feat(semantle): source target pool from google-10000-english dictionary

The ~250-word hand-curated TARGET_POOL was too small for long-term play.
Replaces it with a build-script-generated dictionary:

- scripts/build-semantle-words.js fetches first20hours/google-10000-english
  (no-swears variant), filters to 4–10 ASCII letters, drops the top-200
  most frequent function words, and writes src/modules/semantle/words-data.js
  as a static ES-module export.
- wordlist.js now just re-exports that data via TARGET_POOL + pickFromPool.
- package.json: new build:semantle-words script; chained into `npm run build`
  alongside build:wordle-data so `npm run deploy` regenerates automatically.

Pool size: ~250 → 7953 words. Same ConceptNet verify-and-fallback flow, so
low-quality picks still cost at most one extra concept lookup.

package.json

@@ -9,8 +9,9 @@
   },
   "scripts": {
     "dev": "wrangler dev",
-    "build": "npm run build:wordle-data",
+    "build": "npm run build:wordle-data && npm run build:semantle-words",
     "build:wordle-data": "node scripts/build-wordle-data.js",
+    "build:semantle-words": "node scripts/build-semantle-words.js",
     "scrape:loldle-data": "node scripts/scrape-loldle-data.js",
     "deploy": "npm run build && wrangler deploy && npm run db:migrate && npm run register",
     "db:migrate": "node scripts/migrate.js",

scripts/build-semantle-words.js

@@ -0,0 +1,67 @@
+#!/usr/bin/env node
+/**
+ * @file build-semantle-words — fetches a common-English word list (10k
+ * by Google Ngram frequency, curated by first20hours) and writes the
+ * 4–10 letter alphabetic subset to src/modules/semantle/words-data.js.
+ *
+ * Why this list:
+ *   - Sorted by frequency → easy top-N trimming to drop function words
+ *     (`the`, `of`, `and`) that make terrible guessing targets.
+ *   - Already de-swear-ed. No further blocklist needed.
+ *   - Tiny (~90 KB raw, ~30 KB gzipped) — comfortably inside the Worker
+ *     size budget.
+ *
+ * Source:  https://github.com/first20hours/google-10000-english
+ * Credits: Josh Kaufman (first20hours) — list derived from Peter Norvig's
+ *          Google Ngram analysis.
+ *
+ * Usage:
+ *   node scripts/build-semantle-words.js
+ */
+
+import { writeFileSync } from "node:fs";
+import { resolve } from "node:path";
+
+const SOURCE_URL =
+  "https://raw.githubusercontent.com/first20hours/google-10000-english/master/google-10000-english-no-swears.txt";
+
+// Skip the top-N most frequent words — too common, lousy puzzles.
+const SKIP_TOP_N = 200;
+const MIN_LEN = 4;
+const MAX_LEN = 10;
+
+const root = resolve(import.meta.dirname, "..");
+const dst = resolve(root, "src/modules/semantle/words-data.js");
+
+const res = await fetch(SOURCE_URL);
+if (!res.ok) throw new Error(`fetch failed: ${res.status} ${res.statusText}`);
+const text = await res.text();
+
+const lines = text.split(/\r?\n/).map((w) => w.trim().toLowerCase());
+
+// Preserve original frequency order while filtering — downstream consumers
+// can still sample uniformly, but the index itself stays a frequency rank.
+const words = Array.from(
+  new Set(
+    lines
+      .slice(SKIP_TOP_N)
+      .filter((w) => w.length >= MIN_LEN && w.length <= MAX_LEN && /^[a-z]+$/.test(w)),
+  ),
+);
+
+if (words.length === 0) throw new Error("no words parsed from source");
+
+const body = words.map((w) => `  "${w}",`).join("\n");
+const out = [
+  "// Auto-generated from https://github.com/first20hours/google-10000-english",
+  "// Credits: Josh Kaufman (first20hours) — common English words by Google Ngram frequency.",
+  `// Filter: ${MIN_LEN}–${MAX_LEN} ASCII letters, skip top ${SKIP_TOP_N} most common.`,
+  "// Regenerate with: node scripts/build-semantle-words.js",
+  "export default [",
+  body,
+  "];",
+  "",
+].join("\n");
+
+writeFileSync(dst, out);
+console.log(`wrote ${dst} (${words.length} words)`);

src/modules/semantle/README.md

@@ -29,10 +29,16 @@ ignored (no cost, no stat inflation).
   carries at least one edge.
 
