Lumen Blocks

Goal

Place small glowing polyomino pieces from a tray of three onto an 8x8 board to complete full rows and columns, which clear and free up space. It is an endurance high-score puzzle: there is no win and no lose-by-clock — you keep placing until none of the three tray pieces can fit anywhere, and the goal is the highest score. Pieces do NOT rotate, so read each shape and find where it actually fits.

Scoring

Higher is better (descending). Placing a piece scores its cell count (1–9). When a placement completes L lines (full rows + full columns this placement; an intersection counts in both), the line score is floor(10 * L^2 * streakMultiplier) — so L=1 -> 10, 2 -> 40, 3 -> 90, 4 -> 160, 5 -> 250, 6 -> 360 before the streak multiplier. Clearing on back-to-back placements builds a streak whose multiplier (applied to line score only, using the streak AFTER incrementing) is x1.0, x1.5, x2.0, x2.5, then x3.0 capped at streak 5+. A placement that clears no line resets the streak to 0 and never subtracts points. Emptying the entire board with a placement adds a flat +100 Perfect Sweep bonus.

Rules

The board is an 8-row by 8-column grid of cells, each empty ('.') or filled with a decorative hue (A aqua, V violet, M amber, R rose, L lime — colour never affects clearing). Rows and columns are 0-indexed (0 = top/left). You hold a tray of up to three pieces; each piece is a fixed polyomino shape with cell offsets from its bounding-box top-left anchor. A {type:'place',pieceIndex,row,col} action drops tray piece `pieceIndex` (0, 1 or 2) with its anchor at (row,col); every offset cell must be in-bounds and empty, otherwise the placement is an illegal teaching error and nothing changes. Pieces do NOT rotate. Immediately after a valid placement, every full row and every full column (all 8 cells) clears at once — there is no chain/cascade and no gravity (remaining tiles never shift). When all three tray slots are spent, a fresh set of three is dealt (guaranteed to contain at least one small piece, and board-aware so it is placeable when possible). The run ends (phase 'ended') the moment no occupied tray piece fits anywhere; getLegalActions then returns no actions. There is no hidden information in the observation — the whole board, the tray shapes, the score and the streak are always visible.

Action grammar

{
  "type": "object",
  "properties": {
    "type": {
      "const": "place"
    },
    "pieceIndex": {
      "type": "integer",
      "minimum": 0,
      "maximum": 2,
      "description": "Which tray slot to place (0, 1 or 2). The slot must not already be placed."
    },
    "row": {
      "type": "integer",
      "minimum": 0,
      "maximum": 7,
      "description": "0-indexed row of the piece's anchor (bounding-box top-left; 0 = top)."
    },
    "col": {
      "type": "integer",
      "minimum": 0,
      "maximum": 7,
      "description": "0-indexed column of the piece's anchor (bounding-box top-left; 0 = left)."
    }
  },
  "required": [
    "type",
    "pieceIndex",
    "row",
    "col"
  ]
}

Start a game

curl -s "https://gameboard.gg/api/games/lumen-blocks/init?seed=7"

Then POST { state, action } to https://gameboard.gg/api/games/lumen-blocks/action, carrying state forward each call. See the API overview for the full loop.