Structural Signals #005 · Nave · April 28, 2026

Frostbite

Frostbite happens when blood does not reach the extremities. The body’s core stays warm. The fingers and toes lose feeling, then color, and eventually the tissue itself. The walls of the circulatory system are intact. The problem is reach.

Nave is a website with the same shape: a strong, well-organized core surrounded by a periphery the editorial web never reaches.

The Subject: Nave

By every surface metric, Nave looks healthy. Pages reach each other in roughly three clicks. The link density is well above the threshold where structure carries information. The community algorithm finds eleven well-formed topical clusters at the center of the graph, each one with sharp boundaries and strong internal cohesion. Most of the site is most of those eleven clusters, and most of the site is fine.

What is not fine sits at the edges. The same algorithm finds eight more “communities” at the periphery — six of them containing exactly one page each, the other two micro-clusters of two and four pages. Together those eight account for less than 4% of the site’s pages but 42% of the detected community count, because each is a topological fragment the algorithm refuses to glue to the real clusters. These pages are not orphans. They have inbound links from navigation. They were published, indexed, and made visible. But no editor ever wrote a content link from them to anywhere else, and nothing else writes content links to them. They sit alongside the real structure without ever joining it. The body is warm. The fingers are cold.

This is what happens when content gets shipped without being integrated. A page goes live. It gets a slot in the navigation menu. The team moves on. No editor reads through related pages and adds the contextual links that would make the new page part of a topical neighborhood. The site keeps its existing shape, with one more stranded page added to the perimeter.

Structural Signal: When a community detection algorithm returns more than 30% of its communities as single-page or near-single-page clusters, the topology has a circulation-delivery problem at the edges — pages exist in the index but outside the editorial fabric of any section.

Interactive: zoom, pan, and explore the anonymized topology

The Five Lenses

Skeleton

Good

Circulation

Good

Organs

Needs Work

Health

Needs Work

Nervous System

Needs Work

Skeleton — Size & Connectivity

Average path length of 3.1, density of 0.016. Healthy numbers for a 219-node graph: any page can reach any other in roughly three clicks, and the link density is high enough that the topology carries real structural information. The skeleton itself is sound. This is not a site where pages cannot be found. The problem is not connectivity at the body level. It is circulation at the extremities.

Circulation — Authority Flow

PageRank Gini of 0.56 with the top 1% of pages holding 16% of authority. A notably healthy distribution — well below the 0.70-0.85 range where authority concentrates in a few dominant hubs. Nave distributes authority across multiple centers of gravity rather than funneling everything through a single homepage. The body has plenty of blood. The question for the next two lenses is where it actually flows.

Organs — Community Structure

Nineteen detected communities with a modularity score of 0.73 — among the strongest community structures in this series. The number to look at is what those nineteen contain. Eleven are real topical clusters, ranging from 55 pages down to 7. Inside those eleven the boundaries are sharp and the internal cohesion is high. Then come the other eight: six are single-page communities, one is a two-page cluster, one is a four-page cluster. Together those eight account for less than 4% of the site’s pages but 42% of the detected community count, because each is a topological fragment that the algorithm refuses to glue to the real clusters. The community structure is well-defined at the center and frays at the edges.

Health — Content Isolation

2% orphan rate (5 pages with zero inbound links), 4% dead-ends, and 7% island rate. The orphan and dead-end numbers are modest — better than most enterprise sites. The 7% island rate is the more telling figure: roughly sixteen pages live in connected components detached from the main graph, reachable from navigation but not via editorial link traversal. These are not exactly the same set as the singleton communities — island rate is a connectivity measure, singleton rate is a clustering measure — but both report the same underlying cause: pages added to the CMS that no editor wove into the topical web.

Nervous System — Depth & Bridges

Participation coefficient of 0.14. On average, only one in seven of a page’s editorial links crosses a community boundary. That is lower than Episode 4’s archipelago at 0.30 — but the more important number for Frostbite is the asymmetry behind the average. The eleven real communities link to each other through dozens of cross-community pages. The eight marginal communities have almost none. They sit beside the structure with circulation moving past them, never participating in it. Depth metrics for this graph compute sparsely (a measurement artifact at this scale), but the participation coefficient combined with the singleton concentration tells the same story without them.

What Would We Fix?

Frostbite calls for circulation, not surgery. The healthy core does not need to be rebuilt. The eight marginal communities and the orphaned pages need to be reattached to the editorial neighborhoods they topically belong to.

The optimizer identified 93 strategic link additions that do exactly that — 5 to rescue orphaned pages, and 88 cross-community editorial links that pull stranded pages into the nearest real cluster. No links removed. The participation coefficient rises. The singleton communities dissolve, not because the algorithm changes its mind, but because the pages now have topical neighbors to belong to.

The original topology as crawled. All edges are from the live site.

Rescue Orphaned Pages

5 pages have zero inbound editorial links, making them invisible to crawlers and AI agents that follow link structure.

+5 inbound links from community hub pages All pages reachable through editorial link structure

Orphan Rate
2.3%0.0%

Restore Circulation to the Periphery

88 cross-community editorial links that pull stranded pages back into the real topical clusters around them. Participation coefficient would rise from 0.14 toward the 0.30+ range.

+88 cross-community editorial links Editorial circulation reaches the eight marginal communities, not just the core

Edge Count
763856

Dissolve the Singleton Tail

8 marginal communities (6 single-page, 2 micro-clusters -- 42% of all detected communities) represent pages stranded outside any meaningful topical cluster. Linking them into their nearest neighbor eliminates the fragmentation tail.

Integrate 8 marginal communities into adjacent clusters Periphery rejoins the main editorial graph

Island Rate
7.3%< 2%

Structure is Signal

Nave’s topology tells the story of a site that ships pages faster than it integrates them. The center is well-built. The community detection algorithm finds eleven real topical clusters with strong internal cohesion — products, science, careers, investors, news, the rest — and the cross-cluster links that exist do their job. The problem is what happens to the pages that get published outside the integration loop: a press release here, a one-off product page there, a careers detail buried at a URL that nobody links back to. Each one becomes a stranded page, a community-of-one, a finger in the cold.

Ninety-three strategic link additions would warm the periphery. Five rescue orphans. Eighty-eight pull stranded pages into adjacent topical clusters. The structure does not change. The circulation does. Singleton communities dissolve into the real clusters around them, not because the algorithm changes its mind but because the editorial signal arrives.

This is the fifth topology in this series, and the pattern we see most often is not chaos or neglect — it is uneven editorial attention. Episode 1 found content islands. Episode 2 found gravity wells that trapped traffic. Episode 3 found a monolith that refused to differentiate. Episode 4 found an archipelago with too few bridges. Nave adds a new variant: a healthy editorial body with cold, stranded extremities.

The fix is always smaller than the problem looks. Nave’s topology does not need a redesign. It needs 93 links — a 12% increase in edge count — placed where pages exist outside the conversations they should be part of.

Methodology & Disclaimer — This analysis was performed using web topology crawling and network science methods including PageRank, Louvain community detection, and betweenness centrality. Navigation, header, and footer links are excluded to isolate editorial linking structure — only in-content links are analyzed. All data represents publicly accessible page structure only — no content, metadata, or user data was collected or stored. All identifying information has been anonymized. Structural patterns are presented for educational purposes only.