 Because ConceptNet has no random-word endpoint, the target pool ships in
-`wordlist.js` (~250 curated English words, 4–10 letters, all alphabetic).
+`words-data.js` — ~8k common English words (4–10 ASCII letters) derived
+from Google Ngram frequency via the public
+[google-10000-english](https://github.com/first20hours/google-10000-english)
+list, with the top-200 most frequent function words stripped.
 Each new round picks locally, verifies via the concept endpoint, and falls
 back to an unverified pick after a few misses.
 
+Regenerate the list with `npm run build:semantle-words` (chained into the
+main `npm run build` that `npm run deploy` invokes).
+
 Every guess costs **two** ConceptNet calls (concept edges + relatedness)
 issued in parallel. Typical latency ~300–600ms round-trip from Cloudflare
 Workers; `api-client.js` enforces a 5s timeout and surfaces a "Upstream
@@ -44,7 +50,8 @@ hiccup" message on failure.
   plus lower-level `concept` / `relatedness`) with `UpstreamError` metadata.
   Preserves the earlier word2sim response shape so the rest of the module
   didn't need rewriting.
-- `wordlist.js` — curated local target pool and `pickFromPool()`.
+- `words-data.js` — auto-generated dictionary (regenerate via `npm run build:semantle-words`).
+- `wordlist.js` — thin wrapper exposing `TARGET_POOL` and `pickFromPool()` over the dictionary.
 - `state.js` — KV persistence for game + stats. Target stored lowercased.
 - `lookup.js` — guess normalization and shape validation.
 - `format.js` — warmth-percent and emoji-bucket formatters.
@@ -84,4 +91,5 @@ across all rounds.
 ## Credits
 
 - Similarity + vocabulary: [ConceptNet 5](https://conceptnet.io) by Robyn Speer et al.
+- Target dictionary: [google-10000-english](https://github.com/first20hours/google-10000-english) by Josh Kaufman, derived from Peter Norvig's Google Ngram analysis.
 - Game concept: [Semantle](https://semantle.com/) by David Turner.

src/modules/semantle/wordlist.js

@@ -1,75 +1,18 @@
 /**
- * @file Curated target-word pool for the semantle module.
+ * @file Target-word pool for the semantle module.
  *
- * ConceptNet has no random-word endpoint, so we ship a hand-picked list of
- * common, game-friendly English nouns/verbs/adjectives (4–10 ASCII letters).
- * The list is small on purpose — every entry is a reasonable Semantle target,
- * which matters more than raw size. Expand freely as the game matures.
- *
- * Entries are lowercase and alphabetic only (the similarity endpoint accepts
- * these directly as `/c/en/<word>` concept IDs).
+ * The raw dictionary lives in `words-data.js` (auto-generated by
+ * `scripts/build-semantle-words.js` from the public google-10000-english
+ * list, filtered to 4–10 ASCII letters with the top-200 most frequent
+ * function words stripped). This module just exposes the list and a
+ * uniform random pick so api-client.js can stay focused on HTTP.
  */
 
