx^t∣t−1)\hat{x}_{t\mid t} = \hat{x}_{t\mid t-1} + K_t(z_t - H\hat{x}_{t\mid t-1}).
private equitybayesiannowcastingvaluationkalmanhierarchical
2 min read
OCT12025

Secondaries as Temporal Arbitrage

Spotting trades whenever D=−1VdVdrD = -\tfrac{1}{V}\tfrac{dV}{dr} and pacing curves fall out of sync.
private equitysecondariesliquiditydurationstochastic-discount-factorportfolio-math
2 min read
September2025
SEP282025

Market Beta for Private Market Valuation

Showing how β=Cov(rPE,rpub)Var(rpub)\beta = \tfrac{\mathrm{Cov}(r_{PE}, r_{pub})}{\mathrm{Var}(r_{pub})} mutates once marks are smoothed kk quarters.
Private EquityRiskBetaInteractiveRegressionBottom-up
2 min read
SEP242025

Explore vs Exploit in the Age of AI

Balancing exploration vs exploitation by tuning ϵ\epsilon so E[R∣πϵ]\mathbb{E}[R \mid \pi_\epsilon] stays above ship-it.
StudyPracticeEngineeringAIProductivityStrategy
1 min read
SEP212025

Risk-Free Rates in PE

Building WACC inputs with rf=yUST−πer_f = y_{UST} - \pi_e anchoring ke=rf+β⋅ERPk_e = r_f + \beta \cdot \mathrm{ERP}.
Private EquityRisk-Free RateWACCDCF SurfaceHamadaLBO Monte CarloInteractiveSimulation
1 min read
SEP142025

Hawkes Processes for Portfolio Rollups

Modeling capital-call waves with λ(t)=μ+∑ti<tαe−β(t−ti)\lambda(t) = \mu + \sum_{t_i<t} \alpha e^{-\beta(t-t_i)} triggers.
Private EquityPoint ProcessesHawkesInteractiveSimulation
1 min read
SEP72025

The Denominator Effect Playbook

Watching policy bands break when wt=NAVtNAVt+∑jLj,tw_t = \tfrac{NAV_t}{NAV_t + \sum_j L_{j,t}} spikes and forces secondaries.
private equitysecondariesstrategydenominator-effectliquidityportfolio-math
1 min read
August2025
AUG312025

Geometry of Fund Multiples vs Rates

When IRR\mathrm{IRR} solves ∑tCFt(1+r)−t=0\sum_t CF_t (1+r)^{-t}=0 but TVPI=NAV+DistPaid In\mathrm{TVPI}=\tfrac{NAV+\mathrm{Dist}}{\mathrm{Paid\ In}} tells another story.
Private EquityIRRTVPIInteractiveSimulationCopulas
1 min read
AUG242025

PME, Direct Alpha, & Replication

Contrasting PME=∑CFt(1+rb)T−t∑Callst(1+rb)T−t\mathrm{PME} = \tfrac{\sum \mathrm{CF}_t (1+r_b)^{T-t}}{\sum \mathrm{Calls}_t (1+r_b)^{T-t}} with α=ln⁡(PME)T\alpha = \tfrac{\ln(\mathrm{PME})}{T}.
Private EquityBenchmarkingInteractiveKaplan-SchoarDirect AlphaReplication & PME+
2 min read
December2024
DEC272024

Retrospective 2024

2024 wrap where ∇plan2025\nabla \text{plan}_{2025} is set by residual et=goalt−shipte_t = \text{goal}_t - \text{ship}_t.
goals
7 min read
November2024
NOV82024

Advent of Code 2020 21-25

Resolving allergens with I=⋂kIkI = \bigcap_k \mathcal{I}_k and CRT shuttles using t≡ai(modni)t \equiv a_i \pmod{n_i}.
advent of coderacketalgorithms
1 min read
October2024
OCT102024

Advent of Code 2020 16-20

Parsing tickets via x∈⋂jRjx \in \bigcap_j \mathcal{R}_j and tiling monsters so f(x)=Rx+tf(x)=R x + t lines up.
advent of coderacketalgorithms
2 min read
September2024
SEP152024

