About

A letter matrix (word search) embeds n target words into a W × H grid of alphabetic characters. Words may run in 8 compass directions: horizontal, vertical, and both diagonals, each potentially reversed. The generator first places words using a backtracking algorithm, then floods remaining cells with uniformly distributed random letters from the English alphabet. Failure to account for placement collisions produces unsolvable or degenerate grids. This tool validates every placement against existing letters before committing, guaranteeing a solvable puzzle for every generation.

Grid density matters. A 10×10 matrix with 15 long words will fail to place them all. The tool calculates fill ratio r = total word characters ÷ total cells and warns when r exceeds 0.6. Cognitive research suggests optimal puzzle difficulty sits near r ≈ 0.35 to 0.45, balancing challenge against frustration. Custom word lists accept only alphabetic inputs stripped of whitespace and normalized to uppercase.

Formulas

Word placement validity requires that for each character index i in a word of length L, the target cell coordinates remain within grid bounds and do not conflict with previously placed letters.

col_i = col₀ + i ⋅ dx, row_i = row₀ + i ⋅ dy

Valid if 0 ≤ col_i < W and 0 ≤ row_i < H and (grid[row_i][col_i] = NULL or grid[row_i][col_i] = word[i])

The fill ratio measures grid density:

r = n∑k=1 L_kW × H

Where col₀, row₀ are the starting coordinates, dx and dy are direction deltas (−1, 0, or 1), W is grid width, H is grid height, L_k is the length of the k-th word, and n is total word count. The adjacency check for drag selection uses Chebyshev distance: max(|Δcol|, |Δrow|) ≤ 1.

Reference Data

Grid Size	Total Cells	Recommended Words	Max Word Length	Difficulty	Avg. Solve Time	Fill Ratio Target	Directions Used
8×8	64	4 - 6	8	Beginner	2 - 4 min	0.30 - 0.40	H, V
10×10	100	6 - 10	10	Easy	3 - 6 min	0.30 - 0.45	H, V, D
12×12	144	8 - 14	12	Medium	5 - 10 min	0.35 - 0.50	H, V, D
14×14	196	10 - 18	14	Medium-Hard	8 - 15 min	0.35 - 0.50	H, V, D, R
16×16	256	12 - 22	16	Hard	12 - 20 min	0.35 - 0.55	H, V, D, R
18×18	324	14 - 26	18	Expert	15 - 30 min	0.35 - 0.55	All 8
20×20	400	16 - 30	20	Master	20 - 45 min	0.30 - 0.55	All 8
Direction Key: H = Horizontal, V = Vertical, D = Diagonal, R = Reverse
8 Compass Directions
Right	dx = 1, dy = 0		Left	dx = −1, dy = 0		← →
Down	dx = 0, dy = 1		Up	dx = 0, dy = −1		↑ ↓
Down-Right	dx = 1, dy = 1		Up-Left	dx = −1, dy = −1		Diagonal
Down-Left	dx = −1, dy = 1		Up-Right	dx = 1, dy = −1		Anti-diagonal

Frequently Asked Questions

When a word is being placed, each target cell is checked. If the cell is empty (NULL), the letter is written. If the cell already contains a letter from a previously placed word, placement is allowed only if that existing letter matches the current word's letter at that index. This intersection strategy increases grid density and creates satisfying overlaps. If any cell conflicts, the algorithm retries with a different starting position and direction, up to 500 attempts per word before skipping it.

The selection path must satisfy three conditions: (1) all cells are collinear, meaning they follow a single straight line in one of the 8 compass directions; (2) each consecutive cell pair has a Chebyshev distance of exactly 1; and (3) the extracted string from the selected cells exactly matches one of the hidden words that has not yet been found. The tool rejects curved paths, gaps, and already-discovered words.

A word of length L requires L contiguous cells along a single direction. Available empty cells may be scattered rather than aligned. Additionally, previously placed words create blocking patterns. The algorithm tests all 8 directions from random starting positions. If no valid placement exists after 500 random attempts, the word is reported as unplaceable. Reducing word count or increasing grid size resolves this. The fill ratio indicator warns when the grid is becoming too dense.

Larger grids increase peripheral noise - the ratio of distractor letters to target letters rises. A 10×10 grid with 8 words might have a fill ratio of 0.40, meaning 60% of letters are random distractors. A 20×20 grid with the same 8 words drops to 0.10 fill ratio, embedding words in 90% noise. Additionally, larger grids reduce the effectiveness of edge-scanning strategies since words starting mid-grid become harder to locate visually.

Yes. Random filler letters can accidentally form valid words not in the target list. This is a known property of word search puzzles and is not considered a defect. The tool only validates against the placed word list, so accidental formations are ignored by the scoring system. To minimize false patterns, the filler algorithm uses weighted letter distribution favoring less common consonants (Q, X, Z, J) at reduced frequency matching natural English letter distribution.

The tool supports grids up to 20×20 (400 cells) with up to 30 words. DOM rendering at this scale produces 400 interactive cell elements, well within browser performance limits. The placement algorithm runs in under 50ms for worst-case scenarios on modern hardware. Beyond 20×20, mobile viewports cannot render legible single-character cells without horizontal scrolling, which degrades the puzzle experience.