Chapter 6: Accelerate the Knowledge Transformation Cycle with curation and feedback
Curation and feedback
Core ideas
Humans paper over conflicting docs with relationships; AI inherits the same ambiguity as unpredictability unless you establish authority.
The Knowledge Curation Pyramid elevates trust: Well-Retrieved → Well-Sourced → Well-Informed → Canon (curated truth AI should treat as
authoritative).
Canon is not immutability; it is what the model will assume is true. Governance and friction match risk (preventive vs detective control patterns).
Separate concerns: Canon (truth), knowledge base (access), and Segments (performance-oriented slices for specific workflows).
Metadata enrichment lets one curated source safely power multiple AI products; scale roles and stages when growth, compliance, or model demands require it.
Principles from the chapter
Unresolved ambiguity in your context becomes unpredictability in your products and processes.
In an AI-first organization, when you upgrade the Canon, you instantly retrain the AI workforce relying upon it.
Speculative Canon is helpful if labeled, but Canon requires battle-testing to be confident in its maturity.
The Canon ensures truth. The knowledge base ensures access. The Segment ensures performance.
Metadata enrichment of raw documents enables a single source to be discovered and applied to multiple AI-first products and processes.
The stability of your Canon is determined by the intentional friction you introduce. High-risk domains require high friction (preventive control), while low-risk domains thrive on low friction (detective control).
Treat all data in a Segment as viewable to anyone or any system interacting with the associated AI process.
Read the chapter for…
The winter rail-controller scenario, full pyramid level definitions, architecture of pools and segments, and governance triggers for formalizing roles.