Atticus Li — Better Decisions

VWO can report a Bayesian probability, a frequentist significance level, a sequential recommendation, a fixed-horizon result, or lower-rigor dynamic evidence.

Booking.com says it runs approximately 1,000 parallel experiments at any given moment.

Apple’s Product Page Optimization is a particularly clear public example of Bayesian experimentation built into a self-service product.

A Google experimentation paper from 2010 describes power calculations, A/A calibration, clustered data, counterfactual logging, multiplicity, launch ramps, expert review, and a knowledge repository.

Netflix says its teams run thousands of A/B tests each year, yet only a small percentage of ideas become product launches.

For most of my life, “building software” meant finding someone technical enough to translate an idea into code.

What Intelligence Analysts Know About Evidence That Growth Teams Don't TL;DR Intelligence analysts almost never get a single conclusive source.

What Forecasting Tournaments Say About Trusting Your Gut TL;DR Philip Tetlock's Good Judgment Project — originally sponsored by IARPA, the U.S. intelligence community's research arm — ran a multi-year forecasting tournament that measured, rather than theo…

People new to Claude Code tend to go looking for the right prompt — some ideal phrasing, a magic combination of words that reliably produces a good result.

This is the question that stalls the most people before they've written a single prompt: should I learn to code first, so I actually understand what's happening?

"Vibe coding" is an easy phrase to dismiss on the name alone — it sounds like a shortcut, or a toy version of the real thing.

Most people's mental model for an AI tool is a chat window: you ask, it answers, and the answer lives on the screen until you copy it somewhere yourself.

Any spend limit that depends on process memory has a blind spot on infrastructure that doesn't persist state between runs — and CI is exactly that infrastructure.

Prompt caching gets pitched as close to free money: pay a small premium once, then get the same content back at a fraction of the price on every later call.

Any tool that injects environment variables into the sessions and subprocesses it runs — which most agentic coding tools do, deliberately — creates a namespace a project's own configuration can collide with.

Scheduled jobs that call an LLM API on a timer have a blind spot most teams don't think to check: metered, pay-per-call billing has no ceiling on how many times an unattended job fires while nobody's watching.

Making the cheapest of three plan tiers the visual hero of a pricing page is supposed to be the safest move in conversion optimization — one real experiment from my portfolio shows it can quietly commoditize the whole page and pull conversion down instead…

Borrow a scarcity tactic from ticket-selling sites, drop it into the checkout page for an essential recurring service, and the evidence says it can cost you completions instead of winning them.

A double-digit-looking loss on a stepper redesign turned out to be neither a loss nor a win — it was noise wearing a costume, and the only professional response was to say so.

A homepage built to serve five kinds of visitors was, in practice, being used by one.

A pattern that already won once is not evidence it will win again somewhere else in the funnel — it's a better-odds hypothesis, and the discipline is in testing that transfer instead of assuming it.

The most-cited finding in behavioral pricing research predicted a clean win here, and a well-powered experiment against real customers returned a shrug.

A pricing page can win by making the price harder to notice, not easier to justify.

Reordering three prices beat a full pricing-page redesign, and the margin wasn't close.

Most experimentation programs don't die from a shortage of ideas — they die from testing four of them at once, calling it speed, and discovering that the result cannot identify which idea helped or hurt.

Deleting a few sentences of explanatory copy from a mobile modal produced a double-digit conversion lift — proof that on some pages, the explanation is the confusion.

Two navigation systems were fighting for the same thumb on the same mobile screen, and the fix that won wasn't a clever new idea — it was a decade-old, publicly available piece of UX research that nobody had actually tested against this specific site (Exp…

Most website personalization examples assume that more relevance creates a better experience.

The best lead magnet examples do not win because they are PDFs, quizzes, calculators, or templates.

Color psychology in marketing is usually explained as a dictionary: blue means trust, red means urgency, and green means growth.

Evaluate an A/B testing case study by reconstructing the experiment before judging the headline.

A useful navigation A/B test should compare visibility, accessibility, and removal as separate mechanisms.

Pricing page optimization should make the next decision easier without hiding information buyers need to choose confidently.

Landing page navigation should match the decision the visitor has already made.

Good checkout page design removes choices that compete with payment while preserving the information and controls needed to finish safely.

The useful lesson in most A/B testing examples is not the winning design.

Sequential Testing and the SPRT: How to Stop a Test Early Without Cheating Meta description: Peeking at a fixed-sample A/B test inflates false positives.

Triangulation Over Isolation: Building Confidence from Weak, Convergent Signals Meta description: A single underpowered test never proves anything alone.

The Meta-Analysis Your Experimentation Program Is Missing Meta description: Most programs audit individual tests, almost none audit the program itself.

Why Most "Wins" Don't Replicate: The Winner's Curse, Applied to Growth Teams Meta description: The winner's curse means shipped A/B test wins systematically overstate their true effect.

The Confidence Tier Model: How to Decide When Your Data Isn't Enough Meta description: Most testing programs are built for traffic they don't have.

Decide What Counts as a Win Before You Test Meta description: Medicine proved that picking your primary metric after seeing the data is a structural bias.

How to Tell If a Growth Hire Understands Risk Meta description: A great win story tells you almost nothing about judgment.

I ran a routine cleanup pass on this site's infrastructure recently — the kind of housekeeping that usually turns up nothing worth mentioning.

Subscription or Pay-Per-Token?

Every "examples of behavioral economics" list on the internet teaches you the biases.

How to Resolve a Git Merge Conflict Without Guessing What the Other Side Was Trying to Do A conflict-free merge is not the same thing as a correct one.

Why Trusting an AI Assistant's Memory Is the Wrong Default (And What to Check Instead) Confidence isn't the same thing as current.

How to Get Your AI Coding Assistant to Catch Its Own Mistakes Before You Do A draft that reads as finished has only passed the test that matters least.

Behavioral Economics Examples Roughly a third of tested ideas at Microsoft improve the target metric.