Which metrics matter most for a roguelike?

Runs, not sessions. A roguelike is a loop of short attempts, so the run is the unit to measure. Five numbers cover most questions: runs per player, run length, where runs end, the win rate and the unlock rate.

Read each one for new players and for returning players separately. A first run that ends in 4 minutes means something different from a 4 minute run by a player on their 50th attempt.

Five core roguelike metrics, how to compute each one and what it tells you.
MetricHow to compute itWhat it tells you
Runs per playerrun_start events per player, per day and in totalWhether the loop pulls players into another attempt
Run lengthMedian duration on run_end, by resultWhether a run fits the session players want
Where runs endrun_end by floor or stage, with the causeDifficulty spikes and unfair deaths
Win raterun_end with result win, divided by all run_endHow hard the full game is, by character and level
Unlock rateShare of players with each unlock, by run numberWhether meta progress arrives at a good pace

Sources: Manalens product documentation (events, properties and funnels).

What events should a roguelike send?

Four events give you all five metrics. Send one event per moment and put the details in properties. Do not create one event per floor or per boss, because then every report needs a long list of event names.

Give every run an id and a run number, so you can follow one run from start to end and compare the 1st run with the 20th. Add the build number to every event, so you can compare builds after each patch.

  • run_start, with run_id, run_number, character, difficulty and seed.
  • floor_reached, with run_id, floor and the time since run_start.
  • run_end, with run_id, result (win, death or quit), floor, cause, duration in seconds and run_number.
  • unlock, with the unlock id, run_number and total playtime, for each character, item or mode the player gets.

Sources: Manalens product documentation (event properties).

What does Slay the Spire teach about run data?

That run data finds balance problems that playtesters do not report. At GDC 2019, Mega Crit cofounder Anthony Giovannetti said the studio built its own metrics server. The game sent anonymous data at the end of every run, on death or victory, with the cards, relics, events, enemies, choices and run time.

He also showed how the data can mislead. Early on, two very active playtesters played far more runs than anyone else, so plain averages mostly measured those two people. Weight your averages per player, or cap the runs you count from each player.

The GDC Vault page says the game sold over 1 million copies in its first year after its Steam Early Access launch. The sequel kept the habit: Mega Crit wrote that Slay the Spire 2 had more than 25,000,000 runs attempted in its first week.

Sources: GDC Vault and GDC talk video (2019); Mega Crit Neowsletter (13 March 2026).

What win rate should a roguelike have?

There is no target, but public achievement data shows the shape. Steam shows the share of players who have each achievement. In Slay the Spire, the share that beat the game falls from 56.0% with the Ironclad to 26.0% with the Watcher. Only 14.9% completed Ascension 10 and 7.3% Ascension 20. In Balatro, 71.5% of players won a run and 12.2% won on Gold Stake.

These are shares of players, not of runs, and they count all the years since release. A large share that never wins the first time is normal. A curve with no hard top end, where nearly everyone clears everything, can mean that dedicated players run out of goals.

Share of players with each achievement on Steam, read 18 November 2026.
GameAchievementWhat it needsShare of players
BalatroAnte Up!Reach Ante 489.8%
BalatroHeads UpWin a run71.5%
BalatroHigh StakesWin on Gold Stake or higher12.2%
Slay the SpireRubyBeat the game with the Ironclad56.0%
Slay the SpireAmethystBeat the game with the Watcher26.0%
Slay the SpireAscend 10Complete Ascension 1014.9%
Slay the SpireAscend 20Complete Ascension 207.3%

Sources: Steam global achievements for Slay the Spire and Balatro (read 18 November 2026).

How do I measure unlock progression?

Plot the share of players who have each unlock against their run number. A roguelite needs a steady drip of new things, so this curve shows whether the drip is too slow or too fast.

If most players get the second character in run 3, the early game feels generous. If half of them quit before the first unlock, the wait is too long. Compare the curve with your runs per player: when a big drop in runs comes right before an unlock, move that unlock earlier and test the next build.

Balatro's developer, LocalThunk, told GameDiscoverCo that he used 1 hour retention as the measure of the hook, and that it rose as the game changed after the first public demo. For a roguelike, the first hour holds the first few runs and the first unlocks, so it is a good place to start.

Sources: GameDiscoverCo (18 March 2024); Manalens product documentation (retention).

Where do I see runs, deaths and unlocks?

Manalens is the analytics and growth engine built for games. Send run_start, run_end and unlock from the Unity, Godot or web SDK. Build a funnel from run_start to each floor, and compare D1 and D7 retention by build.

Sources: Manalens product documentation.

Frequently asked questions

What is the most important metric for a roguelike?

Runs per player in the first hour and the first day. It shows whether the loop pulls players into another attempt. Read it next to where runs end, because a fast drop after an unfair death can hide a good loop. Balatro's developer used 1 hour retention as the measure of the hook.

What is a good win rate for a roguelike?

There is no standard rate. Steam achievement data gives reference points: 71.5% of Balatro players won a run, 56.0% of Slay the Spire players beat the game with the Ironclad, and only 7.3% completed Ascension 20. Aim for a first win most players can reach, with harder goals after it.

Should I count runs or sessions?

Count both, but report runs. A session can hold many runs or part of one. Send run_start and run_end with a run id, and the result, floor, cause and duration. Then run length and win rate come from run_end, and you can still read session counts beside them.

Why can run data mislead me?

Because a few players can play most of the runs. Mega Crit found that two very active playtesters dominated early Slay the Spire averages. Weight averages per player, split new and experienced players, and check where in a run an item appears before you call it too strong.

Sources and implementation