Fragmentation Is Not Reach  🔗 **Source:** [Fragmentation Is Not Reach](https://www.jonoalderson.com/conjecture/fragmentation-is-not-reach/) > [!note]+ Document Notes > > [!info] Analysis generated 2026-08-27 by Claude (Opus 5) via readwise-deep-read. > > **Source:** jonoalderson.com — https://www.jonoalderson.com/conjecture/fragmentation-is-not-reach/ · **Published:** 2026-08-27 · **~3,141 words** · article · **Author:** Jono Alderson (technical SEO / web strategy). > > Register key: [STATED] author-stated · [EVIDENCE] shown · [INFERENCE] reviewer inference · [RECOMMEND] reviewer recommendation. > > Full article read; 8 section headings, no links captured. > > # 🏷️ Title: Fragmentation is not reach > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References [EVIDENCE] No outbound links captured. Named only: **Procter & Gamble** (Tide, Gillette, Pampers) as the good-fragmentation case. The argument runs on reasoning, not citation. ## 📝 Summary **The thesis, in the title:** owning more web properties is not the same as reaching more people. [STATED] *"Fragmentation is not reach. Sometimes it is just fragmentation."* **The historical setup.** [STATED] *"There's a reason supermarkets charge brands for shelf space."* Marketing treated distribution as occupying space — another shop, billboard, reseller, domain. [STATED] *"The web removed the shelf, but we kept its logic."* And for a while search rewarded it: more domains, more chances to rank. **Why it stopped working.** [STATED] *"Google has little incentive to fill a result set with several versions of the same company saying roughly the same thing."* An ambiguous query wants a shop, a review, a news story, a guide — *different answers to different needs.* Five sites from one company are *"five candidates for the same answer."* **The two mechanisms that make consolidation win:** - **The fragmentation tax** — [STATED] *"every new boundary strands some proportion of the organisation's existing capability on the other side."* The same engineers, content team, SEO team now support N platforms; *"none of those resources multiplied when the domain was registered. They were divided."* - **Compounding** — [STATED] *"Fix a template once and thousands of pages get better… 2 + 2 = 5, while 1 + 1 + 1 + 1 = 3. A connected system creates value in the joins. A fragmented one pays to recreate them."* **The AI escalation, which is the piece's forward-looking core.** [STATED] *"The emerging web is not giving us smaller shelves. It is replacing the shelf with a decision."* Search showed a list; a generative system compresses hundreds of sources into three sentences; an agent *"eventually has to pick something."* [STATED] *"You can still get traffic from position three. You cannot be the third service an agent chooses when it only needs one."* **Bottom line.** [INFERENCE] One of the sharpest pieces in this batch — a real reframe (fragmentation as a *tax*, not reach), a mechanism (semantic distance should set architectural distance), and an honest counter-section on when separation is right. It converges with the AI-search claim running through this batch and it directly names a WordPress-ecosystem pattern. ## 📖 Main Content ### The core reframe: architecture is an expense, not a footprint [STATED] *"More shelves. More products. More chances to win. Except search stopped behaving like a supermarket."* [INFERENCE] The move that makes the whole piece work is turning a thing organisations count as an *asset* — number of web properties — into a *liability*. Every property is recorded on a slide as reach; Alderson recomputes it as a standing cost against a fixed pool of engineering, content, SEO and reputation. [STATED] *"They duplicate technology, governance, content, analytics, reputation-building and search effort, with more places for everything to drift."* [INFERENCE] This is structurally the **cost-allocation** argument from the Dries pieces in this batch, applied to marketing architecture. Dries: architecture silently decides where infrastructure cost falls. Alderson: brand architecture silently decides where a fixed capability pool gets *divided*. **Both say a technical decision, made for local reasons, allocates a scarce resource nobody was tracking** — and becomes visible only years later as decay. ### "We separate things, then spend money teaching them to behave as though they were connected" [EVIDENCE] The strongest single observation, and it is empirical rather than rhetorical. Organisations split into N sites, then build [STATED] *"a design system to make the sites look related, data feeds to keep product information aligned, cross-domain analytics to reconstruct the customer journey, single sign-on to disguise the joins, and shared navigation to make the estate feel less fragmented."* [INFERENCE] Every item on that list is a second expense incurred to undo the first. **The fragmentation created work, and re-connecting created more.** This is the tell that the separation was not load-bearing: if you have to rebuild the connections you severed, the boundary was ornamental. [RECOMMEND] The diagnostic falls straight out — *count the systems that exist only to make separate properties feel connected; that is the bill for a boundary that earned nothing.* ### Authority cannot be photocopied — the piece's best single argument [STATED] Several overlapping properties, someone suggests cross-linking them for a *"compounding effect."* The rebuttal: *"nothing has been multiplied. Whatever reputation, links, recognition and trust the organisation has earned still has to come from somewhere outside the system."* And the image that lands it: [STATED] *"the digital equivalent of moving money between four bank accounts and congratulating yourself on having quadrupled your wealth."* [INFERENCE] This is a genuine logical correction, not just a metaphor. **Authority is a claim external parties make about you** — links, mentions, demand. Redistributing it internally changes its location, not its quantity. [STATED] *"Authority can be distributed. It cannot be photocopied."* [INFERENCE] And he closes the loop precisely: [STATED] *"create several sites; weaken them by dividing attention and reputation; link them together to compensate; then treat those links as evidence that the division was worthwhile. You have built a dependency loop and mistaken it for strategy."