Advent of Code 2020 11-15

Sim seating automata until ∣St+1−St∣=0\lvert S_{t+1}-S_t\rvert=0 then docking with v=∑ibi2iv = \sum_i b_i 2^i bitmasks.
advent of coderacketalgorithms
2 min read
SEP132024

Advent of Code 2020 6-10

Tracking group answers via ∣∪iSi∣\lvert \cup_i S_i \rvert and grid paths with ft+1=ft+Δf_{t+1}=f_t+\Delta.
advent of coderacketalgorithms
2 min read
SEP112024

Advent of Code 2020 1-5

Racket runs for a+b=2020a+b=2020, seat IDs 8r+c8r+c, and expense reports as (n2)\binom{n}{2} hunts.
advent of coderacketalgorithms
2 min read
SEP102024

Canto I: Momentum Basics

Comparing baselines to rt(k)=∑i=1kwiRt−ir_t^{(k)} = \sum_{i=1}^{k} w_i R_{t-i} trend signals.
algorithmic tradingcantomomentumbacktesting
9 min read
August2024
AUG132024

Gomper

Layering color fields where c(x,y)=∑iwiηi(x,y)c(x,y)=\sum_i w_i \eta_i(x,y) from Poisson-disk seeds.
Generative ArtPoisson Disk SamplingSimplex Noise
1 min read
AUG32024

Of a Momentous Fixed Point

A plotting toy where mt+1=βmt+(1−β)rtm_{t+1} = \beta m_t + (1-\beta) r_t keeps pulsing.
Generative ArtStonksMomentumBuy!Sell!
1 min read
AUG12024

Bebop Rift

Ribbon improvisations where c(t)=∑ksin⁡(2πfkt+ϕk)c(t)=\sum_k \sin(2\pi f_k t + \phi_k) drives the palette.
generative artharmonicswave
1 min read
AUG12024

Magic Pig Detective

A noisy sleuth built from xt+1=Axt+εtx_{t+1} = A x_t + \varepsilon_t pink-noise textures.
generative artMelvins
1 min read
AUG12024

An Imaginary Country

Drifting landscapes with h(x,y)=∑kαkPerlin⁡(2kx,2ky)h(x,y) = \sum_k \alpha_k \operatorname{Perlin}(2^k x, 2^k y) fog.
generative artnoiseTim Hecker
1 min read
July2024
JUL252024

Rothko Generator

Stacking shader layers c(z)=∫01w(t)exp⁡(−∥z−μt∥2/σt2) dtc(z)=\int_0^1 w(t)\exp(-\|z-\mu_t\|^2/\sigma_t^2)\,dt.
generative artMark RothkoSympathetic Magicinteractive
1 min read
JUL252024

The 3-Fiber Kullback-Liebler Poisson Experiment

Weaving Poisson arrivals λxyz\lambda_{xyz} with DKL(P∥Q)D_{KL}(P\|Q) texture maps.
generative art3DKullback-Leiber DivergencePoisson
1 min read
JUL182024

Playing With Chaos

Tuning x˙=σ(y−x)\dot{x} = \sigma(y-x)-style systems until attractors fill R3\mathbb{R}^3.
generative artchaos theory
1 min read
June2024
JUN292024

Blog V2

Shipping MDX so post(x)=Markdown(x)+∫React(x) dx\text{post}(x) = \text{Markdown}(x) + \int \text{React}(x)\,dx.
new softwareoverview
1 min read
JUN282024

Black-Scholes-Merton Model

Walking from dSt=μStdt+σStdWtdS_t = \mu S_t dt + \sigma S_t dW_t to C0=S0N(d1)−Ke−rTN(d2)C_0 = S_0 N(d_1) - K e^{-rT} N(d_2).
OptionsInteractive
1 min read
December2023
DEC272023

Retrospective 2023

Reviewing 2023 via ΔOKR=∑i(actuali−targeti)\Delta \mathrm{OKR} = \sum_i (\text{actual}_i - \text{target}_i) and shrinking σfocus2\sigma^2_{\text{focus}}.
goalsgpt4summitracketpowerlifting
3 min read
May2023
MAY152023

