# Game heatmaps and death maps for level design

> Send player position with death and quit events, count them per grid cell and draw a heatmap over the level to find where players struggle or get lost.

By Yos Riady. Updated 2026-11-18. Published 2026-11-18. Canonical URL: https://manalens.dev/blog/game-heatmaps-death-maps

A death map counts where players die, per grid cell, and shows it over the level. In Tomb Raider: Underworld, a study of about 28,000 players found up to 3,050 deaths in one cell of a single map unit (Drachen and Canossa, 2009). Send position with each death, count by cell, then fix the hot spots.

## Key points

- In one Tomb Raider: Underworld level, 53% of 50,060 deaths came from falling and only 3.6% from enemies.
- Send x and y as two numbers with each death event, then count events per grid cell of about 5 by 5 meters.
- Count players as well as deaths per cell, because one stubborn player can die 40 times in one spot.

## What is a game heatmap?

A heatmap is a count of events per area of a level, drawn as colour over the map. A death map counts deaths. A position map counts where players are at fixed times. Both show where the level works against you.

Bungie used both for Halo 3. WIRED reported that Microsoft researchers analyzed more than 3,000 hours of play by some 600 players. A death map of the multiplayer level Valhalla showed deaths skewing toward one base, so designers changed the terrain. In the Jungle level, dots of player position every 5 seconds showed testers wandering, until a terrain change kept them moving forward.

Sources: WIRED (21 August 2007).

## What does a death map tell me that a death count does not?

Where the deaths happen, and whether one cause or many cause them. A count per level tells you a level is hard. A map tells you which jump or which fight makes it hard.

The Tomb Raider: Underworld study shows the difference. In one Arctic level, 53% of 50,060 deaths came from falling, 32.15% from traps and only 3.6% from enemies. The death map of the Valaskjalf map unit then showed four hot spots. One cell of about 5 by 5 meters held deaths from a Thrall attack and a tricky jump. Another held mostly trap deaths. Each spot pointed to a different fix.

*Causes of death in one Tomb Raider: Underworld level, from Drachen and Canossa.*

| Cause | Deaths | Share |
| --- | --- | --- |
| Falling | Not given as a count | 53% |
| Traps | 16,091 | 32.15% |
| Enemies | 1,789 | 3.6% |
| All causes | 50,060 | 100% |

Sources: Drachen and Canossa (2009); Drachen and Canossa, Evaluating motion (IJART).

## What should a position event contain?

The event name, the level, the x and y position, and the cause. Add z for 3D levels with floors. Round positions to whole units, because you will group them into cells anyway.

Send position with events that matter: deaths, quits, failed attempts, hint requests and stuck checks. Do not send position every frame. A death event is enough for a death map, and a timed sample, such as every 30 seconds, is enough for a "where are players lost" map.

Keep property values flat. In Manalens, event properties are strings, numbers, booleans or null, so send x and y as two numbers, not as one vector.

``` Godot
var p = player.global_position
Manalens.track("player_death", {"level": 3, "x": round(p.x), "y": round(p.y), "cause": "fall", "attempt": attempt})
```

Sources: Manalens track event documentation; Godot SDK documentation.

## How do I turn events into a heatmap?

Group the events into grid cells, count them, then colour each cell. A cell of a few meters, or a few tiles, is a good start. The Tomb Raider study used cells of about 5 by 5 meters.

In Manalens, run the query in the Explorer. It reads raw_events with ClickHouse SQL and returns up to 200 rows, so sort by count and keep the top cells. Save it as a Table chart on a dashboard and click Export CSV. Then draw the cells over a screenshot of the level, in your engine or a spreadsheet. There is no built-in heatmap chart, so the drawing step is yours.

Count players as well as deaths. One stubborn player can die 40 times in one spot. If deaths are high and players are few, the spot is hard for a few. If both are high, it is a wall for everyone.

``` Explorer SQL: deaths per 5 by 5 cell
SELECT
  JSONExtractRaw(properties, 'level') AS level,
  intDiv(toInt64(JSONExtractFloat(properties, 'x')), 5) * 5 AS cell_x,
  intDiv(toInt64(JSONExtractFloat(properties, 'y')), 5) * 5 AS cell_y,
  uniqExact(message_id) AS deaths,
  uniqExact(anonymous_id) AS players
FROM raw_events
WHERE event = 'player_death'
  AND toDate(original_timestamp) BETWEEN {{date_from}} AND {{date_to}}
GROUP BY level, cell_x, cell_y
ORDER BY deaths DESC
LIMIT 200
```

Sources: Manalens Explorer and dashboards documentation; Drachen and Canossa (2009).

## How do I read the hot spots?

Compare each hot spot with what you designed. A boss room should be hot. A corridor should not. A spot that is much hotter than the spots around it is the first suspect.

Shadow Blade, a Unity platformer, did this with deaths drawn over the level in the editor. The team looked for points with a huge jump in death count compared with the rest of the level, and balanced those challenges for a smoother pace.

Then split the hot spot by cause. One cause in one spot usually means one fix: widen a jump, move a trap, or slow an enemy. Many causes in one spot can mean too much at once.

