Feature Flag Combinatorics: The State-Space Explosion

Feature Flags

Number of flags N 4

state = 0110

Auto-run random tests

Test speed 4/s

Coverage

0 / 16 states tested (0%)

Total states 2^N

Tests run

Untested states

Pairwise-suffices

~24

How it works

Each feature flag is a boolean, so the set of possible app configurations is the power set of the flags — every vertex of an N-dimensional hypercube. Adding one more flag does not add one more case; it doubles every case that already existed:

states(N) = 2^N
states(4) = 16     states(10) = 1,024
states(6) = 64     states(20) = 1,048,576

The 3D scene projects that hypercube: each flag gets its own spatial axis, and a vertex is lit when its exact combination of flags has been exercised by a test ("run one test" or the auto-runner picks a uniformly random vertex). An edge joins two vertices that differ in exactly one flag — flipping a single toggle. Watch the Untested states counter as N grows: coverage collapses long before you reach a handful of flags, which is exactly why teams accumulate untested "dark corners" of flag combinations in production.

N flags	Exhaustive 2^N	Pairwise ≈ C(N,2)·4
3	8	12
5	32	40
7	128	84

Pairwise (combinatorial) testing is the standard mitigation: instead of covering every vertex, you only require every pair of flags to jointly take every one of their 4 value combinations (00, 01, 10, 11) somewhere in the suite — which catches the overwhelming majority of real interaction bugs at roughly C(N,2)·4 cases, growing quadratically instead of exponentially. That is why real flag-management platforms (LaunchDarkly, Firebase Remote Config, Optimizely) push teams to keep the *live* flag count small and retire stale flags quickly — every flag left in the codebase permanently doubles the untested state space of every flag added after it.

Current state Tested (covered) Untested

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