Train recognition.
Build Qur’an anchors.

Exact Figure asks, “Do you know the count?” Closest Figure asks a slightly different question: “How good is your estimate, and can you improve it using feedback?” Because of that, a near miss should not be treated the same way as a random guess.

1. Measure the distance

Let a be the actual number of Ayahs and g be the player's guess. The raw distance is:

2. Scale that distance to the Surah

Ten Ayahs away means very different things on a 286-Ayah Surah and a 3-Ayah Surah. Closest Figure therefore uses a tolerance window based on the Surah's own size:

Normally the tolerance is half of the true count. The minimum of three protects very short Surahs from becoming disproportionately harsh.

3. Convert the miss into normalized error

This converts “how many Ayahs off?” into a scale that can be compared across Surahs of very different lengths.

4. Apply the scoring curve

An exact guess is always worth 100. If e ≥ 1, the guess has moved outside the scoring window and earns zero. The exponent 1.5 makes the curve stricter than a straight line, so genuinely close guesses remain valuable while vague guesses lose value more quickly.

5. Higher and lower create convergence

When a guess is below the answer, you’ll see Higher. When it is above the answer, it responds with Lower. The exact distance remains hidden so the feedback narrows the search without solving the round.

6. The best guess carries the round

Every try receives its own proximity score. The strongest score is kept, then five points are removed for each additional try:

Best-of scoring preserves the strongest evidence of knowledge while the additional-try cost still rewards reaching the answer efficiently.

7. Worked example

Suppose the actual count is 78 Ayahs:

The best single-guess score is 92. Three tries means two additional-try penalties, so the current round value is 82 points.

8. Edge cases

The scoring curve is only one part of the system. Closest Figure also needs predictable behavior around unusual input, skipping, exhausted tries, and repeated requests.

9. Implementation

The scoring function is deliberately pure: numbers go in and a score comes out. That keeps it straightforward to test, reuse, and compare between client previews and authoritative server scoring.

const FALLOFF_EXPONENT = 1.5;
const TOLERANCE_RATIO = 0.5;
const MIN_TOLERANCE = 3;
export const TRY_PENALTY = 5;

export function closestFigureScore(
  guess: number,
  actual: number
) {
  const distance = Math.abs(guess - actual);

  if (distance === 0) return 100;

  const tolerance = Math.max(
    MIN_TOLERANCE,
    actual * TOLERANCE_RATIO
  );

  const error = distance / tolerance;

  if (error >= 1) return 0;

  return Math.round(
    100 *
      Math.pow(
        1 - error,
        FALLOFF_EXPONENT
      )
  );
}