The macro-system coins vocabulary deliberately -- a term is minted when a concept has been measured into existence and needs a stable name. The glossary is the index of that vocabulary, and it is global: the manuals, this website, and the AI portal all point at the same maintained source rather than keeping copies that would drift.
Where the glossary lives
The maintained source is labtalk/ai_portal/AI_GLOSSARY_V1.md in the source
tree. It is a pointer index, not a corpus: each term gets one line and a
home; definitions live in the homes. It is registered in the AI portal's
recall graph (node doc.ai_glossary, reached from the onboarding trigger),
named in the developer manual's reading order, and summarized here. This page
deliberately does not duplicate the entries -- a copied glossary is a stale
glossary (the no-perishable-literals rule applied to itself).
If you are reading this from outside the source tree, that path does not resolve
for you. The Coined Vocabulary Term Index
carries every term, its one-line gloss, and its home, so the vocabulary is
reachable here without a copy existing anywhere. It is generated from the
maintained source by tools/fullstack_docs/glossary_sync.py, and a drift check
re-derives it and fails if the two disagree -- so it cannot go stale in the way a
hand-kept copy would. Definitions still live in the homes; the index points, like
the glossary it projects.
What kind of terms it holds
- Ontology and memory -- the two atoms (chat, project), synapses, recall graph, frontal memory, seed budget.
- Coordination primitives -- quip, claim-aif, handoff, aside, recursion markers.
- Durable principles -- the golden rule (verify before you assert), no perishable literals, promote-final-tests, widows-and-orphans care, evidence tiers (planned / chartered / source-evidenced / runtime-proven).
- Relational doctrine -- two house graphs, walkers, two name planes, refresh-driven slaving, canonical workspace posture, the .dtschema / .dtgraph / .erz file plane.
- Learning doctrine -- learning micro-systems and the learning macro-system, NON LMS, demonstrated negation, the red pen, and the house learning motto, the iterative example: "we regroup, go back and amend, and move forward" -- a deficiency shaped by its era is not lived with once the future arrives. Its founding instance: the original teaching dataset was generated before memo fields were imaginable, and when memo-resident workspaces made the gap visible, the ruling was to regenerate the dataset with memo fields rather than let the old ceiling shape new design.
- The team model -- AI agencies as coworkers (Class A / Class B), one member id per deployment, coauthor of record.
Why a glossary is doctrine here
Part of this system's subject matter is learning theory itself. Naming is a learning act: a coined term is a compressed lesson, and the glossary's rule -- one line and a home, definitions live in the homes -- is the same pointer-over-copy discipline the engine applies to data. The vocabulary is taught the way everything else is: each term cites where its evidence lives.