Level Analysis (Game)
Level Analysis (LevelAnalysis) reports each game level's pass rate, fail rate, stuck rate, and score distribution — the core model for numeric balance tuning and churn point localization. It corresponds to the backend AnalyticsService.LevelAnalysis.
Game-specific model. It depends on level events reported by the C# SDK (
level_start/level_finish/level_fail) and thelevelfield. See the C# SDK Integration Guide.
1. What Questions It Answers
- Which level has the lowest pass rate? (stuck point, churn hot spot)
- What is each level's 3-star rate (
star3_rate)? Is the difficulty curve reasonable? - Where does the number of challengers drop sharply? (players churned en masse at the previous level)
2. Key Metrics
Each LevelStat (single level) contains:
| Metric | Meaning |
|---|---|
| attempt_count | Number of attempts (level_start) |
| finish_count | Number of completions (level_finish) |
| fail_count | Number of failures (level_fail) |
| pass_rate | Pass rate (finish / (finish+fail)) |
| stuck_rate | Stuck rate (1 - pass_rate; higher = harder / churn point) |
| avg_score | Average score |
| star3_rate | 3-star rate (share with context['stars']==3) |
| player_count | Players who challenged the level (distinct) |
3. Backend API
gRPC: AnalyticsService.LevelAnalysis
| Field | Type | Description |
|---|---|---|
time_range | TimeRange | Analysis time range |
app_id | uint32 (optional) | Filter by app |
level_id | string (optional) | Specific level ID (empty = all levels) |
Response LevelAnalysisResponse: levels[] (each LevelStat, sorted by player_count descending).
HTTP (admin proxy)
POST /admin/v1/analytics/level-analysis
Content-Type: application/json
{
"timeRange": { "startMs": 1718169600000, "endMs": 1718774399000 }
}
4. Typical Scenarios
Difficulty curve tuning
Look at pass_rate and stuck_rate in level order: the ideal curve rises gradually and smoothly. A sudden jump in some level's stuck_rate indicates a difficulty cliff — lower that level's difficulty or strengthen guidance in the preceding levels.
Churn point localization
Watch how player_count decays along the levels: where the challenger count drops sharply between adjacent levels is the churn point, usually corresponding to a level with a high stuck_rate.
3-star rate evaluation
star3_rate reflects a level's challenge for top players. A too-high 3-star rate (everyone gets three stars) means it is too easy; a too-low one means the 3-star condition is too harsh.
5. Notes and Caveats
- Depends on standard level events:
level_start/level_finish/level_failmust be reported, with thelevel(level ID/name) field on the events, otherwise no statistics are possible. - 3-star rate depends on context:
star3_ratereads from the event'scontext['stars'], so the client must report the star rating. - Pass rate basis:
pass_rate = finish / (finish + fail), excluding unfinished in-progress attempts. - player_count decay is the key signal: pass rate alone is not enough; combine it with player-count decay to locate real churn points.