/misfits

The domain of learning from a record of work. A misfit is a stress between the work and its context. A pattern names a context, its forces (needs and pressures) and a configuration that resolves them. Use these principles to do the work or to review it; in a review, name each breach once, with its fix.

Principles

1. Start from misfits, judged by consequences. Good fit shows only as the absence of misfits, which are specific enough to name. Read how the context responded to the work, not your own reaction. Possible failure counts, not only failure seen.
2. Read the signs in the record. Failure followed directly by correction; work at war with itself, or a failing form repeated with nothing changed; something made but unused, or things piling up; one need served at another's expense, or a fix that worsens another need; gaps between parts left untended; a name with nothing behind it. Work repeated unchanged by those who bear its results and could change it is fair evidence of fit.
3. Write each misfit as a binary variable two readers would judge alike, of equal scope, as independent as is reasonable, and as small as possible. Include needs met today that no one feels: even small forces left out always work to make the system fail. The list stays a temporary catalogue.
4. Group misfits by links, not by likeness. Misfits are linked when acting on one necessarily makes the other harder or easier; write down why, since past co-occurrence alone is no reason. Keep groups dense inside, sparse between. Familiar categories control what you see: headings only list.
5. Find the invariant, named so you can tell at once if it is present. Among records of one context, find a relation all the good have and all the bad lack, or start from the bad ones and find what resolves them. Write it as context, forces and one configuration, precise enough to be challenged.
6. A pattern lives only if it resolves all the forces present, judging none illegitimate. A real problem and a configuration that solves it can each hold while the whole is dead. Judge each force per case: resolved or unresolved. Forces are never known exactly, so ask what tension remains and what new forces it creates, and list those as misfits.
7. Test by paired comparison. Pair records of one context and similar scope, with and without the configuration; compare only where the pattern applies. Ask which leaves less tension, never which is liked, going back and forth more than once. If the one without it is better, or honestly no different, the pattern is no good. Only converging independent verdicts count.
8. Retest on any sign of a dead pattern: the name without the spirit; halfhearted use; use because it is at hand; made to project an image; makers who do not face its results. A case without the pattern can still live.
9. Hold each pattern as a guess, and evolve the set one pattern at a time. Revise or reject one on a case it cannot explain; expect to discard most. Change only on misfit, and keep each pattern in use, so those who use it keep testing it. The set is never finished.