The Five Lenses We Read Every Site Through

One graph, five fixed questions

When we turn a website into a graph, we can end up with thousands of nodes and tens of thousands of links. That is already far too much to read by eye. A picture of it looks like a hairball, and staring harder does not help. So we stop looking and start measuring.

The measuring is not freeform. We read every site through the same five lenses, in the same order, every time. That discipline is the point. If we invented a fresh set of questions for each site, we could never say one site is more fragmented than another, or that an industry trends toward a particular flaw. Because the lenses are fixed, the answers line up. A reading from one episode sits next to a reading from another, and the comparison means something.

The five lenses borrow their names from anatomy, because a website behaves more like an organism than a document — which is also why we call the examination a Digital MRI: Skeleton, Circulation, Organs, Health, and Nervous System. Each one asks a single question and answers it with a standard network-science measure. This note walks through all five, what each measures, and what a failing reading looks like.

Skeleton: how big and how connected

The question: how large is this thing, and how tightly is it held together?

The Skeleton lens measures the raw frame. Page count and link count give the size. Graph density (the share of all possible links that actually exist) and average path length (the typical number of clicks between any two pages) give the connectedness. Diameter, the longest of those shortest paths, marks the far corner.

A healthy skeleton keeps pages close as the site grows. The warning sign is a site that adds pages without adding the links to hold them together: density falls, the average path stretches, and content drifts to the edge where neither a person nor a crawler casually arrives. When the typical page sits more than a handful of clicks from everything else, the frame has outgrown its joints.

Circulation: where authority pools

The question: how does importance flow, and how few pages does the whole site lean on?

Not every page carries equal weight. Links act like endorsements, and endorsements compound, so some pages accumulate far more structural importance than others. We measure this with PageRank, the same idea that ranked the early web, and we look at how evenly or unevenly that importance is spread.

Then we measure betweenness centrality: how often a page sits on the shortest path between two others. Pages with high betweenness are the bridges that traffic routes through. The failure mode this lens catches is over-concentration. When a small handful of pages hold most of the authority and sit on most of the paths, the site has single points of failure. Thin or off-topic hubs then drag everything downstream with them, and removing a few pages would fragment the whole graph.

Organs: does the math match the org chart

The question: does the content sort itself into clear topical neighborhoods, or is it one undifferentiated mass?

Left to its own links, a site sorts itself into clusters: groups of pages that link densely among themselves and sparsely to everything else. Finding those clusters is called community detection, and the cleanliness of the split is scored by modularity. We cover the mechanics of how this works in a forthcoming note on community detection; here it is one of the five readings.

A good organ structure looks like a real org chart: several balanced neighborhoods, each a coherent topic. Two patterns fail the lens. One is a single mega-community swallowing most of the site, which means the content never really separated into topics. The other is a scatter of tiny singleton clusters, which means pockets of content broke off and now sit on their own, related to nothing.

Health: the pages nothing reaches

The question: how much content is dying in isolation?

This is the lens that finds stranded content, and it counts three distinct things. Island rate is the share of pages cut off from the main body of the site entirely. Orphans are pages with no inbound editorial link: they exist, but nothing points to them. Dead-ends are the reverse: pages that receive links but offer none onward, so the trail stops the moment you arrive.

The important finding, which holds across very different sites, is that orphans and dead-ends are independent problems. A site can have almost no orphans and a third of its pages dead-ending, or the exact reverse. They feel like the same flaw and they are not, so fixing one does nothing for the other. When the island rate climbs past a few percent, a real slice of the site has gone structurally invisible: the content is published, but nothing leads there.

Nervous System: wiring, not just rooms

The question: is this a well-designed building, or a pile of rooms that happen to share an address?

The final lens looks at how the whole thing is wired together. Depth measures how many clicks from the homepage a typical page sits; content buried too deep is hard for both people and crawlers to reach. Bridges are single links whose removal would split a section off from the rest, and a site held together by many such bridges is fragile by design. Participation coefficient measures how evenly a page’s links spread across different neighborhoods.

The pattern this lens catches is the sealed silo. Neighborhoods can each be internally healthy and still never link across to one another, so an agent that lands in one section has no editorial path into the next. Good wiring is shallow and cross-connected: neighborhoods that reach each other, not just rooms stacked behind one long corridor.

Why the lenses stay fixed

Any one lens can read green while another reads red. A site can have a sound skeleton, dangerously concentrated circulation, clean organs, poor health, and a silent nervous system, all at once. That is exactly why we keep all five and keep them in order. A single number would hide the trade-offs; five readings show where a site is strong and where it is one missing link away from a problem.

Holding the framework fixed is what turns individual readings into a series. The same five questions asked of one site become a profile, as in a single-site teardown like The Terminus; asked of ten sites in an industry, they become a benchmark, as in an industry report. That is what a Digital MRI produces for one site and what the reports aggregate across a cohort, and it is what every episode in this series is quietly running underneath.