The Four Pillars

Four pages out of nearly a thousand hold more than half of this site’s authority. They form a tier of their own, each carrying between 12% and 16%, and the top 1% of pages together hold 73%. That is the kind of number we have only ever found on sites that were also broken in every other way: orphaned, fragmented, flat. Rotunda is not. Its content is reachable, its communities are balanced, its graph forms a single connected body, and its cross-section flow is the healthiest we have measured. By every lens except one it is among the best-built sites in this series. And more than half of its authority rests on four pages.

The Subject: Rotunda

We’ll call this site Rotunda, an enterprise financial-services property of roughly 980 pages joined by about 4,200 internal editorial links. It is, on inspection, a well-made site. Content is organized into 32 distinct communities of sensible sizes. Almost nothing is orphaned, almost nothing dead-ends, and the entire graph forms a single connected body: 100% of pages in one component. Run the standard health checks and Rotunda passes nearly all of them.

The exception is where the authority lives. In a healthy link graph, internal authority spreads across many pages in many sections, so that no single page is irreplaceable and no single page speaks for the whole site. Rotunda’s authority does not spread. It pools onto four pages. The visualization makes the imbalance plain: a handful of nodes sit swollen at the center of an otherwise even, well-distributed field, drawing a weight wildly out of proportion to their number.

Interactive: zoom, pan, and explore the anonymized topology

OrphansHubsBridgesSilosIslandsReset

The Insight

We call this pattern The Four Pillars. A rotunda is a building whose dome rests on a small ring of columns; pull one and the roof comes down. That is this topology. The top-ranked page holds 16.2% of all PageRank, the second 14.1%, the third 13.8%, and the fourth 12.1%. Four pages, roughly 56% of the site’s entire link equity. A fifth column carries 8.5%, and then the distribution falls off a cliff: the next page down the list holds 3%. The remaining 970-odd pages share what is left.

This is a different failure from the concentration we have seen before. When a site hoards authority and orphans its content and strands its sections, the concentration is a symptom of collapse: equity climbs to the few pages tall enough to catch it because nothing else is wired to hold any. Rotunda has none of that decay. Its pages are linked and reachable, and the mechanism of the funnel is visible in the graph itself: nearly every page on the site carries an editorial link back to the same central hub. The authority pools on four pages not because the rest of the site is broken, but because the internal linking deliberately routes almost everything through the same few doors, and those doors open onto the pillars. The structure is sound. It was simply built to lean on four columns.

The danger is concentrated risk. Four pages are doing the structural work of a few hundred, which makes each one a single point of failure no audit would flag. Retire one of those pages, change its URL in a redesign, let one fall out of the index, and roughly a seventh of the site’s authority goes with it. For an AI agent reading the site by following links, the picture is narrower still: it will weight those four pages as almost the entire company, because the link graph tells it they are. Whatever the four pillars happen to say becomes the site’s whole story, and whatever the other 970 pages say is, by comparison, a whisper.

Structural Signal: When four pages hold more than half of a site’s authority and the top 1% holds nearly three-quarters, while orphan rate, dead-end rate, and connectivity all come back healthy, the topology has a four-pillars problem: a well-built structure whose entire authority rests on a handful of load-bearing pages.

The Five Lenses

Skeleton – Size & Connectivity

976 nodes, 4,186 edges, density 0.0044, average path length 2.0, diameter 4. This is a clean skeleton. The density is modest but the graph is thoroughly connected: every page falls into a single component, and any two pages are on average two clicks apart. Unlike Episode 7 (Minaret), whose short paths came from a thin, centerless sprawl, Rotunda’s come from a properly wired body of content. The bones are good. Everything that follows happens on top of a sound structure, which is exactly what makes it interesting.

Circulation – Authority Flow

PageRank Gini of 0.77, the top 1% of pages holding 73% of authority, four pages holding roughly 56% between them. This is the most concentrated circulation we have measured on a structurally healthy site. Episode 7 (Minaret) ran higher on paper at 0.87, but as a symptom of collapse, on a graph where 81% of pages were orphans; among sites whose other lenses come back green, nothing comes close. For comparison, Episode 8 (Parapet) ran a healthy 0.38 and Episode 5 (Nave) a worrying 0.56; Rotunda sits far past both. What is unusual is not just the height of the concentration but its shape: not one dominant hub with a gentle slope behind it, but four co-equal giants between 12% and 16% each, a fifth column at 8.5%, and then a sheer drop to nothing — the next page holds 3%. Click Hubs to see the cliff. Healthy enterprise sites distribute authority across many hubs in many sections. Rotunda runs more than half of it through four.

