What is a Pattern Library?

Organisations mostly learn by anecdote. The most recent disaster becomes policy. The loudest story in the room becomes the rule. A single charismatic failure can steer a firm’s behaviour for a decade, while a quiet pattern that has repeated thirty times goes unnoticed because no one story about it was ever vivid enough to retell. This is not a memory problem. Most firms remember their stories all too well. It is an evidence problem: the stories were never counted.

A Pattern Library is the collection of recurring decision patterns an organisation has mined from its scored outcomes, admitted only once enough evidence has accumulated, and never from a single anecdote.

From anecdote to pattern

An anecdote is one outcome plus a narrator. It is compelling in proportion to how vivid it is, which has nothing to do with how representative it is. A pattern is different in kind: a shape that recurs across many outcomes, under describable conditions, often enough that chance is no longer the best explanation.

The raw material for that distinction is scoring. A decision that was recorded, and later scored against what was recommended and what was decided, becomes a data point rather than a war story. Enough of them, accumulated in an Outcomes Ledger, make mining possible at all: without scored outcomes there is nothing to count, and the organisation is back to arguing from its loudest memories.

The n-threshold discipline

The defining rule of a pattern library is the one about admission: a pattern is only admitted once enough evidence has accumulated. Below the threshold, an observed shape is held as a hypothesis, watched, and deliberately kept out of recommendations. This is the library’s immune system, and it matters for a reason that compounds. Small samples produce confident nonsense: three engagements can align by chance, and a coincidence promoted to a rule does not just mislead once. Because patterns feed future recommendations, a false pattern quietly misprices every decision it touches until someone notices, and the very authority of the library makes noticing slower.

The threshold converts “we think we have seen this before” into “the record shows this, and shows it this many times”. And admission is not tenure. A pattern must remain challengeable and falsifiable: it names the conditions under which it should hold, so new outcomes can contradict it, and when they do, the pattern is demoted rather than defended. A pattern that cannot be argued with is not knowledge. It is a superstition with a database.

What an admitted pattern feeds

Once admitted, a pattern goes to work in one specific place: the next decision that matches its shape. A commitment with a familiar profile is no longer priced from zero; it arrives with the pattern attached, along with the evidence behind it, as one input among the others. The pattern is a warning with receipts, never a veto: a person still decides, and can decide against it. This is what separates a pattern library from a filing cabinet. It is institutional memory made operational: the organisation’s experience showing up inside the decision, at the moment it is useful, rather than waiting in a repository for someone to think of rereading it.

Pattern libraries and compound patterns

The library has a sibling, and the two are easy to conflate. A pattern library looks backwards: it is mined from the organisation’s own scored outcomes, accumulated across time, and it answers “what has tended to happen to decisions shaped like this?”. Compound patterns look at the present: cross-domain detectors watching for conditions forming right now that no single function can see from where it sits. One learns from history; the other watches the horizon. In ONX the two meet at the same standard and the same destination: every pattern is explainable, challengeable and traceable to its evidence, and both route into a decision made by a person, not into an automation.

One concrete example

Clearly illustrative, with no customer implied. Over several years a services firm scores its engagement outcomes as a matter of routine. A shape begins to recur: engagements sold with an aggressive start date, on the strength of a client’s stated but never measured claim about data readiness, slip, and slip in the same way, in the same phase. The first such project is a war story. The fifth is a hypothesis. Only when the record holds the shape enough times is the pattern admitted to the library, and from then on every proposed commitment matching it arrives with the pattern attached: not a refusal, but the firm’s own counted history, in the room at the moment of choice. Meanwhile the firm’s single most famous disaster, a genuine one-off retold at every offsite, creates no pattern at all, however often it is told. That asymmetry is the entire point.

A firm that runs this discipline long enough changes what it is allowed to believe about itself. Its rules are the ones its record can defend, its exceptions are visible as exceptions, and its experience compounds instead of merely accumulating. That is decision intelligence operating at the level of the organisation’s judgement, one admitted pattern at a time.