The Gardenscapes A/B Tests That Changed Levels After Release

Two players compare different Gardenscapes board layouts while both screens show Level 93, illustrating A/B testing and alternate versions of the same level.

Discover how Playrix used A/B testing, player behavior, retention data, and alternate level versions to change Gardenscapes levels after release.

A Gardenscapes level was not necessarily finished once it reached players. At least during the game’s earlier years, Playrix continued monitoring levels after release and used real player behavior to determine whether the original design was producing the intended results.

When the data revealed a problem, Playrix could do more than make a minor adjustment. The company could create multiple versions of the same numbered level, give those versions to different groups of players, and compare the results. In that system, difficulty, move limits, and board logic were not isolated choices but parts of the same broader design process.

This gives us a rare look at something players normally never see: how different versions of a Gardenscapes level could be tested after the level was already live.

Testing Continued After a Level Was Released

Before release, each level went through design and testing intended to place it within a planned difficulty curve. Moves, board settings, and other parameters could be adjusted until the level performed roughly as expected.

But internal testing could only go so far.

Once thousands of real players reached the same level, Playrix had access to a much larger data set. The company could compare the intended difficulty with what was actually happening during real gameplay.

If the results differed significantly from expectations, the level could become a candidate for further testing.

What Playrix Measured

The historical process described by Playrix relied on several types of player data rather than looking at difficulty alone.

Difficulty: the relationship between failed attempts and total attempts on the level.
Player drop-off: players who had not completed the level for seven days and had shown no other activity in the game.
Attempts to win: how many attempts players needed before completing the level.
Monetization metrics: including purchases made while players were on the level and the amount of coins used to complete it.

This distinction matters. A high failure rate did not automatically mean that a level needed to be changed. Difficult levels are part of the structure of Gardenscapes.

The bigger question was what that difficulty caused players to do.

If players kept trying and eventually moved forward, the level might still be working as intended. If too many players stopped progressing and left the game, the same level could become a much more serious problem.

The Three Test Groups

When Playrix wanted to test a possible change, players could be divided into three groups, with each group receiving a different version of the same numbered level.

Group 1: played the existing version of the level.
Group 2: played an easier version of that level.
Group 3: played a substantially redesigned version.

The original level served as the control. The other groups allowed Playrix to see what happened when the difficulty or the design itself changed.

Players did not need to know that another version existed. Two people could therefore reach the same level number while seeing different boards as part of the test.

After enough data had been collected, Playrix could compare the versions and decide which one produced the strongest overall results.

Level 77: When Keeping Players Was More Important

Level 77 is one of the clearest documented examples of this process.

The original version had a player drop-off rate of approximately 2.29%. Because Level 77 appeared very early in the game, losing players at that point was especially important.

A player who quits at Level 77 does not simply fail to complete one puzzle. That player may never reach hundreds or thousands of later levels, additional game systems, events, or other parts of Gardenscapes.

Playrix therefore tested two alternatives alongside the original level: an easier version and a substantially redesigned version.

The results created an interesting tradeoff.

The redesigned version reduced player drop-off more effectively, meaning more people continued playing. However, some of its monetization results were weaker.

The easier modification of the original level performed better on some financial measurements but did not solve the player drop-off problem as effectively.

Playrix ultimately selected the redesigned version.

At this particular point in the game, retaining more new players was considered more valuable than preserving the stronger short-term monetization results of the other version.

Level 93: A Different Test Produced a Different Decision

Level 93 shows why these tests were not simply about finding the easiest possible board.

Again, Playrix tested different versions because player drop-off was too high.

The two Level 93 videos below make the differences easy to see. In the first version, the board has 9 moves, a goal of 6 lemonade glasses, a more open layout, water through the center, and a different distribution of chains.

In the second version, the same Level 93 has 10 moves and a goal of 5 lemonade glasses. The board is much more compact, contains far more chains, and uses a completely different layout.

The substantially redesigned version performed very well at keeping players in the game. Fewer players stopped progressing at Level 93.

But there was a significant downside: monetization metrics dropped too far.

An intermediate version produced a different balance. It reduced player drop-off while avoiding such a large decline in the other metrics being monitored.

This time, Playrix chose the intermediate version rather than the completely redesigned one.

The comparison between Levels 77 and 93 is important because there was no single rule such as “always choose the version with the highest completion rate.” The preferred version depended on the overall behavior it produced.

The Tests Were About More Than Difficulty

If the only objective had been to make a difficult level easier, Playrix could simply have added enough moves until most players passed it.

The testing process shows that the decision was more complicated.

Changing a level could affect how many times players attempted it, whether they spent coins, whether they made purchases, how quickly they progressed, and whether they continued playing Gardenscapes at all.

A version that improved one metric could make another worse.

That is why comparing several real versions against actual player behavior after release was more useful than assuming that one particular change would automatically improve the level.

Different Players Could See Different Versions of the Same Level

For players, one of the most interesting consequences of this system is that a level number did not always identify one permanent board.

During a test, different groups could receive different versions at roughly the same time.

After the test, the selected version could replace the previous one while keeping the same level number.

This means an old screenshot or YouTube video does not necessarily show the only version that ever existed.

A video might show more or fewer moves, a different arrangement, or a board that looks substantially different from the version another player remembers.

In documented historical testing, that kind of difference was possible without changing the level number.

Why Consecutive Difficult Levels Also Mattered

Playrix also looked beyond individual levels. The overall difficulty curve mattered because several extremely difficult levels appearing close together could create a very different experience from the same levels spread farther apart.

A player might accept one unusually difficult puzzle. Reaching another immediately afterward, followed by another, could create enough accumulated pressure to affect retention.

This meant the data could reveal problems that were not obvious when each level was examined separately.

The position of a level within the larger progression was therefore part of the analysis, not just the design of the board itself.

What This Tells Us About Old Gardenscapes Levels

The historical examples confirm something important: released Gardenscapes levels were not always treated as permanent.

Real player data could lead Playrix to create alternative versions, compare those versions across different groups, and ultimately replace an existing board.

That makes comparisons between old and newer gameplay especially interesting. When an older video does not match a later version of the same level, a redesign is one possible explanation.

It also means that one player’s experience cannot always be used to prove exactly what every other player saw at the same level.

What This Does Not Prove About Gardenscapes Today

There is an important limitation to this evidence.

The detailed process described here comes from Playrix material presented in 2017, when Gardenscapes was a much smaller game with far fewer levels than it has today.

It confirms that Playrix used player analytics, alternative level versions, and test groups in Gardenscapes during that period.

It does not prove that the company uses exactly the same three-group structure, the same metrics, or the same decision-making process in 2026.

The game has expanded enormously since then, and its development systems may have evolved as well.

What the Level 77 and Level 93 Tests Really Show

The most revealing part of these examples is not simply that Playrix made levels easier.

It is that the company tested competing versions of the same level against real player behavior and sometimes made different decisions depending on what the results showed.

Level 77 favored the substantially redesigned version because player retention at that early point was critical.

Level 93 favored a more moderate version because it produced a better balance across the measurements Playrix was watching.

In both cases, the level number remained part of the player’s normal progression even though the board behind the same level number could change.

Conclusion

The Gardenscapes A/B tests provide a rare look behind the level-selection screen.

A released level could continue generating data, alternative versions could be created, different groups of players could test them, and the results could determine which version remained in the game.

For some Gardenscapes levels, release was not the end of the design process. It was the beginning of the largest test.

So the next time an old video looks nothing like the level on your screen, do not be too quick to blame the uploader. The level number may have stayed put while the board quietly moved on.