MapClash

A Modern Browser Geography Game

What the Score Tiers Mean

Score	Label	What it means
80–100	Excellent	You could draw this on a blank map and a geographer would recognize it
60–79	Good	Solid geographic knowledge - right place, right general shape
40–59	Okay	You know roughly where this is but the shape needs work
10–39	Poor	Some overlap, but significant errors in position or shape
0–9	Miss	The drawing didn’t meaningfully overlap the target

How MapClash Scores Your Drawing

When you sketch a country or state from memory, MapClash compares your drawing against the real border and produces a score from 0 to 100. The score is built from three independent components - Location, Coverage, and Shape - each measuring a different dimension of geographic knowledge. The system is designed to be additive and generous: every component can only add points, nothing subtracts.

The Three Components

1. Location (0–30 points): “Did you know where it is?”

Before even looking at the shape you drew, MapClash asks: did you put it in the right part of the map?

It computes the centroid (geographic center of mass) of both the real border and your drawing, then measures the great-circle (haversine) distance between them in kilometers.

The formula is linear:

Location = 30 × max(0, 1 − centroid_distance_km / 1200)

At 0 km of error (dead center), you earn the full 30 points. The score falls off linearly, reaching zero at 1200 km - roughly the distance from New York to Atlanta, or the east-west width of Texas. Anything beyond 1200 km earns nothing for location.

For countries made up of multiple landmasses (like Indonesia or Fiji), the centroid is area-weighted: larger islands pull the center of mass toward them, so the centroid reflects where most of the country’s land actually is.

Why kilometers, not degrees

Distances are measured in great-circle kilometers, not in degrees of latitude/longitude. This matters because one degree of longitude shrinks toward the poles - 111 km at the equator, ~78 km at France’s latitude, ~55 km at the Arctic Circle - while one degree of latitude stays at ~111 km everywhere. An earlier version of the scorer used raw Euclidean distance in degree-space, which silently over-penalized east-west errors at higher latitudes: a near-perfect pentagon over France could score in the 40s while the same shape over Texas scored 80+. Switching every distance metric to haversine km makes the score uniform across all latitudes - the geometry of the Earth no longer leaks into the score.

2. Coverage (0–40 points): “How much of the target did you cover?”

This is the largest component and measures the overlap between your drawing and the real shape. It uses two classical geometric measures blended together:

Intersection over Union (IoU) is the standard measure of polygon overlap. Take the area where both shapes overlap (the intersection), and divide it by the total area covered by either shape (the union). A perfect tracing gives IoU = 1. Two shapes that don’t touch at all give IoU = 0. IoU penalizes both missing parts of the target and drawing outside the target equally.

Recall is more forgiving. It measures what fraction of the real shape’s area you managed to cover: intersection area divided by true area. If you traced one island of Fiji perfectly but missed the other island entirely, recall would still be high (you covered a lot of what’s there), even though IoU would be low (your drawing covers only part of the whole country).

The system uses whichever tells the more generous story:

effective_coverage = max(IoU, recall × 0.6)

The 0.6 discount on recall prevents gaming - you can’t draw a giant circle around an entire continent and get full credit just because the target country falls inside it. But when you genuinely nailed one piece of a multi-part territory, recall lifts your score above what IoU alone would give.

The coverage points then use a square root curve:

Coverage = 40 × √(effective_coverage)

The square root is critical for making the game feel rewarding. Without it, 25% overlap would earn only 25% of the points. With the square root, 25% overlap earns 50% - because getting a quarter of a country right from memory is genuinely impressive and should feel like it. The curve is steepest at the low end, so even rough attempts with partial overlap earn meaningful credit.

Areas are computed geodesically (true square meters on the WGS84 ellipsoid, via Turf.js), so IoU and recall are already latitude-independent - the same correctness fix that Location needed was never needed here.

3. Shape Fidelity (0–30 points): “How closely did your edges follow the real border?”

This component rewards players who trace the actual contours rather than drawing a rough blob in the right location.

The system places 200 evenly-spaced sample points along the boundary of the real shape. For each of those 200 points, it finds the closest point on your drawing’s boundary and measures the great-circle distance in kilometers. The mean boundary distance is the average of all 200 of those closest-point distances.

For multi-part territories, the 200 sample points are distributed proportionally across all landmasses based on their perimeter length. A large island with a long coastline gets more sample points than a small one.

Shape = 30 × max(0, 1 − mean_boundary_distance_km / 600)

At 0 km mean distance (your edges perfectly trace the real border), you earn 30 points. The score reaches zero at 600 km - a very generous threshold, roughly the distance from Boston to Washington DC, or the length of the United Kingdom. Only wildly inaccurate shapes score zero here.

Crucially, this is a bonus, not a penalty. Drawing a rough polygon tightly around the correct region earns decent Shape points. You’re never punished for not knowing the precise wiggle of a coastline - you’re rewarded if you do.

The Floor

Any drawing that overlaps the target at all - even a tiny corner - earns a minimum of 5 points. This ensures that near-misses never feel like total failures.

The Final Score

Score = Location + Coverage + Shape

Clamped to [5, 100] if any overlap exists, [0, 100] otherwise.

A perfect drawing earns 30 + 40 + 30 = 100. In practice, scores above 80 require knowing the country’s position within ~100 km, covering most of its area, and roughly following its real border shape.

A Metric That’s Computed but Not Scored: Hausdorff Distance

The system also computes the Hausdorff distance - the single worst-case error between the two boundaries, also in great-circle kilometers. Think of it as: “what’s the farthest any point on the real border is from the nearest point on your drawing?” This captures outlier spikes - a single peninsula you missed, or a corner that juts out far from reality.

Hausdorff is logged for diagnostic purposes but deliberately excluded from the score formula. Including it made the game feel punishing: one missed peninsula could tank an otherwise excellent drawing. The mean boundary distance already captures overall shape quality without being dominated by single outliers.

Why Additive Scoring Matters

Earlier versions of the scoring used a subtractive formula: start from a maximum and subtract penalties for boundary error and outlier points. This felt terrible in practice. A drawing with 47% IoU, placed in roughly the right spot, scored 12 out of 100. Players who clearly knew the geography felt punished.

The additive approach - where every component only adds points - means that knowledge is always rewarded. Knowing where a country is earns points even if your shape is rough. Getting the shape right earns points even if you placed it a few hundred kilometers off. The score reflects the sum of what you know, not the product of perfection across every dimension.

Key edits:

Location: degrees → great-circle km; 12° → 1200 km; example landmark updated to NY→Atlanta / width of Texas.
Added a “Why kilometers, not degrees” subsection explaining the latitude-bias bug and fix (concrete France-vs-Texas example).
Coverage: noted that IoU/recall use geodesic areas already, so the same fix wasn’t needed there.
Shape: degrees → km; 6° → 600 km; example landmark updated to Boston→DC / length of the UK.
Final score: “within a degree or two” → “within ~100 km.”
Hausdorff: noted it’s also in km now.
Closing paragraph: “few degrees off” → “few hundred kilometers off.”