How Playrix Measured Gardenscapes Level Difficulty Before and After Release

Elderly Gardenscapes-style character holding a ruler and stopwatch beside a board comparing planned and actual difficulty curves across levels.

How Playrix measured Gardenscapes level difficulty in 2017 using internal testing, player data, failure rates, retention metrics and A/B testing.

When a Gardenscapes level feels ridiculously difficult, the discussion can quickly come back to the same questions: Was the board bad? Were there too few moves? Was it just bad luck? Or did the game somehow “want” the player to lose?

But that is not what this article is about.

The more interesting question is this: how did Playrix itself know how difficult a level really was?

In 2017, Alexander Shilyaev, then a leading game designer at Playrix responsible for the match-3 side of Gardenscapes, described a process showing that difficulty was not simply a number of moves chosen by a level designer. There was a predetermined difficulty curve, internal testing, real player data collected after release and, when necessary, A/B testing of different versions of the same level.

This is a historical description of the process used by the Gardenscapes team in 2017. It does not confirm that Playrix uses exactly the same system in 2026. It does, however, give us a rare look at how the company measured something that can feel completely subjective to a player: how difficult a level actually is.

Difficulty Was Planned Before the Level Was Built

According to Shilyaev, before a level designer started building a specific board, there was already a broader plan for how Gardenscapes levels were designed to work together.

Playrix used what he described as a project difficulty curve: a predetermined curve defining how difficulty should rise and fall as players progressed through the game.

A level was therefore not designed in isolation. It was not enough for it to simply be “hard” or “easy.” It had to fit the intended difficulty of that particular point in the overall sequence.

The team used a document defining requirements for each level, including its intended difficulty, objectives and the elements that should be used. The designer was given a framework within which the puzzle had to be created.

The Designer Could Adjust Difficulty — But Could Not Know It for Certain

Shilyaev explained that designers could adjust different parameters to bring a level closer to its intended difficulty. The number of moves was an obvious one, but other parts of the level could also affect how difficult it felt in practice.

The key point is that this was still an estimate.

The person who creates a level knows it far better than an ordinary player. The designer understands its structure, knows which areas of the board matter most and has already played through multiple versions of it.

That is why Playrix did not rely solely on the designer's judgment.

A Separate Team Played the Levels Like Regular Players

Shilyaev's presentation describes a dedicated testing department whose job was to play levels as closely as possible to the way real users would play them.

The designer had created the level according to its intended position on the difficulty curve. Testers then played it repeatedly, and their results were used to make final adjustments before release.

Whether the designer could beat the level was not enough. What mattered was how people who had not created it actually performed.

Internal Testing Was Still Not Enough

Even after design and testing were complete, the process was not over.

Shilyaev made it clear that work on a level could continue after release because only real players could reveal how it behaved on a much larger scale.

The designer had one estimate. The testing team provided another. But once thousands of players reached the level, Playrix had something much more valuable: real-world data.

The First Major Analysis Happened 2–3 Weeks After Release

According to the 2017 presentation, Playrix began its first major analysis of a newly released set of levels about two to three weeks after release.

The delay gave the team time to collect enough player data. As users played, the game sent events to Playrix's analytical database, allowing the company to compare what had been predicted during testing with what actually happened.

Designed difficulty and actual difficulty were not necessarily the same thing.

Playrix Compared Two Difficulty Curves

The team compared the predetermined project difficulty curve with the curve created by real player results.

If both followed roughly the same pattern, the levels were performing close to expectations. If they began to drift apart, something needed to be investigated.

A level designed for a certain degree of difficulty could turn out to be significantly harder or easier once real players encountered it.

Instead of relying on individual complaints or impressions, Playrix could see the difference in the overall data.

What Did “Difficulty” Mean in Playrix's Data?

Shilyaev gave the term a specific statistical meaning.

The difficulty metric was calculated as the ratio of failed attempts to the total number of attempts.

So figures such as 80%, 90% or 95% in this historical material were not arbitrary ratings assigned by a designer. They described how often attempts ended in failure.

This allowed the team to place a level on an actual difficulty curve based on player behavior rather than intuition alone.

The Team Also Looked at Sequences of Difficult Levels

One of the problems Shilyaev described was not a single extreme level but a sequence of very difficult levels.

Even if each level individually remained within acceptable limits, several high-difficulty levels appearing back to back could make the overall experience much more frustrating.

The presentation gives examples of consecutive values such as 92%, 93% and 95% difficulty.

The concern was not simply that players would lose repeatedly. It was that after finally beating one difficult level, they could immediately run into another, then another.

According to Shilyaev, sequences like these could discourage players enough to make them leave the game.

Difficulty Was Only One Part of the Picture

Playrix was not looking at failure rates alone.

Shilyaev identified four major categories of metrics:

Difficulty: failed attempts compared with the total number of attempts.
Dumps: 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: data such as conversion and coins spent while trying to complete the level.

A high failure rate did not automatically mean a level had to be changed. The team also wanted to know whether players kept trying, eventually won or stopped playing altogether.

When Players Stopped Playing, That Mattered More

The “dumps” metric was particularly important because it distinguished between a player who kept losing and a player who stopped coming back.

A player could lose ten times and continue trying. That was very different from reaching a level, failing to complete it and then showing no activity for seven days.

The question was therefore not just “How many players lose here?” but also “How many players stop here?”

Playrix Could Test Different Versions of the Same Level

When the data identified a problem, Playrix could use A/B testing to decide what to change.

The process was more sophisticated than simply adding a few extra moves.

Players could be divided into three groups:

one group played the existing version,
another played an easier variation,
and a third could receive a completely redesigned version.

Playrix then collected enough data to compare how the different versions affected difficulty, player retention and monetization.

This Does Not Prove That Every Difficult Level Is an Intentional Paywall

Shilyaev's presentation clearly shows that Playrix monitored monetization alongside difficulty, attempts and player drop-off.

But it does not prove that every difficult Gardenscapes level was designed with the sole purpose of forcing players to spend money.

The Level 93 example shows a more complicated tradeoff: reducing player drop-off could improve one part of the experience while producing weaker monetization results elsewhere.

What the presentation reveals is a system built around balancing several metrics at once, rather than evidence that the game has to rig difficult levels to create pressure.

Why Real Players Could Produce Different Results

A designer could build a level to match the intended difficulty curve, and an internal testing team could confirm that it appeared to be in the right range.

Yet the level could still behave differently after release.

Real players had different skill levels, habits, resources and ways of approaching the board. Once the player pool expanded far beyond an internal testing group, patterns could emerge that testing simply could not predict.

That is why post-release analysis was a regular part of the process described by Shilyaev rather than something used only when a level went badly wrong.

Difficulty Became Data After Release

The most revealing part of Shilyaev's presentation is the full cycle Playrix had built:

planned difficulty curve → level creation → internal testing → release → real player data → comparison → possible changes.

At least in the Gardenscapes system described in 2017, a level was not necessarily finished the moment players received it.

Before release, difficulty was largely a prediction. After thousands of real attempts, it became measurable behavior.

That may be the clearest takeaway from the entire presentation: before release, the difficulty curve was a plan; after release, it became data.

Source

This article is based on the written version of a presentation by Alexander Shilyaev, then a leading game designer at Playrix, at White Nights St. Petersburg 2017: Playrix: Experience Creating Levels and Elements for Match-3.

Join the Gardenscapes Strategy Community

If you enjoy discussing Gardenscapes levels, events, teams, game mechanics, and updates, you can join the Gardenscapes Strategy Facebook community and connect with other players.

Join the Gardenscapes Strategy Facebook Group