Featured
Spot the Lie
Honest chart or misleading? Judge each one, then watch the data get redrawn straight. How many tricks can you catch?
Browse by theme
Calibrated
how good are your error bars? · 2 min
Correlation, Confounded
control for the real driver
GenAI Cost at Scale
what does it cost?
Agent ROI
automate the work
A/B Test Planner
sample & duration
Forecasts Lead
vs. lagging metrics
Cherry-Picking
timeframes lie
Prompt Playground
tokens · cost · clarity
Daylight vs. Mood
pick a city
Meeting Load
by seniority
PGA Putting
drag the distance
Strokes Gained
build a shot
Where Strokes Are Won
drive for show?
The Decoy Effect
irrational by design
Start here
Calibrated
Question 1/10 · Caught 0
Give a range you're 90% sure contains the answer. If your confidence is honest, nine of ten ranges will catch it. Almost nobody's do.
The question
Under the hood
A hit means the truth landed inside your range. Calibration means your hit rate matches your stated confidence: ranges offered at 90% should catch the answer 90% of the time. In the classic studies (Alpert & Raiffa onward), most people's 90% intervals catch roughly half.
Your calibration
0/10 caught at "90%"
Forecasting & decisions
Forecasts Lead, Metrics Lag
A signal turns mid-year. Watch a forecast catch the turn early while a trailing metric reacts late. The blue band is the forecast’s 90% range: it grows with the horizon, and with volatility. Point estimates hide this; honest forecasts don’t.
Takeaway: the more you steer by lagging metrics, the later you see the turn. Forecasts buy you decision time.
Behind the viz: the metric is a trailing moving average of the true signal; the forecast is a leading indicator. Companion to my essay “Metrics Lag, Forecasts Lead.”
Cherry-Picking Timeframes
The same flat, noisy series — three years of it. Slide a window (or hit a button) and watch the “trend” flip from up to down.
Takeaway: the same data can “prove” growth or decline — always ask what window someone chose, and why.
Behind the viz: a fixed series with no real trend; the line is a least-squares fit over the selected window. Companion to my essay “Tell Me Data Lies.”
A/B Test Planner
Enter a baseline conversion rate, the lift you want to detect, and your traffic — get the sample size and how long the test must run.
Takeaway: small lifts on low base rates need a lot of traffic — know your runtime before you start, not after.
Behind the viz: two-proportion sample size at 95% confidence and 80% power, split evenly across two arms.
Correlation, Confounded
Marketing spend and revenue move together, convincingly. Slide to control for the season driving both, and watch the relationship thin out.
Correlation: 0.00
Under the hood
The slider partials out the confounder: both variables are regressed on the seasonal driver and progressively replaced by their residuals. What remains at 100% is the partial correlation — the relationship after the season has explained what it can.
Takeaway: before acting on a correlation, ask what moves both lines. Controlling for it is one slider here; in real work it is the whole job.
AI economics
GenAI Cost at Scale
Pick a model and a workload, and see what generative AI actually costs per day, month, and year.
Takeaway: at scale, model choice and output length dominate the bill — a cheaper model or tighter outputs can cut cost by an order of magnitude.
Behind the viz: cost = requests × (input_tokens × input_price + output_tokens × output_price), using published per-million-token list prices (approximate, mid-2025).
Autonomous Agent ROI
How much does an AI agent return when it resolves work autonomously? Set the workload and see.
Takeaway: agent ROI is driven by volume × resolution rate — automating even a slice of high-frequency work compounds fast.
Behind the viz: hours saved = tasks × auto-rate × minutes ÷ 60 × 30.4; net = labor saved − agent cost; ROI = net ÷ agent cost.
Prompt Engineering Playground
Type a prompt and see an estimated token count, API cost, and a clarity score computed from the actual text.
Takeaway: small wording changes move cost and clarity more than people expect — specificity is cheap leverage.
Behind the viz: tokens use the ~4-chars-per-token heuristic blended with word count; cost uses published GPT-4o input pricing ($2.50 / 1M tokens); clarity is a transparent rubric.
Behavioral & everyday data
The Decoy Effect
A famous experiment (Dan Ariely, often retold by Rory Sutherland). Make your choice in each scenario, then see what adding one “useless” option does.
Scenario 1 — pick a subscription:
Scenario 2 — same offer, with one option added:
Takeaway: a “useless” option you'd never choose can still reshape the choice you do make. Context creates value — design the menu, not just the products.
Behind the viz: based on Dan Ariely's Economist-subscription experiment (“Predictably Irrational”). Adding the print-only decoy pushed combo sign-ups from 32% to 84%.
Daylight vs. Mood
How daylight hours track with a seasonal mood index across the year.
Takeaway: the higher the latitude, the wilder the swing — daylight, not rainfall, is the stronger signal for seasonal mood.
Behind the viz: monthly daylight hours by latitude; the mood index is a modeled value that tracks trailing daylight (normalized 0–100).
Meeting Load Through the Week
Meeting density by day and hour. Toggle the seniority level.
Takeaway: the calendar fills as you climb — mid-week afternoons run hottest, which is why Thursdays feel draining.
Behind the viz: a day-by-hour heatmap of meeting density, scaled by a seniority factor.
Golf analytics
PGA Tour Putting: Make % by Distance
How quickly a made putt becomes a coin flip as distance grows.
Takeaway: the 50/50 line lands right around eight feet — beyond that, pros are managing risk, not expecting makes.
Behind the viz: Tour make-percentage by putt distance, drawn in a minimal, Tufte-inspired style.
Strokes Gained: build a shot
Every spot on the course has an expected number of strokes to hole out. Strokes Gained = (expected before) − (expected after) − 1. Build a shot and see what it was worth.
Ball starts
Ball ends
Takeaway: a “good” shot is simply one that beats the expectation for where you started — that's all Strokes Gained measures.
Behind the viz: expected-strokes baselines approximated from Mark Broadie's PGA Tour data (“Every Shot Counts”). Green distances are in feet; all others in yards.
Where Strokes Are Won
“Drive for show, putt for dough”? Pick a handicap and see where the strokes versus a tour pro actually go.
Takeaway: the long game — off the tee plus approach — is where most strokes are lost; putting is the smallest slice. Broadie's most famous, most counterintuitive finding.
Behind the viz: strokes lost per round vs. a tour pro by category; figures approximated from Mark Broadie's “Every Shot Counts.”