— NOTES

Writing.

Notes on forecasting, engineering, and the products I build.

Aug 5, 2026 · ~15 min

Competition is for losers — full notes on Thiel's Stanford talk

Detailed notes on Peter Thiel's CS183B lecture. The core claim: value created and value captured are independent variables, and competition is the machine that grinds the second one to zero. Airlines vs. Google, why monopolists and competitors lie about their markets in opposite directions, small-market entry, the four monopoly traits, last mover, why science doesn't pay, and the mimetic trap. With timestamps, written for future review.

Jul 2, 2026 · 12 min

Does the American heat move the 2026 World Cup?

A weather × performance study on the first 81 matches. Humid heat visibly drains players and reshapes tactics — but the body tax is absorbed before it reaches the scoreboard. Seven interactive charts, all on real data.

Jun 26, 2026 · ~5 min

Train the body to train the mind

I half-remember reading that exercise — especially team sports like football — is good for the prefrontal cortex: you train judgment and attention while you move. The solid part (aerobic exercise improves executive function) has thick evidence; the 'ball games beat running' part is thinner, but the direction has research behind it. So learning isn't only a sit-down activity — move, ideally inside a game that forces you to think fast.

May 17, 2026 · 7 min

Citing the canon is the load-bearing UX choice

A feng-shui app that footnotes 《阳宅十书》 reads differently from one that doesn't. Notes on building FengShui Blueprint — why every score points to a classical source, and why that decision did more work than the rules engine itself.

May 17, 2026 · 10 min

CME's compute futures: not a perpetual — yet

On May 12, CME and Silicon Data announced the first compute futures, cash-settled against Silicon Data's daily GPU rental indices. The right mental model is electricity or weather derivatives, not Bitcoin perpetuals. But a compute perpetual is coming on a crypto venue within twelve months — and that's the more interesting product.

May 17, 2026 · 9 min

Marketing as three measurement problems: SEO, GEO, CRM

SEO is hygiene now. GEO — whether LLMs cite you — is the next three years of brand discovery. CRM is the data plane that survives platform churn. Notes on why advanced marketing in 2026 is one identity pipeline with three serving surfaces, not three vendor stacks.

May 12, 2026 · 9 min

Non-custodial escrow as a marketplace primitive: design notes from TaskMarket

A 4-state Solidity contract on Base, a binary arbiter call, and no third destination address. How TaskMarket settles in USDC without ever touching user funds — and what each invariant prevents.

May 8, 2026 · 8 min

83 ties was enough to kill TSFMs as my baseline

Backtest notes from UEFA Oracle: a 3-model Time Series Foundation Model ensemble lost to a well-tuned Elo + xG + injury-weighted model on 83 historical knockout ties. Why the data regime mattered more than the model.

May 4, 2026 · 8 min

Migrating a community platform without migrating the database

lakebbs went from Vue 2 + Express + Sequelize to Next.js 16 + React 19 + Drizzle ORM — without changing the MySQL 5.7 schema. Why locking the database during a stack rewrite is the boring, correct decision.

Apr 28, 2026 · 7 min

Workers AI as a triage filter, not a generator

The right place to spend a Workers AI call is at the front of the pipeline as a scoring filter, not at the end as a generator. Cost, latency, and quality reasons — with concrete patterns from Agentic Inbox and IRCC Monitor.

Apr 14, 2026 · 8 min

Forecasting the new Champions League

Notes on UEFA Oracle — a hybrid TSFM + Elo + Poisson model for the 2025-26 UCL, benchmarked against Polymarket.

Apr 5, 2026 · 7 min

Time Series Foundation Models on equities: the hostile case

Why US equities are the stress test TSFMs don't want, and what an honest zero-shot benchmark has to do differently.

Mar 30, 2026 · 5 min

Fermi estimation as a scoping discipline

Before you open a notebook, get the answer within an order of magnitude. If the physics ballpark is wrong, the pipeline can't save you.

Feb 25, 2026 · 6 min

What the Rust compiler refused to let me avoid

Rewriting a trading bot from Python to Rust isn't a syntax change. It's a forced audit of every decision the dynamic version let you postpone.

Jan 22, 2026 · 5 min

Shop-floor latency is a product decision

Two-second lag kills trust in a production console. It's not a performance knob — it's the shape of the product.