CoolFace
Datasetpublic

ryanjosephkamp/ars-magna-greatest-hits

Ars Magna Greatest Hits The funniest and most apt anagrams of people, companies, products, titles, places and phrases, found by Ars Magna and kept by hand. Every row is a real anagram: the words use exactly the input's letters, checked against a pinned revision of English OpenList (368bf0e4460461c985fca8bde49e4062d56c1516), and every word is in the tier the row names. Accented letters fold to their base letter, so Beyoncé has three e's. Nothing typed is ever replaced by… See the full description on the dataset page: https://huggingface.co/datasets/ryanjosephkamp/ars-magna-greatest-hits.

sourceHugging Facemitupdated 5d agoView on Hugging Face
0likes393downloads
Dataset Card

Ars Magna Greatest Hits

The funniest and most apt anagrams of people, companies, products, titles, places and phrases, found by Ars Magna and kept by hand.

Every row is a real anagram: the words use exactly the input's letters, checked against a pinned revision of English OpenList (368bf0e4460461c985fca8bde49e4062d56c1516), and every word is in the tier the row names. Accented letters fold to their base letter, so Beyoncé has three e's. Nothing typed is ever replaced by something else (the literal rule): a number is never read as its name, and a symbol never as a letter or a word. A number or a symbol of the set @ $ & % + # (and ! or ? inside a word) is left out of the letters, and the row says so (reading).

Subsets

ConfigWhat it holdsRows
allEvery published hit.288
peoplePeople: performers, athletes, politicians, public figures.68
companiesCompanies and organisations.25
productsProducts, software, devices, brands.32
titlesFilms, shows, albums, books, games.66
placesCities, countries, regions, landmarks.33
phrasesPhrases and the classics.64

A category with no rows yet is listed here but not offered as a config until it has one.

python
from datasets import load_dataset
people = load_dataset("ryanjosephkamp/ars-magna-greatest-hits", "people")

Fields

FieldMeaning
idThe input's folded letters, its category, and its words as a multiset. Stable.
inputThe text as a person would write it.
categoryOne of people, companies, products, titles, places, phrases.
wordsThe anagram's words, in reading order.
displayThe words joined with spaces.
lettersThe sorted letters the input and the anagram share.
readingHow each number and symbol of the input stood when the anagram was made, as a list of item and reading in the input's order: drop (left out) for every item; null for an input without any, and for a hit from before 2026-09-21, whose digits and symbols were dropped without a record. The literal phase adds self and the leet readings.
prefilter_scoreThe model-free score that put it in front of a judge (ordering, word frequency, length).
judgeOne entry per judge, with its model, rubric version, rationale and date. Rubric v2 scores relation to the input (1–5, the score that decides) and reads (1–3), with tone and subjects labels and a justification. Rubric v1 scored aptness, grammar and memorability (1–5 each) and a total.
justificationOne plain sentence explaining why the anagram fits its input, for a reader who does not know the reference; null when there is none yet.
aboutOne factual sentence saying what the input is, from Wikidata's English description or written for the dataset; null when there is none yet.
wikipediaThe input's English Wikipedia article; null when it has none or none is known.
sensesThe sense a word reads in, in this anagram, as a list of word and sense in reading order: one sentence each, only for a word whose first dictionary sense would not explain the reading or that has no definition; null when no word needs one.
shelfgreatest (Greatest Hits, chosen by hand), interesting (a clear link), or stretch (a loose one).
submitterWho found it, when it was submitted rather than mined; null for a mined hit.
addedThe date it entered the list.
dictionaryThe English OpenList revision it was verified against.
tierThe smallest dictionary tier that contains every word: common, standard, full, or extended for a word the site added on top of English OpenList.
tagsLabels: classic for the ones everyone knows, tone:… and subject:… from the judge, greatest-candidate where the judge scored relation 5.
statusaccepted or featured. Proposed and retired rows are not published.

How rows get here

  1. 1.Inputs are gathered without a model: hand-picked, or taken from the day's most-viewed Wikipedia articles and classified through Wikidata.
  2. 2.The Ars Magna engine enumerates every anagram.
  3. 3.A deterministic prefilter keeps phrases made of everyday words, in their best reading order.
  4. 4.A language model scores each phrase's relation to its input against a fixed rubric (v2).
  5. 5.The pipeline places each phrase on a shelf by that score.
  6. 6.A person approves by merging. Greatest Hits are chosen by hand.

The pipeline, the schemas and the rubric are in the repository.

Changelog

  • —2026-09-21: 288 rows across 6 categories, dictionary 368bf0e44604, rubric v2.