The Great Library of SISO · Research architecture
Registry outside.
Knowledge system inside.
The Great Library is the public catalog and permanent learning record. SISO Knowledge is one independently addressable Research Work: the place where knowledge is acquired, curated, stored, indexed, related, and retrieved. Neither local folder nesting nor a giant repository becomes ontology by accident.
If a thing answers “what exists, which release is selected, and where is the evidence?”, it belongs in the Great Library registry. If it produces or retrieves durable knowledge, it belongs in SISO Knowledge. If it mines source, it belongs in Foundry. If it turns evidence into grounded conclusions, it belongs in Evidence Engines. Large payloads and live state stay in a separately governed data plane.
The four responsibilities
What the Knowledge checkout actually contains
| Class | Zones | Decision |
|---|---|---|
| Canonical | sections, books, module_templates | Keep as authored knowledge and schemas. |
| Production tools | pipelines, queries, scripts | Keep together until independent adoption—not size—proves a repository split. |
| Generated views | _index, graph | Rebuildable projections, never hand-authored truth. |
| Operators | agents, legacy memory | Keep Knowledge-specific definitions; migrate generic capability and runtime state to their owners. |
| Independent Works | Foundry and Evidence Engines checkouts | Represent through registry relationships; physical nesting grants no ownership. |
| Boundary exceptions | Legacy web application and shared design system | Preserve dirty work now; bridge to product/shared-UI ownership later. |
| Data plane | Acquired corpora, transcripts, databases, queues, caches, and logs | Keep outside Git with source, rights, retrieval, and transformation provenance. |
Repository rule
A unit earns its own repository only when it is independently adoptable, ownable, securable, or separately releasable. A large directory, generated dataset, UI route, or convenient local grouping is not enough. Humans and agents may clone the whole Knowledge source baseline or an independent Research Work; future packages can provide smaller materialization units without rewriting Work identity.
Evidence and honest limits
The current public release is an architecture baseline, not a claim that every local corpus or live pipeline state was published. The local checkout had 164 parent-level dirty entries when mapped, plus preserved dirty nested products. The release therefore marks installation unavailable, portability experimental, and reuse rights unavailable until a license is accepted. This is a truthful migration surface, not a cleanup theatre.
Read the pinned human domain map, machine domain contract, data boundary, and migration receipt.