The deep dives

Five longer treatments on the WGRA site: memory that compounds, team scale, agent capability, the knowledge graph, and knowledge bundles.

The nuggets here are bites. When you want the full argument on one of the pillars, the WGRA site has a deep dive for each, longer, worked, with real examples.

• Memory that compounds: why a wiki accumulates where RAG rediscovers, and why failed methods and walked-back claims are preserved, not deleted (the file-drawer effect, filed alongside with a Status note).
• Team scale: collaboration within a group (attribution, conflict resolution) and federation across groups (agent cards, querying a peer’s memory with permission).
• Agent capability: discipline in the artifact, the verification gate, discipline gates, the four built-in operations, and the real case where a typed edge let a reviewer in a separate session catch a bug that recovered 75% more coverage.
• The knowledge graph: typed edges reified to a graph, SPARQL over meaning, SHACL validation, shown on live graphs.
• Knowledge bundles: the wiki packaged as a portable, shareable unit of compiled competence.

These nuggets and those deep dives are two lengths of the same story: graze here, go deep there.

Related

• Karpathy’s llm-wiki, the idea the deep dives build on.
• Our recipe: a collaborative llm-wiki, the implementation behind them.

Sources

• Beyond Memory (Saboia Moreira et al., 2026, Zenodo), the paper behind the pillars.
• llm-wiki-memory-template (GitHub), the template the deep dives document.