Gridjelly
How to Survive Longer in 8×8 Block Puzzles: Tips Tested on 1,800 Bot Games
An 8×8 block puzzle has rules simple enough that it's often hard to see why you lost. So we took Gridjelly's actual game code and had six bots, each with a different way of thinking, play the same 300 games — 1,800 games in total. Because each bot clearly weighs some things and ignores others, the differences between them answer a practical question: what actually keeps a game alive?
How the game is built
The board is 8×8 and you get three blocks at a time. Blocks can't be rotated; any full row or column clears. You get a new set only after placing all three, and the game ends when none of the blocks left in your hand fits anywhere on the board. There are 29 block shapes, each with a weight that sets how often it appears. Of the total weight of 92, big blocks of five or more cells (3×3, 2×3, 3×2, the five-long bars and the big L shapes) account for 19%. The 3×3 square has a weight of 1, making it the rarest piece.
The game also quietly helps you in a few ways (in Classic mode):
- Every new set contains at least one block you can place right now. If not, it redraws up to 20 times, and failing that it slips in a single-cell block.
- When the board is 60% full or more, blocks of three cells or fewer become twice as likely.
- 70% of the time it insists on a set where all three blocks can be placed (with up to six redraws).
In the Daily Challenge everyone has to get the same sequence of blocks, so only the first guarantee stays on; the rest are switched off. That's a good reason to play a little more carefully in the daily.
Six bots and the results
Every bot chooses where to put one block at a time. What differs is how it judges a placement:
- Place anywhere — a random legal spot.
- Chase lines only — the spot that clears the most lines, preferring edges on ties.
- Full board eval — scores the board after the move: cleared lines are good; filled cells and one-cell holes boxed in on all sides are bad; more room for a 3×3 or a five-long bar is good. It's the same evaluation the developers use for testing.
- Board eval minus "room for big blocks" and minus "avoid holes" — each drops one term so we can see what that term is worth.
- Board eval + plan all three — considers spots from best to worst, but only accepts one if every remaining block in the set can still be placed afterwards.
Placing at random ends after 15.2 blocks on average. Chasing lines gets 40.9, and full board evaluation gets 47.3 (610 points on average). The bot that plans all three blocks lasted 165.9 — 3.5 times as long as the plain board-evaluation bot — and its average score rose to around 2,700. It uses exactly the same judgment of a spot; the only addition is checking "will the other two still fit?"
Tip 1: Look at all three blocks before placing the first
This made the biggest difference of anything we tested. When game over happens explains why. Every new set is guaranteed to contain something placeable, so games don't end when a set arrives — they end in the middle of a set. Of the board-evaluation bot's 300 games, 268 ended with one block left in hand and the rest with two. The earlier blocks went into spots that looked good on their own and destroyed the only place the last block could go.
In practice:
- Decide where the most awkward block in the set (3×3, 2×3, big L) will go first, then fit the others around it.
- Don't let a small block grab a good spot early. Small blocks will fit almost anywhere at the end.
- If a move clears lines, check whether the remaining blocks fit on the board after the clear. Clears open space, so changing the order can make a block fit that otherwise wouldn't.
Tip 2: Always keep one 3×3 space open
Remove only the "room for big blocks" term from the board evaluation and survival drops from 47.3 blocks to 26.3 — 44% shorter. Removing "avoid holes" instead costs just 7% (down to 43.9). In this game, what keeps you alive isn't so much avoiding holes as keeping wide open ground where big blocks can land.
The blocks that ended games make the point even more clearly. The 3×3 square was left in hand at game over 11.2 times more often than its frequency would suggest. The 3×3, 3×2 and 2×3 blocks together make up only 5% of the blocks dealt, yet they were 34% of the blocks stuck in hand at game over. Single cells and two-cell bars never ended a game at all.
The five-long bars are the interesting case. By length they look like the hardest pieces, but for this bot they ended games only about as often as they appeared (around ×1). That's because the bot deliberately saves room for them. For the bot with that term removed, the five-long bars climbed up the list too. So the lesson isn't "big blocks are dangerous" — it's "big blocks are dangerous when you don't save room for them." The 3×3 stayed at the top even though the bot saves room for it as well, which shows how hard it is to keep nine empty cells together in one piece.
Practical habits:
- Keep one patch of the board where a 3×3 fits whole, and don't scatter small blocks into it.
- Keep a straight run of five empty cells, horizontal or vertical, somewhere too — that's where a five-long bar goes.
- The common habit of filling from the edges and corners first makes it easier to keep your open ground in one place.
Tip 3: Don't just chase lines
The bot that always clears a line when it can (40.9 blocks) lost earlier than the bot that looks at the whole board (47.3). Clearing lines is good, but if a clear leaves an awkward shape behind, another spot may be better. Breaking up your 3×3 space just to clear one line is usually a bad trade.
Tip 4: You can lose with half the board empty
When the board-evaluation bot lost, the board was on average only 55% full. Losing with nearly half the board empty happens because the empty cells are scattered in pieces that don't match the shapes in your hand. Rather than "there's plenty of room left," ask "what shape is the empty space in?" The shape of your open ground matters more than the count of empty cells.
Tip 5: Avoid one-cell holes, but don't obsess
A one-cell hole boxed in on all sides can only be filled by a single-cell block. It's worth avoiding, but in the data its value (7%) is far smaller than saving room for big blocks (44%) or planning all three blocks (3.5×). Don't break up a wide open area just to avoid a hole. Do remember, though, that the row and column running through a hole can't clear until that cell is filled.
How to read bot numbers. The turn counts and scores here are bot results, not a measure of human skill. The bots look only one move ahead (even the planning bot only looks within the current set) and follow fixed rules, so focus on the differences between bots rather than the absolute numbers. Every bot played the same 300 games (seeds 1–300, Classic mode, difficulty adjustments on), and rerunning the analysis script reproduces the same values. For how points are calculated, see How Gridjelly Combo Scoring Works; for the full rules, see How to play Gridjelly.