Tilewright: seven tiles, one opponent, and a board nobody else uses

01  What it is, and what it deliberately is not

Tilewright is a letter-tile word game you play in a browser tab. You hold seven tiles, you build words across a shared grid, and one AI opponent answers you turn for turn. It is free, there is no account, and there is nothing to install: the first screen is already a dealt game, not a menu.

A match is a complete game rather than an endless score chase. You play until the bag is empty and one side uses their last tile, or until four consecutive scoreless turns end it. A match has a shape: an opening where the board is wide, a middle where you fight over specific squares, and an endgame the racks decide.

What is not in it matters too. No human multiplayer, no chat, no leaderboard, no in-play hint button, no account keeping score on you. The game is supported by advertising, which is why it costs nothing. And it is not a branded board wearing a different name: the grid, the bonus layout, the tile values and the word list are its own.

02  The board is the unusual part

The grid is 13 × 13 — 169 squares — and 45 of them are bonus squares: 8 triple-word, 9 double-word, 4 triple-letter and 24 double-letter. The layout turns rather than mirrors. Rotate it a quarter turn and it is identical; hold it to a mirror and it is not. Almost every board in this genre is mirror-symmetric, which puts premium squares in matched pairs you can plan against from both sides.

The consequences are specific. The eight triple-word squares sit on the four corners and the four edge midpoints, so the edges are where the enormous turns live, and leaving a letter within reach of one is how a comfortable lead disappears. Four of the nine double-word squares form a diamond one step out from the centre, so a three-letter word straddling two of them is worth four times its face value. The centre square is itself a double-word, so the opening word, which must cover it, is doubled by definition.

The tile set is just as specific. The bag holds 96 tiles: 93 letters and 3 blanks. Values run along a ladder of 1, 2, 3, 4, 5, 8 and 11 and nothing else — Q and Z at 11, J and X at 8, K at 5, V at 4, F, H, W and Y at 3. Seven tiles in one turn pay a flat 45 on top, and three blanks in the whole bag, shared between both players, makes a blank genuinely scarce.

03  Three opponents, and why none of them cheats

The opponent comes in three tiers — Apprentice, Journeyman and Wright — and they are not three programs with different rules. They are one move generator with three ways of choosing from its ranked list of legal plays. All three see the same board and the same dictionary you do, and none of them can see your rack.

What separates them is judgement. Apprentice picks from the middle of its list, roughly the 55th to 85th percentile of what it found, and does not think about the rack it will be left holding: it plays a strong word at the wrong moment. Journeyman weights its choice toward the top fifth and does care about keeping a workable mix of vowels and consonants. Wright takes the move with the best combined score and rack value, tracks which letters are still in the bag, closes the board when it is ahead, opens it when it is behind, and plays the endgame out exactly.

That structure is the point: losing to Wright should feel like being outplayed rather than out-cheated. One win unlocks the next tier — no timer, no grind.

04  The Daily Rack, and a number that tells the truth

Once a day, every player gets the same seven tiles on an empty board. The rack is derived from the UTC date rather than picked by a server, so nobody is dealt an easier hand. You get a single submission: your best opening word.

Then the game does something more interesting than scoring you against strangers. It enumerates every legal opening word that rack could have played — usually a few hundred — and reports where yours landed: "your 34 beats 91% of every word playable from today's rack." That counts words, not people. With no server and no other player's score to compare against, "you beat 91% of players" would be a fabricated social signal.

What it reveals after you submit is worth more than the percentile: the best word the rack could have played, and the best word at each length it allows. That ladder is the useful part, because the longest word is not automatically the biggest. The centre star doubles whatever you place on it, so a short word on a letter bonus regularly out-scores a longer word played flat.

Play on consecutive days and a streak builds, kept in your own browser per device, while an archive of about a month of past racks stays readable.

05  Playing it on a phone, and how a turn actually feels

The game is built thumb-first. Tap a rack tile to lift it, tap a square to set it down, tap a tile you placed this turn to take it back. Squares that cannot legally accept your tile visibly refuse it, so you are never guessing why a placement will not take. The score preview updates live, and the primary button becomes "PLAY WORD" only once a legal word is on the board — with nothing placed it reads PASS instead, because a greyed-out button you cannot act on is a dead end.

Two small controls do most of the ergonomic work. Shuffle rearranges your rack, worth doing more often than feels necessary: a rack you have stared at for two minutes is a rack you have stopped seeing. Recall pulls every tile you have laid this turn back, so a plan that stops working costs you nothing but the thought. A zoom toggle fits the whole grid onto a 360-pixel-wide screen, and a mini-map shows where you are looking.

06  How a match ends, and what the results sheet tells you

A game finishes one of two ways. If the bag is empty and you play your last tile, you go out: you gain the value of every tile still on your opponent's rack and they lose it, so going out first is often worth more than the word you were saving. If neither side can score for four consecutive turns, each player instead loses the value of their own leftover tiles. You can exchange tiles rather than play, but only while the bag holds a full rack of seven.

The results sheet is where the game earns its place. It shows both scores, your best word rendered as actual tiles, and the panel that is genuinely rare in a free browser game: the three turns where you left the most on the board. Each redraws the board exactly as it stood when you played, shows your word cold, then fires the word you missed through the same animation a real placement gets. Only gaps of eight points or more count, and it says plainly that it covers the turns it checked.

Finally there is a share card: a 1200 × 630 image with your score, the tier you faced and one word. The whole brag, and no caption needed.

07  Questions people ask

Is Tilewright free, and do I need an account?

Free, and no account. The game is supported by advertising, which is why it costs nothing, and there is nothing to sign up for or install. Your streak, tier progress and best scores are kept in your own browser rather than on a server.

Does Tilewright work on a phone?

Yes — it is built mobile-first, with tap-to-place tiles rather than dragging, a zoom view that fits the whole 13 × 13 board on a 360-pixel-wide screen, and a mini-map for orientation. The page loads once over the network; after that the dictionary and the AI opponent both run inside your browser.

What word list does Tilewright use, and could a word be missing?

It ships ENABLE, a public-domain English word list, compressed and downloaded once. About 160,000 entries ship, after slurs and profanities were stripped out at build time — so the opponent cannot play one either. Because it is a fixed list rather than a live service, an unusual word can genuinely be rejected.

Does the AI opponent see my tiles or cheat?

No. All three tiers see the same board and the same dictionary you do, and none of them can see your rack. They differ only in how they choose among the legal moves they can find — Apprentice picks from the middle of its list, Journeyman weighs score against the rack it keeps, and Wright also counts the tiles left in the bag and plays the endgame exactly.

Is the Daily Rack percentile a comparison with other players?

No — it counts words, not people. The game enumerates every legal opening word the day's rack allows and reports the share of them your submission beat. That figure is computed in your browser from the board itself, because there is no server and no other player's score to compare with.

Can I play against another person, or get a hint during a game?

Neither is in the game. There is no human multiplayer and no in-play hint button. What you do get is recall, rack shuffle and the ability to take back tiles you have placed this turn, plus a panel after the match showing the three turns where you missed the most points.

08  More from the shelf

• Chess — Beat the engine. Post the win.
• Dominoes — Match the ends. Block them out.
• Backgammon — Roll the race. Take the cube.
• Fiveaside — Score past the keeper. Post the scoreline.

Browse every app in the catalogue →