Journal

How a change lands

Four parts before the work, one verdict after it, and a gate that only results can open. The whole path, from a felt problem to the stable core.

Four stations on a line, proposal, landing, verdict, and stable, with a gate before the last. A dot travels along the line and pauses at each station. proposalfour parts, written firstlandingships on a fast trackverdictscored by a personstablethrough the gatethe gate
Fig. 01Every change travels the same line. It is written as a proposal, ships on a fast track, is scored by a person, and reaches the stable core only through the gate.

Oscille is meant to run on a small number of fixed shapes. A change to how the company works starts as a proposal, is scored by a person, and can only reach the stable core through one gate. This is what that looks like from start to finish.

Before the work

A proposal has four parts. The shape exists so that the person scoring the change later can do it without asking the author anything.

  1. The problem, as people feel it, not as someone guesses its cause. “Planning eats a whole evening”, not “we lack a planning tool”.
  2. The change, meaning what will exist that does not exist now. Concrete enough that a stranger could tell whether it was built.
  3. The expected outcome, the visible difference the change should make. It is written before the work starts and never edited afterwards.
  4. How we know, the single check that settles whether the expected outcome happened.

A proposal that needs the word “and” in its change line is usually two proposals. A proposal that cannot be scored from its own text goes back to its author. That is the whole test of whether it is written well.

After the work

Once a change has landed, one person scores it. Each change gets three answers, yes, partly, or no, with one line of evidence anyone could check.

Did it fix the problem it said it would? Judged by what people feel, never by effort spent or code written. Did people have to do less? The number of decisions people have to make went down, or stayed flat while the seed did more. What did it cost? Hours and money now, and computing time once we meter it.

A verdict is written down with its evidence and its date. Undoing a change later is a new change with its own verdict. Cost never fails a change on its own. It puts the first two answers in proportion.

The gate

Three concentric rings, alpha, beta, and stable, with many changes in the outer ring and few in the core. One gate leads into the core. alphabetastablenominated by use, revenue, speedthe gate, generality and replication only
Fig. 02Three tracks. Most changes live in the outer ones and act under uncertainty. Only a yes on the first two questions, at a justified cost, moves something into the core.

A practice or a component becomes stable only when the first two answers are both yes and the cost is justified in the verdict. Convenience, speed, or good sales can put a change forward. They can never promote it.

When the answers disagree, nothing is promoted. The disagreement is written down and stands as a stop, not something to average away. This makes the stable core prefer doing nothing over being confidently wrong, and it binds only the core. The faster tracks must still act under uncertainty, or the whole system would protect itself by doing nothing at all.

The method itself changes only by pull request, and the request has to state the problem, the evidence the current method failed, the replacement, and the ways the replacement is expected to fail. Verdicts keep changing because reality does. The method changes rarely, and never silently.