Organs – Community Structure

Louvain detection returns 32 communities at a modularity of 0.48, and this is where Rotunda looks genuinely good. The largest community holds 15% of pages (145), and below it sit four more substantial sections (118, 118, 112, 95 pages) followed by a long, natural tail of smaller clusters. The modularity is moderate rather than high — the heavy traffic toward the center blurs the boundaries — but the sections themselves are real and sensibly sized. There is no monolith here, none of the single undifferentiated mass that defined Episode 3 (Atrium). The site is organized into genuine topical neighborhoods. The authority concentration is not a failure of how the content is grouped. It is a failure of where the links point.

Health – Content Isolation

2.0% orphan rate (20 pages), 1.0% dead-ends (10 pages), 0% islands, with the entire graph in one connected component. By the standards of this series these are excellent numbers. Episode 1 and Episode 6 were defined by content stranded with no way in; Rotunda strands almost nothing. Nearly every page receives at least one inbound editorial link and offers at least one outbound. This is the lens that makes the circulation result so striking. The site is not concentrating authority because its content is unreachable. The content is reachable. The authority is simply funneled, by design, onto four pages.

Nervous System – Depth & Bridges

Participation coefficient of 0.54, the highest we have recorded, and an average depth of 1.0 — 957 of 976 pages sit exactly one link from the graph’s center. Read those two numbers together and the funnel becomes visible. The participation coefficient means pages link healthily across community lines rather than staying sealed within their own cluster; compare Episode 5 (Nave, 0.14) or Episode 7 (Minaret, 0.12) — Rotunda’s cross-section flow is in a different class. The depth tells you where all that wiring points: nearly every page on the site links straight back to the same central hub, which is precisely how the pillars collect their weight. The cross-community wiring is real. It simply converges, in volume, on four pages and stays there.

What Would We Fix?

Most episodes in this series end with a list of structural repairs, because most of the sites are structurally broken. Rotunda is the rare case where the optimizer has almost nothing to do. Run it and it rescues the 20 orphans, adds 166 cross-community bridge links, and lifts the edge count from 4,186 to 4,372, nudging the orphan rate to zero and trimming dead-ends from 1.0% to 0.6%. Useful, but marginal. The mechanical fixes barely move the site, because by the metrics a structural optimizer reads, Rotunda is already healthy.

The real work is the one number the optimizer cannot touch: the concentration. None of the four pillars is over-linked in the outbound sense the optimizer looks for, so it leaves them alone. Spreading authority off four load-bearing pages is not a connectivity problem to be solved by adding edges in bulk. It is an editorial decision about which pages deserve to be linked from where, made one link at a time, with the goal of giving the deep, healthy, well-organized content Rotunda already has a share of the equity it currently routes past.

OriginalOptimizedShow Changes

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

Rescue 20 Orphaned Pages

A small number by series standards, but 20 pages still receive no inbound editorial link and sit outside the discoverable body of the site. Connecting each to its community hub closes the last gaps in an otherwise reachable graph.

+20 inbound links to orphan pages → Every page becomes structurally discoverable

Orphan Rate

2.0%→0.0%

Add Cross-Community Bridges

166 cross-section links deepen the links between Rotunda's 32 well-formed communities, raising cross-community flow and giving authority more routes to travel sideways instead of pooling at the top.

+166 cross-community bridge links → Related sections gain editorial connections between them

Edge Count

4,186→4,372

Distribute Authority Off the Four Pillars

Four pages hold more than half of all authority and the top 1% holds 73%: four single points of failure on an otherwise healthy site. Deep-linking from the pillars into the content beneath them, and interlinking that content, spreads equity across sections. This is editorial work, not structural, and no optimizer can do it alone.

editorial deep-linking from the four hub pages → No single page carries a structural share of the site

Top 1% authority share

73%→< 40%

Structure is Signal

Every previous episode in this series was a story of signal that failed to spread: content stranded on islands, traffic trapped in a well, a monolith with no internal seams, an archipelago with no ferries, a frozen periphery, a compound fracture, a centerless sprawl, a site that received visitors and refused to pass them on. In each case the structure itself was the problem, and the fix was to build the connections the site was missing.

Rotunda is the harder lesson, because its structure is not the problem. Every previous episode, from the gravity well of Episode 2 to the archipelago of Episode 4, had something visibly wrong with the shape of the graph. Rotunda has a graph almost any of those sites would envy: reachable, balanced, well-connected, one single body. It still funnels more than half of its authority onto four pages, and that single fact would never appear on a conventional audit, because every box that audit checks comes back green.

This is what topology reveals that page-level checks cannot. A site can have clean titles, valid schema, no broken links, no orphans, and a sensible hierarchy, and still rest its entire weight on four columns. Authority is not a property of any one page; it is a property of how the pages point at each other, and Rotunda’s point, overwhelmingly, at the same four. That is what the Digital MRI measures and surface-level SEO never will: not whether the building is well-made, but where its load is carried, and what happens to everything else when it all rests on so few.