From PDE to Price: Solving Black-Scholes and Delta Hedging

Solve LV=0\mathcal{L}V=0, go risk-neutral, then hedge with Δ=∂V/∂S\Delta = \partial V/\partial S so dΠt=rΠtdtd\Pi_t = r\Pi_t dt.
Black-Scholes-Mertonrisk-neutral measuredelta hedgingGreeksEuropean option
2 min read
April2023
APR262023

Jammin' with GPT4

Pair sessions where d flowdt\tfrac{d\,\text{flow}}{dt} tracks E[tokens∣prompt]\mathbb{E}[\text{tokens} \mid \text{prompt}].
GPT
6 min read
APR262023

Outpainting with Dall-e 2

Extending canvases by sampling xt+1∼N(μt,Σt)x_{t+1} \sim \mathcal{N}(\mu_t,\Sigma_t) from p(pixels∣prompt)p(\text{pixels}\mid\text{prompt}).
OutpaintingDall-e 2GPT-4ArtPhotographyPlagiarismPainting
4 min read
January2023
JAN242023

The Black-Scholes Partial Differential Equation

Deriving ∂tV+12σ2S2∂SSV+rS∂SV−rV=0\partial_t V + \tfrac{1}{2}\sigma^2 S^2 \partial_{SS} V + rS \partial_S V - rV = 0 and why it prices hedges.
Black-Schole-Mertonpartial differential equationEuropean derivative
2 min read
December2022
DEC272022

Retrospective 2022

A scorecard where P2022=∑iwi⋅ΔhabitiP_{2022} = \sum_i w_i \cdot \Delta \text{habit}_i.
goalskanbanbookssynth
9 min read
November2022
NOV192022

Itô's Lemma

Applying df=∂tf dt+∂xf dXt+12∂xxf σ2dtdf = \partial_t f\,dt + \partial_x f\,dX_t + \tfrac{1}{2}\partial_{xx} f\,\sigma^2 dt to finish the derivation.
Itô's LemmaItô CalculusGeometric Brownian MotionTaylor expansionlog-normal
3 min read
October2022
OCT212022

Stochastic Differential Equations & Geometric Brownian Motion

Tracking how ΔXt∼N(0,Δt)\Delta X_t \sim \mathcal{N}(0,\Delta t) becomes dSt=μStdt+σStdWtdS_t = \mu S_t dt + \sigma S_t dW_t.
stochastic differential equationsgeometric brownian motionjump diffusionsquare root diffusion
1 min read
September2022
SEP272022

Brownian Motion as a Symmetric Random Walk Limit

Watching Sn/n⇒WtS_n/\sqrt{n} \Rightarrow W_t as Δt→0\Delta t \to 0.
random walkbrownian motionpythonmathematical financeoptions pricingCLTInteractive
2 min read
SEP112022

Kaprekar's Routine

Iterating γt+1=sort⁡↓(γt)−sort⁡↑(γt)\gamma_{t+1} = \operatorname{sort}_\downarrow(\gamma_t) - \operatorname{sort}_\uparrow(\gamma_t) until 6174.
algorithmnumber theoryuniqueness
3 min read
July2022
JUL122022

Montana

Field notes where exposure E∝Af2E \propto \tfrac{A}{f^2} keeps geysers and big skies in range.
montanayellowstonegeysershot springs
1 min read
June2022
JUN262022

Cyborg Lisps

When host stacks embed λ\lambda-calculus so code≡data\text{code} \equiv \text{data} across the cyborg membrane.
lispembedded programming languagesmacrosmetacircular evaluator
8 min read
JUN222022

A Spoonful of Metaprogramming

Binding tidy-eval spells to λ\lambda-expressions so f(x;θ)f(x;\theta) pipelines synthesize themselves.
Rmetaprogrammingtidy eval
2 min read
JUN202022

Searching For Chaos

Scanning coefficient vectors a\mathbf{a} until Lyapunov λmax⁡(a)>0\lambda_{\max}(\mathbf{a}) > 0 lights up.
pythongenerative artchaos theorydifferential equationsconvex hullclustering
5 min read
JUN12022

O/blog/itory First Post

Setting t=0t=0 with blogt+1=blogt+ε\mathbf{blog}_{t+1} = \mathbf{blog}_t + \varepsilon experiments.
introduction
1 min read