Gnomon

Stand uphill. Cast the shadow.

A two-player abstract game about light. A sun circles the board, every stone casts a shadow, and a stone caught in an enemy's shadow is gone. Games take 5–10 minutes.

godot --path gnomon          # from the playground folder

Click a stone, then a highlighted cell. R new game · U undo · H hints on/off · S show tomorrow's shadows · F1 rules · right-click or Esc to deselect. Play against the computer (four levels) or two players on one screen.

How to play

Imagine we're sitting at a 7×7 board. Amber has five round stones along the bottom, Teal has five diamond stones along the top. Around the board is a dial with a sun on it. Amber moves first.

Shadows. Every stone casts a shadow straight away from the sun, two cells long. If the sun is in the north, shadows point south; if it's in the north-east, they point south-west. Shadows run in a straight line, orthogonal or diagonal, and they don't stop at stones.

Eclipse. After you move, every enemy stone standing in one of your shadows is eclipsed and removed. Your own shadows never hurt you, and an enemy's shadow only hurts you after they move. So "standing in a shadow" is a threat, not a death: you get a turn to react. Put it another way: whoever is uphill (nearer the sun, in line, within two cells) wins that standoff.

Your turn. Move one of your stones one step, in any of the 8 directions, onto an empty cell. You must move; you can't pass unless you truly can't.

The sun turns. After each round (one move by Amber, one by Teal) the sun moves one notch clockwise, 45°. The standoffs rotate with it: two stones that are safe now may be a death sentence in two rounds, and the other way round. A pair of enemy stones in line within two cells is a ticking clock; the sun's position decides who dies when it arrives, and you can see it coming.

Dusk. One full circuit of the sun is a day, 8 rounds. At the end of each day the outermost ring of cells goes dark and any stone on it is lost. The board shrinks 7×7 → 5×5 → 3×3. When the third dusk falls, the game ends.

Winning. Eclipse every enemy stone and you win. If the last dusk falls first, whoever has more stones wins; equal is a draw. (Both sides losing their last stones to the same dusk is also a draw.)

Reading the board

• Tinted streaks are shadows, coloured by their owner. Don't step into the other colour's.
• A stone under a dark veil with a red ring and "!" is yours and in an enemy shadow: move it, or break the threat, before they move.
• A stone with a gold ring and "×" is theirs and in your shadow: almost any move that leaves your shadow in place takes it.
• A dashed ring marks a stone that will be in shadow once the sun turns, if nothing changes.
• The hollow ring on the dial is where the sun goes next.
• Pick a stone and the destination dots tell you what the move does: gold = takes a stone and leaves nothing of yours exposed, orange = takes one but exposes you, red = leaves you exposed for nothing, white = quiet. Hover a dot for the details.
• The red-tinted ring of cells pulses as dusk approaches.

A few things worth knowing

• Stones a knight's move apart (like b2 and c4) are never in line, so they never threaten each other, whatever the sun does.
• The sun starts in the north, so Teal begins uphill; Amber has the first move to make up for it. Around round 4 the sun passes south and the advantage swings to Amber. Fights are about the timing of that swing.
• Because the sun is on a fixed schedule, you can always see what the next few rounds will do. Most mistakes are forgetting the sun is about to move.
• Dusk is a second clock. Five stones on the edge need five moves to step inward; you only get one move per round.

Design notes

What I was going for. A game where the rules themselves have weather: one shared, perfectly predictable global state (the sun) that every local fight depends on, so the question is never just "can I take that stone?" but "when, and who will be uphill by then?" I wanted decisions that need you to hold a rotating picture in your head, with the rotation public and deterministic so it's a puzzle rather than a gamble, and a shrinking board so nobody can hide.

What I tried and threw out, with the numbers from AI-vs-AI games. The AI is a plain alpha-beta search, and all of this is reproducible with the tools below.

• Eclipses resolve instantly for both colours after every move. Hard to read (danger only shows when you blunder) and unbalanced: with ordinary alternating turns Teal, who always moves last before the sun turns, won 196 of 240 depth-3 games (82%) across the eight start directions.
• Snake turn order (A B | B A | A B, so the sun-turner moves again). Looked like the principled fix, but it made the result swing wildly with the sun's start direction: at depth 2, Amber won 59–1 from one start and lost 3–57 from another.
• Shadows only take stones on their owner's move + sun starts in the north was the combination that worked: depth 3 vs depth 3 gave 52–48 over 100 games. Search depth matters far more than side: depth 3 beat depth 2 by 79–19 as Amber and 96–4 as Teal, and depth 4 beat depth 3 by 14–2 and 15–1 (only 16 games each way). That's the "skill beats luck" property I wanted.
• Average game: about 24 plies (12 rounds), so most games end during the second day; dusk matters a lot to anyone who hides.

What I'm unsure about.

• Is it fun for a person, not a search algorithm? The AIs are ruthless about tempo; a human might find the rotation hard to hold in their head. The overlays try to carry that, but I haven't watched a newcomer play.
• Shadow length 2 and 5 stones were chosen by feel. I only compared shadow lengths 1–3 under the earlier, instant-eclipse rules (length 1 was lopsided toward Teal, length 3 made games a few plies shorter), never under the final rules. 5 stones means each loss hurts; 6–7 on a bigger board might give more room for strategy.
• Fixed start direction. It balances the AI games, but starting angle is a real lever (north: 52–48; south-west: Teal won 39 of 40). A pie rule, where Amber picks the starting sun and Teal picks a side, could be more elegant, but it asks a newcomer to judge something they can't yet evaluate.
• Endgames with one stone each become cat-and-mouse. That felt good, but long ones could stall until dusk.

Ideas I'd try next. Let players choose the starting sun (pie rule). A second, slower "moon" casting shadows the other way. Stones of different shadow lengths. Hex board with a 6-position sun. A move that lets you spend a stone to hold the sun for a round.

For tinkerers

godot --headless --path gnomon -s tools/test_rules.gd                    # rules unit tests (7000+ checks)
godot --headless --path gnomon -s tools/selfplay.gd -- games=100 p1=3 p2=3   # AI vs AI statistics
godot --headless --path gnomon -s tools/transcript.gd -- depth=3 seed=1  # one game, move by move
godot --path gnomon -- --autoplay    # watch the computer play both sides in the real UI

selfplay options: p1/p2 search depth (0 = random), len shadow length, day rounds per day, sun start direction 0–7 (N, NE, E, SE, S, SW, W, NW), snake=1, mover=0 (instant-eclipse rule). Game logic lives in scripts/rules.gd with no rendering; scripts/ai.gd is the search; scripts/view.gd draws; scripts/main.gd runs the game.