The thin slice

A thin slice is one vertical cut through every layer that delivers a small, usable piece of value. The opposite is a horizontal layer — all of one tier, the whole data model or the whole interface, finished for every feature at once. A layer is shippable to no one: a database with no logic and no screen does nothing a person can use. A slice touches the interface, the logic, and the data for a single feature, so it can be run, shown, and accepted on its own.

The test of a good slice is INVEST: it is Independent, Negotiable, Valuable, Estimable, Small, and Testable. Two older names describe the first slice of a system: a tracer bullet — a minimal end-to-end path wired through every layer, so the architecture is proven early on something real — and a walking skeleton, the smallest implementation that exercises the whole system end to end, then grows rung by rung.

The slice is not a compromise between the two failures; it removes the property that made each one fail. Four forces, and each is the exact answer to one earlier dive.

A slice is small enough to specify only enough and prove. It carries its own definition of done — the one behaviour it adds and the check that accepts it — so it can never be the unspecified, unacceptable diff that sank vibe coding. Generation with no acceptance is impossible here: the acceptance is part of the slice.

Each slice returns reality the moment it ships. The first answer arrives in week one, not at the end, so the silence that doomed the big-bang spec is broken at the first rung. Every assumption is tested while it is cheap to be wrong.

A slice is accepted before the next begins, so correctness is a property held at every step rather than a state hoped for at the end. "Done" is not a milestone far away; it is the gate on each rung. The system is correct now, and stays correct as it grows, because no rung is built on an unaccepted one.

Each slice is built on an already-accepted slice — the exact mirror of vibe coding's compounding entropy. There, confidence leaked rung over rung; here, it accumulates. The hundredth rung stands on ninety-nine proven ones, so the cost of being sure does not grow with the system.

Feedback now has a consequence worth measuring: value arrives early and keeps arriving. Step through the increments below. A thin-slice plan ships one accepted rung of value at each step, from the first. A big-bang plan ships nothing until the single release at the end — the same total, delivered late, and only if the one release works.

Put the slice on the Portal — the learning platform this course builds. The first rung is not "the store". It is one branded Snowflake id: a 14-character identifier whose embedded timestamp, node, and sequence decode back correctly and never collide within a node-millisecond. That one rung is a full vertical cut — the id format is the data, the generator is the logic, and the value a caller receives is the interface.

It is specified by its invariants, proven by a test, and shipped. Only then does the next slice — storing a single event keyed by such an id — begin, built on an accepted rung. Nothing downstream trusts an id until the rung that mints it has been accepted.

# the first rung: one branded id, specified by its invariants and proven
id = Portal.ID.generate(:user)
# => "USR0NbAb1xcFCy"

# the slice carries its own acceptance — it decodes back, and it is unique
assert Portal.ID.decode(id).type == :user
assert Portal.ID.decode(id).timestamp <= now()
refute id == Portal.ID.generate(:user)  # a fresh id, every call

Too little plan ships the unspecifiable; too much plan ships nothing. The thin slice is the minimum between them: the smallest vertical cut that is still worth shipping and still possible to prove. It answers vibe coding by carrying its own acceptance, and the big-bang spec by returning feedback at the first rung; it holds correctness at every step, and compounds confidence instead of leaking it. It is the shape every later rung in the course takes — and the unit the whole Author/Operator loop exists to produce.

Primary sources for the thin, vertical slice, and where it connects in the course.