How does the AI
choose a move?

1. Balance the next line against the next piece

Filling a row clears it and frees space. But a move that clears a line can also leave a tall stack or trap empty cells beneath it. Those problems make later pieces harder to place.

The AI therefore judges both the immediate benefit and the shape of the board it leaves behind. Its goal is to clear as many lines as possible before the stack reaches the top.

2. Imagine possible landings

The AI starts with the settled blocks and the current falling piece. For each distinct orientation, it tries different columns and simulates dropping the piece until something stops it.

Placements that cross a wall, overlap a block, or leave part of the landed piece above the board are discarded. Each remaining placement produces a candidate board: a copy of the current board with that piece added. The actual game has not moved yet.

The AI measures each candidate before completed rows disappear. That lets it count the lines the placement would clear, along with the structure of the stack.

3. Describe a board with a few numbers

A feature is a numerical measurement of a board. Instead of judging every cell separately, the basic AI uses four measurements: height, completed lines, holes, and unevenness.

A column’s height is the distance from its highest filled cell to the bottom, including any holes beneath it. The weights will turn these measurements into a score in the next step.

4. Turn the measurements into a score

A weight controls how strongly a feature affects the decision. Completed lines have a positive weight. Height, holes, and bumpiness have negative weights, so increasing them lowers the score.

Score = −0.434 × height + 0.238 × lines − 0.247 × holes − 0.081 × bumpiness

For each candidate, multiply the four measurements by their weights and add the results. The highest score wins. Scores can all be negative: −8 is better than −10.

After choosing, the AI rotates and moves toward its target, then drops the piece. The game locks it, clears any full rows, and supplies the next piece. The same decision process starts again.

The advanced AI keeps the same search-and-score process but replaces the four measurements with eight. It distinguishes shallow holes from deeply buried ones, notices fragmented surfaces, and measures how much of the current piece contributes to cleared lines.

A transition is a change from empty to filled, or filled to empty. More transitions indicate a fragmented row or column. A well is different from a hole: it is open above, so a piece can still fall into it. Summing its successive depths gives deeper wells a larger penalty.

Each advanced measurement is multiplied by the signed weight shown above, and the terms are added just as in the basic scorer. More detail can help the AI distinguish boards that look equally good under the four basic measurements; it does not guarantee a better choice every time.

6. Understand the hints and the limits

With hints enabled, the outlined piece shows the landing preferred by the selected scorer. Switching features can change the hint because the same candidate boards are judged by different measurements and weights.

Comparison mode gives the basic and advanced AIs the same sequence of pieces. Their decisions produce different boards, and eventually different totals of cleared lines. A longer run or a different sequence can produce a different outcome.

This AI plans one placement at a time. It knows the current piece but has no preview of the next one. Its features estimate whether a board will be useful later; it does not simulate several future pieces to prove that.

It also searches straight-drop landings rather than every movement path. A desirable landing may become unreachable while the piece falls. The score is a practical guide to a good move, not a promise of the best possible game.