ANAGRAM UNIVERSE / LINGUISTIC OBSERVATORY
Discover what your letters can become.
A phrase is a finite universe. Explore its exact transformations, meaningful fragments and unexpected linguistic coincidences — with every letter accounted for.
01 / PRIVATE LABORATORY
A finite alphabet. An open question.
02 / PUBLIC OBSERVATORY
THE ANAGRAM UNIVERSE
Explore the linguistic structures people have chosen to contribute anonymously.
Most Intelligible
How this ranking works
Maximum local intelligibility: 48% normalized word frequency, 32% approximate POS compatibility, 20% lexical-category coherence, minus a long-phrase penalty.
All scores are deterministic local heuristics. No AI inference contributes. Counts represent accepted contributed searches, not people. Results are search-bounded. Small daily timeline cells are suppressed.
Public corpus activation is on hold
The shared database has not been changed. Contributions, public rankings and public phrase pages cannot be exercised until the production-edit hold is cleared and database checks pass. No private searches or invented popularity figures appear here. The private solver above is independent of this hold.
Withdraw a contribution
Use the private receipt supplied at contribution time. No account identity is used to locate contributions.
03 / METHOD & LIMITS
Nothing invented. Everything inspectable.
The conservation law
C_R(x) ≤ C_S(x); U(x) = C_S(x) − C_R(x) ≥ 0.
Generation uses a bounded beam over residual character inventories, not raw permutations. A separate validator recomputes each candidate inventory. Exact candidates consume every letter; subsets leave an explicit residual. Unicode NFKC and language-aware lowercase define identity. Apostrophes are removed; ordinary punctuation becomes spaces; diacritics remain distinct. English and Spanish Latin-letter dictionaries are supported.
Local linguistic proxies, not AI judgments
Intelligibility = 0.35 × normalized Zipf frequency + 0.50 × local grammar score + 0.15 × category cohesion, minus length and structural penalties. Grammar combines transition compatibility (55%) and English/Spanish phrase templates (45%), with penalties for repetition, unfinished function words and excessive function-word density. These are authored, inspectable rules—not measured n-gram probabilities. Single-word grammar uses a 0.9 baseline; unmatched multiword templates receive a 0.25 template baseline. Short unrecognized tokens and long noun chains are penalized. Unrecognized one/two-letter tokens require the abbreviations option. Tags and categories remain approximate. No trained language model, embedding model, factual inference or biographical inference is used.
Humor combines a small comic lexicon (35%), explicit opposing word pairs (25%), source divergence (15%) and intelligibility (25%). Irony weights opposing pairs (70%) and divergent low-cohesion text (30%). Poetry combines repeated initials (35%), repeated endings (35%) and cohesion (30%). Surprise combines rarity (55%) and divergence (45%). Memorability combines intelligibility (40%), initials (30%) and a preference for three-word phrases (30%). These are disclosed computational proxies, not human judgments.
Explicit computational ceilings
At most 64 source letters, eight words, 120 normal or 180 deep results. Normal mode: beam 100, lexical breadth 240, 24,000 expansions, 2.2-second engine deadline. Deep mode: beam 180, breadth 400, 65,000 expansions, 4.5-second engine deadline. The browser worker has a 12-second outer watchdog including dictionary loading. Hardware and dictionary coverage affect discovered results. No exhaustive-search promise.
Lexical provenance & limitations
The adapted English and Spanish lexicons derive from wordfreq 3.1.1 historical frequency data (through approximately 2021), licensed CC BY-SA 4.0. Frequencies are measured corpus estimates; grammar, humor and meaning scores are derived heuristics. Word categories are incomplete. This is not a professional dictionary, grammar checker, exhaustive lexicon or safety classifier.
METERED AI INFERENCE CALLS BY THIS ENGINE: 0 · No inference adapter or paid fallback exists. This statement is not a claim that hosting or development has no cost.