Software Testing Pyramid Economics (2D)

Test Suite Shape

Unit tests 1200

Integration tests 150

E2E tests 20

Pipeline Conditions

Bug injection rate 1.0 bugs/s

E2E flakiness 8%

Live Readout

Elapsed

0.0 s

Bugs injected

Caught

Escaped to prod

CI cost / build

0.0 s

Cost per bug caught

—

How it works

Every injected "bug" is assigned the layer that would realistically catch it — most defects are unit-local (65%), fewer need components wired together (25% integration), and only a handful only surface end-to-end (10% E2E), the real shape of the pyramid.

Each layer's coverage saturates with test count: coverage = 1 − e^−n/scale (scale ≈ 500 for unit, 120 for integration, 25 for E2E — E2E tests are expensive so far fewer are needed to move the needle, but each only exercises one journey). A bug is only detectable once it reaches its own layer or a later one — an integration-only bug cannot trigger inside an isolated, mocked unit test. A later layer that isn't the bug's "true" layer still gets an incidental catch chance (0.3× per layer skipped), same as real regression coverage. E2E detection is further scaled by (1 − flakiness), modelling real flaky-test false negatives.

coverage(layer)   = 1 − exp(−testCount / scale)
P(catch at layer) = coverage × effectiveness(layer, trueLayer)
CI cost / build    = unit·10ms + integ·500ms + e2e·5000ms
cost / bug caught  = (bugs run through CI × CI cost) / bugs caught

Pyramid — particles rise bottom→top through Unit → Integration → E2E → Production; green = caught, red = escaped.
Coverage panel — live coverage% and catch counts per layer.
Cost trend — cost-per-bug-caught (CI seconds) over the run; a thin, top-heavy "ice-cream cone" suite drives this up sharply.

Caught Escaped to prod In flight