LINGUABASE — MAKE A CLUE GAME (machine-readable method file) ============================================================ Companion to https://www.linguabase.org/ — a free, public-domain (CC0) semantic dataset of English. This file is self-contained: tables, the convergence idea, method, measured results, and a pointer to a complete build script. Written to be read by an LLM asked to adapt the method to a different game. THE CONVERGENCE IDEA -------------------- The player sees 4 CLUES and 4 OPTIONS (1 answer + 3 wrong). The clues come at the answer from its DIFFERENT SENSES: road, death, chime, turnpike -> TOLL (the fee sense, the burden sense, the bell sense). The answer connects to all four clues. Each wrong option connects to some clues and has ZERO connection to at least one — HIGHWAY fits road and turnpike but has nothing for death or chime. That zero is the player's elimination path. Zeros matter more than highs: the answer wins by having none, not by scoring highest on any single clue. THE DATA -------- Download (parquet or TSV, identical on three mirrors, linked from linguabase.org/download.html). Tables used here, 6 of 13: associations word, associates a ranked cloud of related words, strongest first senses word, sense_labels the word's canonical senses sense_associations word, sense, associates a separate cloud per sense; key (word, sense) vocabulary word, core_rank rank 1 = most familiar; the core file cuts at the recommended 400,000 threshold word_families word, relatives morphological relatives scores_labels word, content_rating, sensitivity, ... content_rating G/PG/PG-13/R/X; sensitivity 1-5, HIGH = SAFE (5 is the neutral majority) File notes: the shipped files are not sorted by rank and the TSVs carry a header row — sort on core_rank, don't trust file order. List columns are parquet list; in the TSVs they are pipe-delimited (split on "|"). sense_associations carries both the canonical senses (bare labels matching senses.sense_labels) and hundreds of finer analysis senses per word; filter to the canonical labels unless you want the fine grain. STRENGTH (the scoring primitive) -------------------------------- strength(word, clue) = 100 - position of clue in word's ranked associates (1-based); absent or past position 99 = 0. It is a position proxy, not a semantic measurement — the separation floor below is what keeps faint matches from deciding puzzles. THE PUZZLE RULES ---------------- answer from the top 10,000 by core_rank; family-safe (G/PG and sensitivity > 3); 3+ letters; has a cloud; has 2+ senses clues 4 of the answer's top-30 associates, spread across >= 2 canonical senses (sense_associations says which sense each associate belongs to), none a family relative of the answer wrong 3 words from the top 30,000 (same safety filter): each has options strength > 0 on exactly 2-3 clues and 0 on the rest; none a relative of the answer or any clue; together they cover all 4 clues; the answer's strength total beats the best wrong option's by 100+ ("separation") THE SUPPLY (measured, June 2026 release) ---------------------------------------- Funnel over the 10,000 most familiar words: 10,000 -> 9,637 family-safe -> 9,599 with a cloud (3+ letters) -> 8,999 with 2+ senses -> 8,925 with 4 clues across senses -> 5,488 with 3 clean wrong options = one a day for 15.0 years Answer-pool sensitivity (same funnel, wrong-option pool fixed at 30k): top 5,000: 2,915 anchors ( 8.0y) · top 10,000: 5,488 (15.0y) top 25,000: 11,567 anchors (31.7y) · top 50,000: 19,153 (52.5y) These are ANCHORS, not puzzles: the count picks ONE deterministic clue set per answer. An answer with six senses supports several distinct puzzles, so the supply is larger than the anchor count. A REAL PUZZLE (from the build) ------------------------------ answer TOLL (rank 6,455) clues: road, death, chime, turnpike clue senses: fee, burden, bell, fee road death chime turnpike TOLL 88 95 93 89 total 365 highway 88 0 0 89 total 177 freeway 77 0 0 84 total 161 rattles 0 58 61 0 total 119 separation 200 (365 - 177) The play pack carries the matrix, so feedback can name the zero: pick HIGHWAY and the game can say it has no connection to DEATH or CHIME. All 5,488 packs together: 863 KB minified JSON; 30 days = 4.4 KB. WHAT THE MECHANICS DON'T JUDGE ------------------------------ Answer quality. NOT and YES both anchor valid puzzles at this spec — four clues, clean wrong options, a 100-point win. Curate the answer list, or cut sharper with scores_labels (game_suitability 1-7, concreteness) before shipping. THE SCRIPT ---------- https://www.linguabase.org/linguabase-clue-game.py — the complete build (also at linguabase-clue-game.html with a line-numbered viewer). Run it beside the six core parquet files; it writes puzzles.tsv and puzzles.json. Adapt the constants at the top: ANSWER_POOL, DECOY_POOL, CLUE_POOL, SEPARATION_MIN. CITATION -------- Linguabase (2026). DOI: 10.5281/zenodo.20788428 (CC0 / public domain).