Only knights.

Knights move in an L: two squares one direction, one square perpendicular. Every move flips colors — a knight on light always lands on dark and vice versa. That alternation is the heart of every knight tactic.

From a1, a knight has two destinations. From d4, it has eight. The corner is a one-way street; the center is a hub. Your default direction of travel is toward the middle.

A knight can't reach an adjacent square in one hop. Going from b1 to c1 takes three moves. Distance on a knight's board doesn't match grid distance — counting hops is the right unit.

Classic: don't land on a square the opponent's knight attacks — collision means you lose. Capture: landing on the opponent's knight wins outright. Same board, opposite intuition. Practice both — they sharpen different muscles.

Tempo is whose move is doing the work. A move that limits the opponent's options while keeping yours open is a free move. Look for squares that attack their next likely landing while preserving multiple of your own.

In an L-shape: two squares in one orthogonal direction (up, down, left, or right), then one square perpendicular to that. That's the only move the knight has, and it's why knights are the most mobile piece in the opening and the most awkward in the endgame — their moves don't line up with the intuitions you build from other pieces.

From any square well away from the edge, a knight has eight legal destinations — four L-shapes, mirrored left/right. From the corner, it has two. From an edge square (but not a corner), it has three or four. This mobility-by-location asymmetry is the single most important fact in the game.

An isolation game is an abstract-strategy format where each player controls a single piece, and the square a piece leaves is marked permanently unusable. Play continues until one player has no legal move — that player loses. Isolation games are a tiny family (fewer than a dozen well-known ones) but they're a great teaching vehicle because the rules fit on a sticky note and the strategy is deep.

A knight-isolation game is the variant where the piece involved is a chess knight. The knight's awkward L-move combined with the shrinking board makes it much richer than a king- or bishop-based isolation game. The knight can't retreat along its own path — whatever squares you used to get to your current position are gone — so every move is a commitment.

In Classic, landing on the opponent's square is illegal. You win only by leaving them no legal move. The game tends to last longer, play more positionally, and turn on subtle space-management decisions. Good Classic play looks patient — you're squeezing the opponent's mobility down one move at a time.

In Capture, landing on the opponent's square is legal and wins instantly. The game plays faster, more tactically, and often ends in fewer moves. The tension is mutual — either knight could be captured at any moment if the opponent gets a free shot — so you're always watching two things: what the opponent can do to you this turn, and what they'll be able to do next turn.

Practical difference: in Capture, you often have to reject your best positional move because it would leave your knight capturable. In Classic, no move is outright losing-on-the-next-turn — the game is pure slow pressure. Most players find Capture more fun the first few games because the stakes feel higher; Classic grows on you as you start seeing the longer-range plans.

A knight in the corner of an empty board has exactly two legal moves. From a1 those are b3 and c2. That's it. The corner is the lowest-mobility square on the entire board.

In an isolation game, any position where your knight has only two legal moves is one move away from a position where it has one (the opponent just has to block one of the two). And one legal move is one move away from zero, which is losing. So corners — and to a lesser extent, other low-mobility edge squares — are traps.

Walkthrough — the corner squeeze:

1. Your knight is on a1. Legal moves: b3, c2. (Mobility: 2.)
2. Opponent plays to b3 (blocks it). Your only legal move is now c2. (Mobility: 1.)
3. You move to c2. Your a1 square is now dead. Opponent plays to block one of c2's legal moves.
4. The noose tightens a move at a time. Unless you can force the opponent into a worse position right now, this ends with you stuck.

Shortcut: keep your knight where it has at least three legal moves. When mobility drops to two, you're already in trouble — find a move that restores mobility, or find a move that drops the opponent's mobility even faster.

Count two numbers: my mobility (legal moves for my knight right now) and their mobility (legal moves for theirs). Then compare.

• Ahead on both numbers, opponent is in the corner or near it: you're winning. Close down their escape routes without compromising your own position.
• Similar mobility, middle of the board: it's an open position. Play the move that drops their mobility more than yours.
• Behind on mobility, near the edge: you need to complicate. Look for moves that threaten two of their destinations at once, or that force them into your half of the board where you can create more dead squares around them.

The trap to avoid: hoarding your own mobility. A move that keeps your options open is great if it also hurts the opponent; if it's just "safe," they'll use the turn to squeeze you anyway. Isolation games reward moves that do two jobs — improve you AND worsen them — over moves that only do one.

Tempo is a word borrowed from chess. It literally means "a turn" — one player's move. You have a tempo advantage when every move available to your opponent makes their position worse, and at least one move available to you makes yours better. In that scenario, each exchange of moves widens the gap.

The classic tempo-winning move in a knight-isolation game is one that threatens two opponent destinations at once. They can only defend one, so the other goes to you next turn. You didn't gain material (there is no material), but you gained a turn of initiative — a tempo.

