# Deckbuilder metrics: card pick rates, deck win rates and balance data

> Which data to collect in a deckbuilder: card offers, picks, skips and run results, how to read card win rates without bias, and what Slay the Spire learned.

By Yos Riady. Updated 2026-11-18. Published 2026-11-18. Canonical URL: https://manalens.dev/blog/deckbuilder-card-metrics-win-rates

Track every card offer, pick and skip, and every run result with the final deck. Then compare a card's win rate with runs that reached the same floor, not with all runs. Mega Crit balanced Slay the Spire this way, and players attempted more than 25,000,000 runs of the sequel in its first week.

## Key points

- Send card_offered, card_picked and run_end with the floor, so you can compute pick rates and corrected win rates.
- Compare a card's win rate with all runs that reached the same floor, because late cards only appear in strong runs.
- On Steam, 71.5% of Balatro players won a run, but only 12.2% won on Gold Stake or higher.

## Which metrics matter for a deckbuilder?

The run metrics of any roguelike, plus card metrics. Runs per player, run length and win rate show whether the game holds players. Card metrics show whether each card has a place. Slay the Spire's goal, as Anthony Giovannetti put it at GDC 2019, was not that every card is equally good, but that every card is reasonable to take some of the time.

- [Roguelike metrics: run length, completion and unlocks](https://manalens.dev/blog/roguelike-game-metrics)

*Card metrics for a deckbuilder, how to compute each one and what it tells you.*

| Metric | How to compute it | What it tells you |
| --- | --- | --- |
| Offer rate | Times a card is offered, per run | Whether players see the card at all |
| Pick rate | Times picked, divided by times offered | How much players want the card |
| Skip rate | Card rewards skipped, divided by rewards offered | Whether rewards feel worth taking |
| Win rate with the card | Wins with the card in the final deck, divided by runs with the card | How strong the card seems, before correction |
| Deck size at run end | Median cards in the final deck, by result | Whether players thin or grow their decks |

Sources: GDC talk video (2019).

## What events should a deckbuilder send?

Send one event per choice and keep card names in properties. Card games have hundreds of cards, so one event per card would make reports impossible to read.

Add the floor or act to every event. You need it to correct win rates later. Add the character or deck, the difficulty level and the build number, so you can compare balance patches.

- card_offered, with run_id, floor, the card ids offered and the source (combat, shop or event).
- card_picked, with run_id, floor and the card id, or skipped when the player takes nothing.
- card_removed and card_upgraded, with run_id, floor and the card id.
- combat_end, with run_id, floor, enemy, turns, damage taken and deck size.
- run_end, with run_id, result, floor, character, difficulty and the final deck as a list of card ids.

- [Game tracking plan generator](https://manalens.dev/tools/game-tracking-plan-generator)

Sources: Manalens product documentation (event properties).

## Why can a card win rate mislead?

Because a card that appears late only shows up in runs that already got far. Giovannetti gave the example of the card Madness. It appeared in winning decks far more than expected. The reason was not its power: most players got it from an event late in the third act, just before the boss, so only strong runs ever had it.

The fix is to compare like with like. For each card, take the floor where it was picked. Compare the win rate of runs that picked it there with the win rate of all runs that reached that floor. The gap between the two is the card's real effect, as far as the data can show it.

He also warned that a few very active players can make up most of your runs. Weight the data per player, or cap the runs from any one player, before you nerf a card.

1. For each card, find the floor of each pick from card_picked.
2. Find all runs that reached that floor, from floor_reached or run_end.
3. Compute the win rate of those runs, and the win rate of the runs that picked the card there.
4. Compare the two, by character and difficulty, with at least a few hundred runs in each group.
5. Read the gap next to the pick rate and player comments before you change the card.

Sources: GDC talk video (2019).

## How do I find cards that nobody picks?

Sort cards by pick rate when offered. A card that is picked much less often than others of the same rarity is either weak or unclear. A card that is almost always picked may be too strong, or may be the only answer to a common problem.

Then look at the skip rate. Slay the Spire lets players skip card rewards, and skipping is a valid choice, because a thin deck is often stronger. If the skip rate climbs in a new build, the new cards may not fit the decks players build.

Watch out for cards that only fit one strategy. Giovannetti called them build-around cards: weak alone and strong in the right deck. Read their win rate inside the decks they were made for.

Sources: GDC talk video (2019).

