The test pyramid

Before the cheer feature may climb the ladder, it must pass its exams, and the exams come in sizes.

The smallest is a unit testingConcept · lights on your mapunit testingTesting one small piece, usually one function, in isolation: call it with known inputs, check the outputs. Thousands of them, run in seconds, and when one fails the culprit’s address is precise. exam: one function, alone, interrogated. Does deliver_cheer drop a cheer when the runner has finished? Does it survive the runner nobody follows, the one Priya found in week two? There are forty-one of these for the cheer feature alone, they run in under a second, and when one fails it names the guilty function directly.

The middle size is integration testingConcept · lights on your mapintegration testingTesting several components together: the logic actually writing to a real test database, actually calling a fake payment service. Slower and fewer; they catch failures living in the seams.: several parts working together, the delivery logic actually writing to a real test database, actually pushing through a real queue. Slower, fewer, and they catch the failures that live in the seams between parts, which is where failures prefer to live.

At the top, end-to-end testingConcept · lights on your mapend-to-end testingTests that drive the whole product the way a user would: a robot browser clicking, sending, checking what appears. Realest and slowest; a handful cover the journeys that matter most. drives the product like a person: a robot browser logs in as a fake spectator, follows a fake runner, types go faster, and checks the message lands. Realest, slowest, fewest. A handful of these guard the journeys that matter most, and one of them now guards cheering.

Robots and humans

Everything above is automated testingConcept · lights on your mapautomated testingChecks written as code and run by machines, identically, thousands of times a day, never bored on the four-hundredth run. What the pyramid is made of.: checks written as code, run by machines, incapable of boredom on the four-hundredth identical run. Its partner is manual testingConcept · lights on your mapmanual testingA human actually using the product, exploring and judging. Slower and unrepeatable, and the only way to catch what nobody thought to write a check for., a human wandering the feature with intent, and it is not the obsolete one. On Thursday a teammate manually sends themselves cheers for an hour and finds the thing no robot was told to look for: on a cracked old Android, the cheer banner covers the pace display. A robot verifies what you thought of. A person finds what you did not.

The pyramid passes. Next: the strange nightly world where the exams sit themselves.