Crash Fingerprint Clustering

Crash Ingestion

Distinct root causes 4

Fingerprint depth (top-N frames) 3

Stack noise (symbolication variance) 10%

Crash arrival rate 2.0/s

Live Metrics

Crashes processed0

Unique clusters found0

Crash-free users100.0%

How it works

Real crash reporters (Firebase Crashlytics, Sentry, Bugsnag) don't show you every raw crash — they group thousands of reports into a handful of "issues" by hashing part of the stack trace. This sim runs that exact grouping algorithm on synthetic crashes.

fingerprint = hash( frames[0 .. N-1] )
if fingerprint in knownClusters:
    knownClusters[fingerprint].count += 1
else:
    knownClusters[fingerprint] = new Issue(count = 1)

Each falling sphere is one crash report drawn from one of the root causes below — the true, hidden bugs. Its stack trace is the cause's fixed frame sequence, but with stack noise probability each frame can get replaced (obfuscated symbol names, inlining, thread interleaving all do this in production). Only the top N frames (fingerprint depth) are hashed to decide which tower a crash lands on.

N too small — unrelated bugs share a shallow prefix and get merged into one inflated tower (false consolidation).
N too large, or noise too high — one real bug's traces disagree past frame N and fracture into many separate towers (over-fragmentation) even though "Unique clusters found" is an exact count of distinct fingerprints, not a display limit.

Crash-free users tracks a fixed pool of 60 synthetic devices:

crash-free % = (usersWithZeroCrashes / totalUsers) × 100

It only falls, never rises within a run — exactly like the release-health metric shops like Crashlytics report per app version.

Falling crash report Cluster tower (height = crash count) Overflow bucket (display cap only)

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.

Crash Fingerprint Clustering Simulator