A worked example of [[Shared Second Brain|building a shared second brain]] at cross-team scale, traced through pain → cluster → quality bar → method → tool. Extends the [[Shared second brain example - ADR log|ADR log example]] - same organisation, six months later, having outgrown a decision-only log.
## In one line
ADR log answers *what did we decide* but not *who knows about X* or *what tools have we tried* → cross-team expertise cluster, non-negotiables *multi-entry queryability* + *connectivity across dimensions* → linked databases following the [[Expertise Radar]] pattern → [[Notion]] with linked databases.
^one-line
## Starting point
![[Shared second brain example - ADR log#^one-line]]
That works for *what did we decide?*. It fails when the question shifts.
## The acts in one line each
ADR log captures decisions but cannot answer *who is the expert here*, *what tools have we tried*, or *what problems does this technology actually solve*. The trail still goes cold at the edges of the decision record.
^trail-line
Cross-team expertise as a graph - decisions yes, but also experts, tools, problems, and projects, all interrelated. Non-negotiables: *multi-entry queryability* and *connectivity across dimensions*.
^profile-line
Linked databases following the [[Expertise Radar]] pattern - each entity (problem, tool, expert, project, decision) is its own database, cross-referenced. Entries traversable from any direction.
^method-line
## Full walkthrough
### Knowledge trail mapping
Same organisation as the ADR log example, six months later. The ADR log works for the question it was designed for - *what did we decide on this project, and why?* The pattern breaks when the question shifts:
- *Who in the org has experience with event-driven architectures?* - ADRs name decisions but not experts; finding the right person requires asking around
- *What tools have we tried for caching across projects?* - the tool is buried inside decision rationale paragraphs; no way to query across
- *This new project has a notification problem - has anyone solved this before?* - the underlying problem is not a first-class entity; only decisions are
Decisions are documented. Everything else - the *who*, the *what tool*, the *what problem class* - lives in people's heads.
### Knowledge domain profiling
The cluster: cross-team expertise as a graph. Same data as the ADR log but with additional first-class entities - experts (the people), tools (the technologies), problems (the underlying business challenges), projects (the contexts), and the decisions that connect them.
Volatility: low at the entity level (the same Postgres expert remains a Postgres expert; the same tool remains the same tool). High at the connection level (which projects used which tools, who solved what problem when).
Connectivity: explicitly high - the whole point is being able to traverse from any node to any other.
The quality bar:
- **Multi-entry queryability** - the same data must be reachable by problem, tool, expert, project, or decision
- **Cross-dimension connectivity** - every entity is a first-class node, not buried inside another entity's prose
- **Maintainability** - adding a new decision must not require re-tagging across the entire graph
- **Agent-readiness** - structured enough that an agent can pull adjacent context
Non-negotiables: *multi-entry queryability* and *cross-dimension connectivity*. Without those, this is just an ADR log with extra fields.
Steward candidates: the senior engineers who already track tool usage and expertise informally - the same brokers who emerged in the ADR log act.
### Knowledge methodology matching
A linked-database pattern following the [[Expertise Radar]] convention - one database per entity (problems, tools, experts, projects, decisions), cross-referenced. The discipline:
- Every decision tagged with its problem (one-to-one or one-to-many)
- Every decision tagged with the tools considered (the *with* axis - many-to-many)
- Every expert tagged with the projects they led and the tools they have hands-on experience with
- Every problem tagged with the projects where it appeared
- Querying by traversal - pick any starting node, walk the graph
The method exists independently of any specific tool. Linked databases in Notion, page properties in Confluence, the graph plugin in Obsidian - the pattern is the contract.
### Tool (illustrative, after the workshop)
[[Notion]] with linked databases - relations between Problems, Tools, Experts, Projects, and Decisions; queryable from any entry; comfortable for non-engineers to read; in the same workspace as task management. The same workshop would land on a different tool if the room's non-negotiables included offline-first ([[Obsidian]] with graph plugins) or compliance-grade audit trail (a more rigid DMS).