diff --git a/docs/plans/2026-02-23-word-frequency-filter-design.md b/docs/plans/2026-02-23-word-frequency-filter-design.md
new file mode 100644
index 0000000..ff159c2
--- /dev/null
+++ b/docs/plans/2026-02-23-word-frequency-filter-design.md
@@ -0,0 +1,77 @@
+# Word Frequency Filter — Design
+
+## Overview
+
+Reduce noise in word mode results by frequency-filtering the noun and adjective
+lists. Words are sorted by frequency rank (Google Books corpus) so the solver
+can slice to a cutoff. A new UI dropdown lets the user choose vocabulary size.
+
+## Build Step Changes
+
+Modify `scripts/build_word_lists.py` to:
+
+1. Download `frequency-alpha-alldicts.txt` from hackerb9/gwordlist (~246K
+   dictionary-verified words with frequency rankings from Google Books).
+2. Assign a frequency rank to each noun and adjective. Words not found in the
+   frequency list get rank `infinity` (pushed to end).
+3. Sort `nouns.txt` and `adjectives.txt` by frequency rank (most common first).
+4. Re-sort `nouns_plural.txt` to match the new noun order (index-aligned).
+
+## UI Changes
+
+New dropdown visible when `mode=word`, label "Vocabulary":
+
+```html
+<select id="solve-vocab">
+  <option value="3300">Common</option>
+  <option value="6400" selected>Standard</option>
+  <option value="12000">Extended</option>
+  <option value="0">Full</option>
+</select>
+```
+
+Values are approximate noun counts at each frequency tier:
+- **Common** (3,300): top 5K frequency words — very clean, fast
+- **Standard** (6,400): top 10K — good balance (default)
+- **Extended** (12,000): top 20K — broader coverage
+- **Full** (0 = no limit): all words — maximum coverage, may be noisy
+
+The dropdown is hidden for non-word modes (same pattern as plural checkbox and
+filter dropdown).
+
+## Backend Changes
+
+### `solve_word_dictionary()` — new parameter
+
+```python
+def solve_word_dictionary(
+    ...,
+    vocab_size: int = 0,  # 0 = no limit
+) -> Generator[SolveResult, None, None]:
+```
+
+At the start of the function, slice the word lists:
+
+```python
+if vocab_size > 0:
+    nouns = nouns[:vocab_size]
+    plurals = plurals[:vocab_size]
+    adjectives = adjectives[:vocab_size]
+```
+
+Since the lists are frequency-sorted, this keeps only the N most common words.
+
+### `SolveRequest` — new field
+
+```python
+vocab_size: int = 0  # 0 = no limit
+```
+
+Passed through to `solve_word_dictionary()`.
+
+## Frontend Changes
+
+- `index.html`: Add vocabulary `<select>` inside `.solve-controls`
+- `dom.js`: Export `vocabLabel` and `solveVocab`
+- `popover.js`: Show/hide `vocabLabel` when mode changes
+- `solver.js`: Include `vocab_size: parseInt(solveVocab.value)` in request body