*Hot spot patterns and a first fix to test.*

| Pattern | Likely meaning | First fix to test |
| --- | --- | --- |
| Many deaths, one cause | One hard element | Tune that element |
| Many deaths, many causes | Too much at once | Remove one threat |
| Many deaths, few players | A few players stuck | Add a checkpoint or hint |
| Many quits, few deaths | Players lost or bored | Clearer path or signpost |

Sources: Game Developer, Closing the loop (2 February 2014); Drachen and Canossa (2009).

## Which other maps are worth making?

Quit maps, hint maps and position samples. Each one uses the same grid query with a different event name.

A quit map counts where players were when they closed the game or went back to the menu. It finds places that are dull or confusing, not only deadly. A hint map counts requests for help. In the Tomb Raider: Underworld data, the area with the most causes of death also had the most help requests, which made it a strong candidate for a challenge review.

A position sample every 30 or 60 seconds shows flow. Halo 3 researchers made snapshots of where players were at five minutes, one hour and eight hours. If players move too fast, the game may be too easy. If they are spread across the map, they may be lost.

Sources: Drachen and Canossa, Evaluating motion (IJART); WIRED (21 August 2007).

## How do I check that a fix worked?

Compare the same cells before and after the patch, per player. Filter the query by build. If deaths per player in the hot cell fall and the level completion rate rises, the fix worked. If deaths only move to the next cell, the wall moved with them.

Halo 3 testing ran on normal difficulty, because the research aimed at the average to below-average player. Do the same: compare players on the same difficulty setting, or add difficulty as a property and split by it.

Sources: GameSpot, GDC 2009 (26 March 2009); Manalens app events documentation.

## Where can I count deaths per cell?

Manalens is the analytics and growth engine built for games. Send player_death with level, x, y and cause from the Unity or Godot SDK, then group by cell in the Explorer and export the table as CSV. Plans are priced by monthly active players, so a death event per attempt adds no event bill.

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

Sources: Manalens product documentation.

## Frequently asked questions

### Do I need a heatmap tool to make a death map?

No. You need a position on each death event, a query that counts events per grid cell, and an image of the level. Export the counts as CSV and draw them over the image in your engine or a spreadsheet. A dedicated tool only saves the drawing step.

### How big should a heatmap grid cell be?

Start with a cell about the size of one challenge, such as one jump or one room. The Tomb Raider: Underworld study used cells of about 5 by 5 meters. Smaller cells spread the counts thin. Larger cells hide which element causes the deaths.

### Should I track position every frame?

No. Send position only with events that matter, such as deaths, quits, failed attempts and hint requests. Add a timed sample, such as every 30 seconds, if you want to see where players get lost. Every frame adds noise and data without a better answer.

### How many players do I need for a useful death map?

A few dozen players can show a clear wall, because walls are loud. Halo 3 tests brought in about 20 people at a time. Count players per cell, not only deaths, so one player who retries 40 times does not make a spot look hard for everyone.

## Sources

- [Drachen and Canossa: Analyzing spatial user behavior in computer games using GIS (2009)](https://andersdrachen.com/wp-content/uploads/2011/01/p182-drachen.pdf)
- [Drachen and Canossa: Evaluating motion, spatial user behaviour in virtual environments (IJART)](https://andersdrachen.com/wp-content/uploads/2011/01/05_drachen_ijart.pdf)
- [WIRED: Halo 3, how Microsoft Labs invented a new science of play (Clive Thompson, 21 August 2007)](https://www.wired.com/2007/08/ff-halo-2/)
- [GameSpot: GDC 2009, quality testing Halo 3 (Randolph Ramsay, 26 March 2009)](https://www.gamespot.com/articles/gdc-2009-quality-testing-halo-3/1100-6206812/)
- [Game Developer: Closing the loop, validating design decisions using player data for a platformer (Amir H. Fassihi, 2 February 2014)](https://www.gamedeveloper.com/design/closing-the-loop-validating-design-decisions-using-player-data-for-a-platformer)

## Related guides

- [Game marketing measurement guide](https://manalens.dev/blog/game-marketing-measurement-guide)
- [Playtest analytics: what to measure](https://manalens.dev/blog/playtest-analytics)
- [Why players quit your game](https://manalens.dev/blog/why-players-quit-game-churn)
- [Difficulty tuning with data](https://manalens.dev/blog/game-difficulty-tuning-analytics)
- [Steam achievement data: see where players quit](https://manalens.dev/blog/steam-achievement-data-player-drop-off)
- [Game tracking plan generator](https://manalens.dev/tools/game-tracking-plan-generator)
- [Game marketing guides](https://manalens.dev/blog/topic/game-marketing)
- [Horror game metrics: session length, completion, streamer reach and refunds](https://manalens.dev/blog/horror-game-player-metrics)
- [What makes a game cozy? A 25-point design checklist for developers](https://manalens.dev/blog/cozy-game-design-checklist)
- [UGC and mods: effect on game sales and retention](https://manalens.dev/blog/ugc-mods-sales-retention)
- [Game marketing software: the 8 tool categories a studio needs](https://manalens.dev/blog/game-marketing-software-categories)
