This hub gathers the As Written / Scarlet Dame technical-and-conceptual stack across two registers. The conceptual mainline: a [collective-memory thesis] (Narrative_CollectiveMemory) — RDF knowledge graphs over vector databases for traceable organizational truth — extended by the [prompts-as-ontology-targets idea] (Narrative_PromptsAsOntologyTargets), where prompts/skills are versioned objects compiled differently per worldview. The implementation register names concrete choices: [MCP as the integration layer] (Observation_MCPStack), [SortaRich on Cloudflare Workers] (Obs_TechStack) with an AI-generated codebase, and [per-repo bearer-token scoping] (Obs_BearerTokenScope). Beneath the optimistic graph-as-memory framing sits a sharper, more skeptical premise about the models themselves — [LLMs as continuation machines] (Observation_LLMContinuation) — attached as related rather than broader, signalling it qualifies rather than supports the mainline. The cluster reads as the foundational why (anti-hallucination, ontology-driven prompts) plus the how (serverless, MCP, scoped auth), with one philosophical undercurrent about what LLMs actually are.
Witness phrases
verbatim RDF knowledge graphs over vector databases maintain organizational truth
verbatim AI collective memory prevents hallucination and ensures consistency across teams
verbatim LLM prompts as ontology compilation targets: versioned, scoped, and recompiled when the worldview shifts
verbatim LLMs are continuation machines, not thinking machines.
verbatim As Written uses MCP (Model Context Protocol) as the integration layer for organizational memory tools.
verbatim SortaRich built on Cloudflare Workers serverless architecture; entire codebase generated via AI prompts.
verbatim Bearer API keys scoped to {owner}/{repo} in URL path
verbatim Skeptical undercurrent: LLMs framed as mere continuation machines following narrative vectors — qualifies the upbeat 'AI collective memory' thesis, and is attached as related, not broader.
verbatim Two registers coexist without merging: conceptual/strategic narratives (graphs, ontology-compiled prompts) vs. concrete infra observations (Cloudflare Workers, MCP, bearer-token auth).
verbatim Self-referential edge: the codebase is itself AI-prompt-generated, while the product's thesis is that prompts should be ontology-compiled and worldview-scoped — an unstated tension about whether the stack practices its own model.
In this area
NarrativeNarrative_CollectiveMemory Co-located organisational memory as competitive infrastructure
NarrativeNarrative_PromptsAsOntologyTargets LLM prompts as ontology compilation targets: versioned, scoped, and recompiled when the worldview shifts
ObservationObservation_LLMContinuation LLMs are continuation machines, not thinking machines. They follow narrative vectors embedded in training data.
ObservationObservation_MCPStack As Written uses MCP (Model Context Protocol) as the integration layer for organizational memory tools.
2 further nodes are not published in the public projection. Each node links from the entity index.