The Mathematics of Random Letters

Drawing a random letter from 26 candidates is a discrete uniform distribution over a finite alphabet. Each letter holds exactly 1/26 ≈ 3.85% probability. Each draw is independent: the letter "A" appearing once has zero influence on whether "A" appears next. This independence is what separates random text from natural language, where letter sequences follow deep statistical patterns shaped by vocabulary, grammar, and phonetics.

Random Letters versus Natural Language

In English text, letter frequencies are profoundly unequal. Samuel Morse quantified this in the 1830s while designing telegraph codes: he counted letters in a printing house's type cases and found that E appears roughly 13% of the time, T about 9%, and Z barely 0.07%. His frequency ranking (ETAOIN SHRDLU) became legendary among typesetters and cryptographers. A random letter generator ignores all of this. Every letter appears with identical 3.85% probability, producing sequences that look nothing like natural language. That contrast is precisely what makes this tool useful for testing whether a system treats all inputs equally.

The Coupon Collector's Surprise

How many random letters must you draw before you have seen all 26 at least once? Most people guess around 30. The mathematical answer is approximately 100. Specifically, the expected number is 26 × H(26) ≈ 26 × 3.85 ≈ 100.2, where H(26) is the 26th harmonic number (the sum 1 + 1/2 + 1/3 + … + 1/26). The first few letters arrive quickly. The last one or two take agonizingly long. With 20 unique letters collected, you still expect roughly 26 × (1/6 + 1/5 + 1/4 + 1/3 + 1/2 + 1) ≈ 64 more draws to complete the set. The coverage chart above tracks this journey in real time.

This phenomenon, known as the coupon collector's problem, appears throughout computer science, ecology (species sampling), and quality assurance (how many items to test before covering all defect types). Philippe Flajolet and Robert Sedgewick provided the modern analytical framework in their landmark text on analytic combinatorics.

Cryptographic Fairness

Each letter is selected by crypto.getRandomValues(), the Web Cryptography API. A random byte is generated and mapped to the 26-letter range using rejection sampling: values outside the largest exact multiple of 26 (which is 234 within the 0-255 byte range) are discarded and redrawn. This eliminates the modulo bias that would otherwise make A through D slightly more likely. The result is a mathematically perfect uniform distribution over the alphabet, generated entirely in your browser.

In the Classroom

Have each student visit /letter/100 and generate 100 random letters. Ask them to tally the frequency of each letter and compare with the expected 3.85 per letter. Some students will see 8 or more of one letter and zero of another. The class discussion: is the generator biased? The histogram provides the answer visually as more letters accumulate toward equal heights.

For a deeper exercise in cryptanalysis, inspired by the 9th-century Arab scholar Al-Kindi who pioneered frequency analysis, have students compare their random sequences against a paragraph of English text. The random sequence has flat frequency; the English paragraph shows dramatic peaks at E, T, A. This contrast introduces the concept that randomness has structure, and that structure can be measured. The tool requires no accounts and stores no student data.

Private by Architecture

Every letter is generated inside your browser. The server delivers the page; your device creates the randomness. Your history lives in localStorage on your machine. Sharing the URL shares the tool configuration only. The recipient draws their own independent letters from their own device entropy.