Procedural Memory

How to do something — a workflow whose order is part of its content, not a set of independently retrievable facts.

Why it matters in a memory layer

Atomisation is the write path's job: break statements into facts that can be retrieved on their own. It is exactly the wrong operation for a procedure. Four steps are not four facts, and a memory layer preserves their order only because the extractor was told to make an exception.

Retrieval never faced this. A document arrives whole and comes back whole, so a recipe's steps stay in sequence because nothing had the chance to reorder them.

The subtler problem is that a remark about a procedure gets the same type. On this course's corpus two memories are typed procedural and one is "the diff step matters most" — split it on commas and you get a well-formed two-step workflow that does not exist. Dropping it beats parsing it, because a wrong procedure is not noise; it is something someone might follow.

Connections

Taught in: Procedural Memory · The Taxonomy That Actually Routes

Used in: Learning From Outcomes · Resolving 'Last Week' · Retrieving Procedures · Atomic Memories · The Typed Memory Model

Do not confuse with: Semantic Memory