Why it matters in practice: end-of-game isolation positions often come down to a single-tempo race. Both players' knights are cornered, both players have one or two legal moves, and whoever runs out first loses. The tempo swings you built five moves earlier are what decide that race.

The simplest reliable technique:

1. Pick your two or three candidate moves from the current position. Don't try to evaluate all eight — just the ones that pass a first-glance sniff test.
2. For each candidate, mentally make that move, then mentally make the opponent's best reply. Their best reply is usually the move that most reduces your mobility in the new position.
3. Count your remaining mobility after their reply. That's your depth-2 score for the candidate.
4. Pick the candidate with the highest depth-2 score. If two are tied, pick the one that also drops the opponent's mobility the most.

This is exactly what a chess engine does at depth 3 — your move, their best reply, your evaluation. Done by hand it's slow and fallible, but it's dramatically better than picking the first move that looks good.

Don't plan deeper than two moves by default. Three moves ahead (mine-theirs-mine) is the upper limit for most humans without a notebook. Four moves ahead is professional-chess territory and requires specific training. Most positions in Only Knights can be played well at depth 2; the few that can't are usually late-game squeezes where you can count to the end of the game directly.

A little. Research on expert chess and go players consistently finds narrow transfer: these players are better at tasks that closely resemble the game they've trained on (pattern recognition in similar layouts, multi-step planning on gridded problems) and no better at tasks that don't. Playing a thousand games of chess does not reliably make you better at chemistry, writing, or taxes.

What abstract strategy games do reliably build:

• The "what happens if I do this" reflex. Players who've drilled turn-based games for a while tend to imagine consequences before acting, even outside the game. That habit is small but genuinely transferrable.
• Comfort sitting with an unsolved position. Most people make snap decisions to escape ambiguity; strategy games reward pausing in ambiguity until the right move reveals itself. The ability to not-decide for another ten seconds is itself a skill.
• Resistance to tilting. Playing through a losing position without giving up early is a learned skill. Isolation games that drag on after you've lost tempo are good practice for it.

Don't expect Only Knights to fix bad life decisions, but don't dismiss it either. The thinking style is real; the transfer is narrow but not zero.

Speed is Streak mode with a 10-second per-move clock. When it's your turn the clock starts at 10.0s and counts down inside the turn banner. If it hits zero before you move, you lose that match — the same as any other loss in the streak. The clock resets to 10.0 the moment it becomes your turn again.

The clock is server-authoritative in multiplayer: the server holds the truth about the deadline, so you can't buy extra time by messing with your device clock. The 10-second budget is deliberate — it's long enough to find a solid move on a familiar position, short enough that hesitation is expensive. If you're new to Only Knights, start in Streak (no clock) and move to Speed once the L-move patterns feel instant.

Speed gives you 10 seconds per move and resets at the start of every turn. Bullet and Blitz borrow a different idea from chess: each player gets a total budget for the whole match that ticks down only while it's their turn. Bullet is 1 minute per player, Blitz is 3 minutes per player.

Because the clock is shared across the whole match, you can spend 20 seconds on a hard opening choice and still have time to breeze through the obvious closeout — as long as the total stays on your side of zero. The budget resets to a full minute (or three) at the start of every new match in the streak, so a long run doesn't turn into an unwinnable endurance contest. If your clock hits zero on your turn you lose the match and the run ends — same as any other loss in Streak / Speed.

Bullet averages about 5 seconds per move across a typical ~12-move match; Blitz averages about 15 seconds. Both are server-authoritative in multiplayer, so the opponent's device clock can't cheat you out of a flag. Rough pick: use Speed if you're practising quick pattern recognition, use Blitz if you want a chess-feeling game where you can think on the hard moves, use Bullet if you want the frantic feeling of a real bullet game.

Every day, the start screen shows a Today's puzzle card with a tiny board preview, the date, and your current daily streak. Tap Solve today's puzzle to load the position on the main board. The puzzle is client-side and date-seeded — every visitor on the same calendar date sees the same position, Wordle-style.

Think of it as dropping into a match already in progress: dead squares are already on the board, your knight is already somewhere, and the AI opponent is already wherever it ended up. Your job is just to survive. Win the match — trap the opponent, or in Capture, land on them — and the day is solved. Lose and it's a miss. There's no ply budget, no timer, and no requirement to play the "right" opener.

Each puzzle does have a computed best first move though. Play it and you earn the Perfect-solve bonus badge on top of the win; miss the best opener and you still get credit for the win with a plain ✓ Solved badge. Either way, one attempt per day — you can't re-roll after committing. The ruleset alternates by date (even-numbered days Classic, odd-numbered Capture) so nobody camps a single style. The streak counter increments each day you win and resets the first time you lose or skip a day. Streak, log, and the day's cached puzzle all live in your browser's localStorage; there is no server-side record and no leaderboard (yet).

