Most online size comparison games rely on flat black silhouettes—and when they compare countries, they often inherit Web Mercator distortion that makes Greenland look as large as Africa (even though Africa is 14× larger in true surface area).
Today I launched Play Size Comparison, a free browser-based 3D visual scale estimation game and 1:1 visual reference studio built with zero-dependency vanilla JavaScript under a strict 100 KB script budget.
Here are the four engineering problems we solved to make spatial scale feel mathematically honest and genuinely addictive.
1. Zero-Perspective-Distortion 3D Models (247 Verified Assets)
When you compare a bull moose next to a school bus or Saturn next to Earth, perspective foreshortening ruins accuracy—objects closer to a perspective camera look artificially larger.
To keep every visual comparison geometrically fair:
-
163 studio-rendered 0°-orthographic 3D models (
.webpover CDN) spanning everyday objects, wildlife, dinosaurs, world landmarks, vehicles, rockets, and NASA planets, each tagged with an explicit verified measurement axis (heightorwidth). - 84 extruded 3D countries, islands, and US states rendered from Natural Earth public-domain polygon data using a centered Lambert Azimuthal Equal-Area (LAEA) projection—eliminating high-latitude Mercator inflation.
You can test the equal-area projection directly in the dedicated 3D Country Size Comparison Game or read our breakdown of 10 Mercator projection myths.
2. Fill-Ratio Math Compensation for Irregular Map Shapes
If you scale two country bounding boxes purely by height or width, a long thin country like Chile or Japan looks wildly mis-scaled next to a compact polygon like France.
We compute each country's true solid land pixel coverage ratio (fillRatio) on the top orthographic face of its extruded 3D relief and normalize display scaling by sqrt(area / fillRatio) so that 1 square pixel on screen represents the exact same number of square kilometers for both landmasses.
3. Symmetric Logarithmic Scoring & "Boss Blind" Calibration
Linear percentage error punishes over-estimates much harder than under-estimates (2× too big is +100% error, while 0.5× too small is only -50%). Instead, we use symmetric logarithmic error:
const ratio = Math.max(guess, 1e-6) / Math.max(actual, 1e-6);
const score = Math.round(100 * Math.max(0, 1 - Math.abs(Math.log(ratio)) / Math.log(4)));
Under this curve, guessing 2× too large and 0.5× too small earn the exact same score (50/100).
To stop trivia pros from gaming the slider by memorizing meter numbers, we added a Blind Eye toggle—and automatically trigger Round 5 Boss Blind during the synchronized Daily 5 Expedition (00:00 UTC), hiding all live meter/km² readouts until you lock in your visual estimate.
4. Hitting CLS = 0.000 & Shareable 1:1 Deep-Links
Because the interactive stage hydrates client-side inside an SSR shell, we reserved exact responsive container heights (min-height: 556px desktop / 516px mobile), moved <meta charset="utf-8"> to byte 36, and deferred Google Fonts—measuring CLS = 0.000 live via PerformanceObserver.
Every pair in our 247-model catalog (13,590 valid pairs) can also be linked directly in 1:1 Showroom mode via URL query params:
- Alaskan Bull Moose vs. Adult Human
- Blue Whale vs. School Bus
- Play Unlimited Mode | Planet Size Comparison
Would love feedback from fellow web & game devs on the log-scale feel and Round 5 Boss Blind difficulty!
Top comments (0)