-// biome-ignore format: keep the list compact and grep-friendly
-export const TARGET_POOL = [
-  // nature / geography
-  "ocean", "mountain", "forest", "desert", "river", "valley", "garden", "island",
-  "beach", "cave", "meadow", "glacier", "volcano", "canyon", "jungle", "lake",
-  "hill", "plateau", "cliff", "harbor", "coast", "swamp", "prairie", "tundra",
-  "delta", "creek", "stream", "pebble", "boulder", "horizon", "iceberg", "dune",
-  // weather / time
-  "winter", "summer", "autumn", "spring", "morning", "evening", "midnight",
-  "sunrise", "sunset", "thunder", "rainbow", "blizzard", "breeze", "drought",
-  "storm", "shadow", "twilight", "decade",
-  // people / relations
-  "friend", "family", "mother", "father", "brother", "sister", "child",
-  "stranger", "neighbor", "partner", "sibling", "elder", "infant",
-  // arts
-  "music", "dance", "poem", "story", "painting", "theater", "cinema", "novel",
-  "symphony", "sculpture", "sketch", "ballet", "opera", "concert",
-  // objects
-  "computer", "phone", "camera", "robot", "engine", "wheel", "pencil", "hammer",
-  "mirror", "bicycle", "umbrella", "lantern", "compass", "anchor", "blanket",
-  "candle", "cushion", "kettle", "ladder", "needle", "paper", "pillow",
-  "scissors", "telescope", "throne", "vase", "window", "zipper", "bottle",
-  "basket", "bridge", "tower",
-  // animals
-  "eagle", "tiger", "dolphin", "rabbit", "snake", "salmon", "wolf", "horse",
-  "butterfly", "elephant", "panda", "falcon", "sparrow", "penguin", "octopus",
-  "beetle", "crow", "dragon", "hawk", "jaguar", "kangaroo", "lion", "monkey",
-  "otter", "parrot", "raccoon", "squirrel", "turtle", "whale", "bear", "fox",
-  "shark",
-  // food
-  "apple", "bread", "cheese", "coffee", "sugar", "pepper", "potato", "orange",
-  "honey", "chocolate", "cinnamon", "almond", "berry", "butter", "grape",
-  "lemon", "olive", "tomato", "walnut", "yogurt", "ginger",
-  // emotions / abstract
-  "love", "anger", "fear", "courage", "sorrow", "wonder", "dream", "memory",
-  "silence", "laughter", "hope", "delight", "regret", "trust", "justice",
-  "freedom", "honor", "peace", "victory", "promise", "secret", "truth",
-  "wisdom", "mystery", "destiny", "patience", "loyalty",
-  // professions
-  "teacher", "doctor", "artist", "farmer", "pilot", "soldier", "writer",
-  "sailor", "hunter", "engineer", "chef", "dentist", "nurse", "judge",
-  "scientist",
-  // body
-  "shoulder", "finger", "elbow", "throat", "pulse", "tongue", "ankle", "spine",
-  "muscle", "nerve", "eyelid", "beard", "tooth", "thumb", "knee",
-  // verbs / actions
-  "gather", "wonder", "linger", "stumble", "whisper", "shimmer", "wander",
-  "rescue", "gallop", "imagine", "pursue", "retreat", "thrive", "squeeze",
-  "shatter", "tremble", "travel", "borrow", "descend", "inherit", "vanish",
-  "forget", "invite", "resist", "settle",
-  // adjectives
-  "ancient", "gentle", "fragile", "massive", "glowing", "rugged", "silent",
-  "frozen", "steady", "brittle", "clever", "gloomy", "cheerful", "noisy",
-  "silver", "golden", "radiant", "distant", "graceful", "humble", "vivid",
-  "quiet", "sacred", "sudden", "tender", "wealthy", "generous", "majestic",
-];
+import WORDS from "./words-data.js";
 
-/**
- * @returns {string} — a random lowercase word from the pool.
- */
+export const TARGET_POOL = WORDS;
+
+/** @returns {string} — a random lowercase word from the pool. */
 export function pickFromPool() {
-  return TARGET_POOL[Math.floor(Math.random() * TARGET_POOL.length)];
+  return WORDS[Math.floor(Math.random() * WORDS.length)];
 }