On the start screen, tap Play with a friend. A room opens, you'll see an invite link, and whoever opens that link first joins you. The link is in the form practice.saltnfork.com/games/only-knights/?room=abc1234 — you can send it over any chat app. Only one friend at a time: if a third person opens the link after the room is full, they see a "room full" message.

Tap Play a stranger on the start screen and you drop into a lobby queue. When another player is also in the queue, the server sends both of you a match proposal — a lightweight handshake that says "here's a potential opponent, do you both want to play?" The MatchRoom itself isn't created until both sides have accepted, which means declining never leaves the other player staring at a ghost room.

If you tapped Play a stranger yourself, your side of the proposal is accepted automatically — your click was the consent. You just keep seeing the "Searching…" spinner until the match begins. The only time you'll see an explicit Accept / Decline modal is if you opted into Notify Me and are playing solo while waiting in the background; there, a proposal should pop up with the other player's handle and give you 30 seconds to decide.

Decline three proposals in a row within an hour and you're removed from the waiting list with a clear notification, so nobody ends up proposing into a black hole. You can re-opt in whenever you're ready. Handles are locked to 2–8 letters or digits — no symbols, no spaces, no Unicode — and the server runs the same filter as the client, so nothing weird can leak onto your opponent's screen. If the 60-second search window expires with nobody around, you get the usual fallbacks: Try again, Notify me, Send invite link, or Play against the computer.

They get a 90-second grace window. During those 90 seconds the match is paused on your side and their seat is marked as reconnecting — if they come back (or refresh the tab) within the window, the match resumes from exactly where it was, with the same clock state on each side. If they don't return in 90 seconds, the match is awarded to you by forfeit and you drop back to the start screen so you can either host a new room or play solo.

Yes. A reconnect token is saved in your browser's localStorage when you join a room. If you refresh, close the tab and reopen it, or navigate away and come back within the match, the page re-enters the same seat automatically — you don't need the invite link a second time. Clearing site data or joining a different room replaces the token.

No. Multiplayer matches are tracked separately: they don't touch your solo personal-best streak or the leaderboard submission, and a multiplayer loss won't break a solo streak. You'll see the recent MP results in the Recent multiplayer matches list on the start screen (stored locally in your browser — no account, no server-side history).

Two separate timers can run in a multiplayer match. The 10-second per-move clock applies only in Speed mode. In Bullet (1 minute per player) and Blitz (3 minutes per player) each side has a total-match budget that ticks down only on their turn. Independent of either, every room has a 60-minute idle lifetime — if there's no activity (no moves, no settings proposals, no chat-like signals) for 60 minutes, the room closes and both players drop out. Any activity resets the 60-minute timer, so a normal back-and-forth match will never hit the idle cap. The HUD shows the remaining room lifetime next to the room code so you're not surprised.

The side panel shows a collapsible Move history list for the current match. Moves use coordinate notation — no chess experience required. a1→c3 means "a knight moved from a1 to c3." a1×c3 means "a knight moved from a1 to c3 and captured the opponent knight that was there." The arrow is a plain move, the × is a capture. Pairs are numbered — the left column is whoever moved first this match, the right column is whoever replied. Hover or tap a move to flash its from- and to-squares on the board; useful for reviewing the turning point of a losing match. The list clears on every new match and isn't stored anywhere — refresh the page and it's gone.

Once you've focused a board square (click it, or Tab until the outline appears), the arrow keys move focus one square at a time — ArrowUp and ArrowDown step along a file, ArrowLeft and ArrowRight step along a rank. Dead squares are skipped automatically. Press Enter or Space on the currently-focused square to tap it (same as clicking). Esc clears a solo selection without making a move. H shows a hint while it's your turn in solo. Works identically in multiplayer, minus the hint.

The speaker icon in the top-right corner of the header toggles sound effects. Short synthesized cues play when a knight moves (one tone for your move, a quieter lower tone for your opponent's) and at match end (a rising two-note chime for a win, a descending pair for a loss). The preference is remembered in your browser's localStorage, so it sticks between sessions. Touch devices default to muted; mouse-based devices default to unmuted — both defaults are overridden as soon as you touch the toggle. There are no bundled audio files; everything is synthesized via the Web Audio API on first user gesture.

At the end of a run (or a multiplayer match) there's a Download image button in the result card. It builds a 1080×1080 PNG containing the final board on the left, the full move history next to it on the right, and a headline strip underneath with your result, streak (or multiplayer series score), mode, and ruleset. The file downloads directly to your device so you can post it, save it, or send it however you like — no share-sheet or copy-link middle step to fight with. The board is rendered on canvas directly from the game state, so knights show as labelled discs and the result is identical across browsers.