Rarity Methodology

Every rarity score on RNGdle.net is calculated from a complete analysis of all 1,000,001 supported numbers (0 to 1,000,000). Nothing is estimated or sampled.

Supported range and representation

The supported range is 0 to 1,000,000 inclusive. Numbers are analyzed by their natural decimal representation with no leading zeros, so 7 is treated as a one-digit number, not "000007". Digit-based traits (length, palindrome, sequences) follow directly from that representation.

Step 1 — Extract features

Each number is reduced to a set of boolean traits plus two distribution features (digit sum and count of distinct digits). Traits are organized into groups so correlated properties can be de-duplicated later.

Step 2 — Exact probability and information

For each trait f, we count how many of the 1,000,001 numbers match it. The probability is count / 1,000,001 and its information content is -log2(probability). Rarer traits carry more bits and contribute more to the score. Below are the exact counts used by the current algorithm.

Trait Group Matches Probability
Single-digit number range 10 1 in 100,000
Two-digit number range 90 1 in 11,111
Three-digit number range 900 1 in 1,111
Four-digit number range 9,000 1 in 111
All digits identical repetition 45 1 in 22,222
Five or more of one digit repetition 487 1 in 2,053
Quadruple digit repetition 11,340 1.13%
Triple digit repetition 138,024 13.80%
Double pair repetition 223,803 22.38%
Repeated digit repetition 457,731 45.77%
Adjacent digit run repetition 35,884 3.59%
Only two distinct digits repetition 4,294 1 in 233
Palindrome symmetry 1,989 1 in 503
Sequential ascending sequence 22 1 in 45,455
Sequential descending sequence 26 1 in 38,462
Ascending segment sequence 28,826 2.88%
Descending segment sequence 29,815 2.98%
ABAB pattern sequence 243 1 in 4,115
ABCABC pattern sequence 891 1 in 1,122
Alternating parity sequence 34,875 3.49%
Zigzag digits sequence 136,425 13.64%
Harshad number arithmetic 95,428 9.54%
Happy number arithmetic 143,071 14.31%
Prime factorization 78,498 7.85%
Semiprime factorization 210,035 21.00%
Perfect square factorization 1,001 1 in 999
Perfect cube factorization 101 1 in 9,901
Power of two factorization 20 1 in 50,000
Fibonacci number factorization 30 1 in 33,333
Triangular number factorization 1,414 1 in 707
Factorial factorization 9 1 in 111,111
Highly divisible factorization 2,809 1 in 356
Binary palindrome representation 1,998 1 in 501
All ones in binary representation 18 1 in 55,556
Hex repdigit representation 59 1 in 16,949

Distribution features

Digit sum and distinct-digit count are scored from their exact distributions across the supported range. Daily game trait links such as “Digit sum 23” point here because these values vary by number rather than belonging to one fixed boolean-trait row above.

Step 3 — Avoid double counting

Some traits are highly correlated — "all digits identical", "quadruple digit" and "repeated digit" overlap heavily. Traits are grouped into range, repetition, symmetry, sequence, arithmetic, factorization, representation. Within each group only the strongest trait scores in full; the rest are attenuated by decreasing weights (1, 0.35, 0.15, 0.07, 0.03) so a family of related traits cannot stack indefinitely.

Step 4 — Score to percentile

Group contributions are summed into total bits and scaled to an integer rarity score. Because every number was scored during the build, we know the exact score distribution and can map any score to a precise percentile. The current build has a maximum score of 1,677 across 807 distinct score values.

Tier Percentile Minimum score
Singular Top 0.1% 603
Extreme Top 1% 399
Exceptional Top 5% 303
Rare Top 10% 266
Distinct Top 25% 218
Notable Top 50% 175

Step 5 — Versioning

The scoring configuration is versioned. Every result records the algorithm version it was produced with, so a shared number keeps its meaning even after the algorithm is updated.

Current algorithm version: 1.1.0 · Built 2026-09-06

A known limitation — length

Only ten of the 1,000,001 supported numbers have a single digit, so "is a single-digit number" carries about as much information as a strong pattern, and the score treats it that way. Short numbers also have very few distinct digits and very small digit sums, which are rare in their own right. The result is that short numbers dominate any raw ranking: sixty-five of the hundred highest-scoring numbers in the range have three digits or fewer.

This is measured correctly and is not adjusted away — but it is worth being clear about what it measures. Length is not a property of a number, it is a property of where the range was cut off. Widen the range to ten million and 3 would score higher still, while 3,003 would not move. Everything else the score counts — palindromes, factorisations, digit patterns — is intrinsic to the number and does not depend on the boundary.

Individual scores are left untouched, so a short number's page reports exactly what the algorithm computed. Where a ranking is presented, though, the boards start at four digits, because otherwise they would rank the boundary rather than the numbers — see the leaderboard and the examples on the pattern pages.

Cultural features

Meme numbers, year-like numbers and date-like numbers are interesting but subjective, so they are intentionally kept out of the mathematical rarity score. If added later, they will be shown separately and clearly labelled.