wiki-boundary

wiki-boundary

Deterministic ranking of the pages everybody links to that have little in them. Borrowing C10 of [[roadmap-emprunts]] §2.17. Read-only, no LLM: one query over the persisted knowledge graph.

The score proposes attention. It does not establish importance. A high score says exactly one thing: many pages point here, and there is not much here when you arrive. That is a reason to look, never a verdict on the page, its author, or its priority. Say this out loud when you present results — it is the whole contract.

Pre-condition

The knowledge graph must exist AND be recent enough to carry substance measurements: wiki-meta/graph/knowledge-graph.json. If the tool refuses with "no substance measurements", the graph predates this feature — offer to run /wiki-graph, then retry.

Always look at graphAnalyzedAt in the result. The graph is a snapshot. A month-old graph ranks pages that may have been rewritten, moved or deleted since. If it is more than a couple of weeks old, say so before presenting the list, and offer to rebuild.

Steps

1. Call the tool

mcp__obsidian-router__find_boundary_pages({ vault, limit: 10 })

Arguments, all optional:

• --limit N → limit (default 10, ceiling 100).
• --min-inbound N → minInbound (default 1). Raise it on a big vault to keep only real crossroads.
• --exempt-types a,b,c → exemptTypes, replacing the default list. Pass the defaults plus the additions — ["redirect","source","answer","index"], not just ["index"] — or you will silently un-exempt the rest. See step 1-bis.
• --exempt-statuses a,b,c → exemptStatuses. No default — closed pages (status: superseded, retired…) stay visible and labelled unless the user asks to hide them. Exact match, case-insensitive; a page with no status: is never exempted; superseded-in-part is not swept up by superseded. This asymmetry with types is a decision, not an oversight: the candidate default ["superseded"] was rejected after adversarial review, because the ADR contract that standardises the token covers decision pages only, and a global filter would silently erase legitimate historical hubs.
• --all-types → exemptTypes: [], scoring every page including the ones held out by default. Use only when the user explicitly asks to see everything; warn them that migration stubs and capture records will dominate.
• --as-of YYYY-MM-DD → asOf. By default recency is measured against the graph's own build stamp, which makes the ranking a pure function of the graph file.