## How hard should a deckbuilder be?

Most players should win once, and few should win at the top level. Steam achievement data shows the shape. 71.5% of Balatro players won a run, 30.2% won on Black Stake or higher and 12.2% on Gold Stake. In Slay the Spire, 56.0% beat the game with the Ironclad, 44.0% with the Defect, and 7.3% completed Ascension 20.

These are shares of players over the life of each game, so a new game will start lower. Use your own data in the same way: the share of players who have a first win by their 10th run, and the share who reach each difficulty level.

*Share of players with each achievement on Steam, read 18 November 2026.*

| Game | Achievement | What it needs | Share of players |
| --- | --- | --- | --- |
| Balatro | Heads Up | Win a run | 71.5% |
| Balatro | Low Stakes | Win on Red Stake or higher | 43.6% |
| Balatro | Mid Stakes | Win on Black Stake or higher | 30.2% |
| Balatro | High Stakes | Win on Gold Stake or higher | 12.2% |
| Slay the Spire | Ruby | Beat the game with the Ironclad | 56.0% |
| Slay the Spire | Sapphire | Beat the game with the Defect | 44.0% |
| Slay the Spire | Ascend 20 | Complete Ascension 20 | 7.3% |

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

## Where do I read pick rates and win rates?

Manalens is the analytics and growth engine built for games. Send card_offered, card_picked and run_end from the Unity, Godot or web SDK. Filter by build, character and floor, and query the events with SQL for floor-corrected win rates.

- [Game tracking plan generator](https://manalens.dev/tools/game-tracking-plan-generator)

Sources: Manalens product documentation.

## Frequently asked questions

### How do I measure if a card is too strong?

Compare the win rate of runs that picked the card at a floor with the win rate of all runs that reached that floor. A plain win rate with the card misleads, because late cards only appear in runs that already went far. Mega Crit saw this with the card Madness in Slay the Spire.

### What is a good card pick rate?

There is no single good rate. Compare each card with others of the same rarity and cost. A card picked far less often is weak or unclear, and one picked almost always may be too strong. Mega Crit aimed for every card to be reasonable to take some of the time.

### How many runs do I need before I change a card?

Enough that one or two players do not decide it. Mega Crit found two very active playtesters dominated its early Slay the Spire data. Weight results per player, and wait for a few hundred runs that reached the floor where the card appears before you act on a win rate.

### What share of players should win a deckbuilder?

Steam achievements give reference points: 71.5% of Balatro players won a run, and 56.0% of Slay the Spire players beat the game with the Ironclad. Only 12.2% of Balatro players won on Gold Stake. Aim for a reachable first win and a long tail of harder goals.

## Sources

- [GDC: Slay the Spire: Metrics Driven Design and Balance, talk video (published 26 June 2019)](https://www.youtube.com/watch?v=7rqfbvnO_H0)
- [GDC Vault: 'Slay the Spire': Metrics Driven Design and Balance (Anthony Giovannetti, Mega Crit, GDC 2019)](https://www.gdcvault.com/play/1025731/)
- [Mega Crit: The Neowsletter, March 2026 (13 March 2026)](https://www.megacrit.com/news/2026-3-13-neowsletter-issue-20/)
- [Steam global achievements: Balatro (read 18 November 2026)](https://steamcommunity.com/stats/2379780/achievements)
- [Steam global achievements: Slay the Spire (read 18 November 2026)](https://steamcommunity.com/stats/646570/achievements)

## Related guides

- [Game marketing measurement guide](https://manalens.dev/blog/game-marketing-measurement-guide)
- [Deckbuilder games on Steam: market data](https://manalens.dev/blog/deckbuilder-games-steam-market-data)
- [Deckbuilder demos, Next Fest and launch](https://manalens.dev/blog/deckbuilder-demo-next-fest-launch)
- [Roguelike metrics: run length, completion and unlocks](https://manalens.dev/blog/roguelike-game-metrics)
- [Game tracking plan generator](https://manalens.dev/tools/game-tracking-plan-generator)
- [Deckbuilder game guides](https://manalens.dev/blog/topic/deckbuilder-games)
- [Steam achievement data: see where players quit](https://manalens.dev/blog/steam-achievement-data-player-drop-off)
- [Game heatmaps and death maps for level design](https://manalens.dev/blog/game-heatmaps-death-maps)
- [First session funnel and FTUE analysis](https://manalens.dev/blog/first-session-funnel-ftue)
- [Steam achievements and trading cards: do they lift sales?](https://manalens.dev/blog/steam-achievements-trading-cards)