* — naming the circularity where an intervention's cost is misread as its justification. [INFERENCE] Same shape as the GraphRAG entity-duplication finding in this batch: split one entity into `Acme` / `AcmeBusiness`, and the traversal (human or machine) misses what's attached to the twin. **Fragmenting an entity degrades exactly the retrieval the fragmentation was meant to serve.** ### The AI section is where the argument stops being about SEO [STATED] *"AI makes the compression harsher because the interface no longer has to show a list of destinations at all."* Traditional search was *"relatively generous"* — organic, ads, shopping, maps, videos, snippets. A generative system retrieves from hundreds of sources and returns three sentences. An agent asked to buy, book, or call an API *"has to decide which product, provider, catalogue, account or version of Acme to use."* [INFERENCE] The load-bearing distinction: [STATED] *"That is a different failure mode from ranking third. You can still get traffic from position three. You cannot be the third service an agent chooses when it only needs one."* **Ranking is a distribution; selection is a single draw.** Fragmentation was survivable when the interface showed ten results; it is fatal when the interface returns one action. [INFERENCE] This is the **same claim Daryn Strauss made** in this batch — people increasingly get answers from AI tools rather than from a results page — arriving from the opposite direction. Strauss: expose an MCP server so the agent can read you. Alderson: **be one coherent entity so the agent can *select* you.** [INFERENCE] They're two halves of one instruction: *readable* (Strauss) and *unambiguous* (Alderson). And it pairs with the Cloudflare post from this batch, which did exactly what Alderson prescribes — a single blog, one MCP surface, four curated tools — the consolidated-and-agent-readable case in practice. ### Semantic distance should determine architectural distance [STATED] The operational rule, stated repeatedly: *"Semantic distance should usually determine architectural distance. Things which are genuinely far apart can benefit from stronger boundaries. Things which are close together usually benefit from being represented as parts of a coherent whole."* [EVIDENCE] With the concrete test: [STATED] *"If you need a brand architecture diagram to explain the names, the names are not doing enough work."* And the naming critique — *Business, Pro, Enterprise, Solutions, Services* — [STATED] *"corporate soup outside the organisation… Somebody searching for 'Acme business login' does not care which division owns the account system… They want the bloody login."* [INFERENCE] This is the most portable thing in the piece and generalises well past marketing. **The distance between two concepts should set the strength of the boundary between them, and a boundary stronger than the distance justifies is pure overhead.** It applies to codebase module boundaries, documentation structure, API surface design — anywhere someone draws a line that the underlying difference doesn't support. ### The honesty section is what makes it credible [EVIDENCE] Rather than a consolidation crusade, he spends a full section on when separation earns its cost: [STATED] *"Procter & Gamble does not need Tide, Gillette and Pampers to pretend they are one thing. A luxury product should not inherit the commercial framing of the budget version. A regulated financial service may need to behave very differently from the consumer app."* [EVIDENCE] And he pre-empts the obvious misreading: [STATED] *"This is not a promise that moving everything onto one domain will magically improve rankings. Consolidations can destroy useful distinctions, botch migrations and create a vast corporate landfill with excellent canonicalisation."* [INFERENCE] The distinction he draws — [STATED] *"federated where difference creates value and consolidated everywhere else"* — is the actual thesis, and it's a spectrum, not the binary the title implies. **The good version keeps the surface varied and the machinery shared:** [STATED] *"Several brands can share product data, identity systems, analytics, infrastructure, governance and content operations while presenting genuinely different customer experiences."* ### The politics section names why this is hard [STATED] *"Separate properties acquire separate owners, targets and incentives… Nobody wants their site to become a subsection of somebody else's, because ownership of a website feels suspiciously like ownership of importance."* And the line that generalises: [STATED] *"Sprawling web estates are rarely just collections of websites. They are maps of the compromises that created them."* [INFERENCE] This is Conway's Law stated for web architecture — structure mirrors the org chart — and it's why the technical argument, however correct, rarely wins. [INFERENCE] It connects to the Dries governance thread: there, *whoever funds the infrastructure controls it*; here, *whoever owns the domain owns the status*. **Both describe technical structure ossifying organisational power, and both note the technical fix is really a political negotiation.** ## 🗒️ Notes - **The `/business/` prescription is the concrete takeaway:** [STATED] *"Audience segmentation is a repeat offender. Business customers may need different pricing, proof points, purchasing journeys, support and language. None of those things inherently requires another domain. They may simply require a good /business/ section."* [INFERENCE] A URL path versus a domain is the cheapest possible boundary, and it's almost always sufficient. - **"Technical separation does not require search separation"** — the sharpest sub-point. [STATED] *"A marketing page, login screen, dashboard, support centre and application may need different experiences and technology. They do not all need to compete as public discovery surfaces."* [INFERENCE] Backend architecture and public-discovery architecture are independent axes, and conflating them is a common, expensive error. - **The six-question test at the end is a usable artifact:** does separation make the proposition clearer, help customers understand what they're buying, occupy a different market position, create a different *kind* of answer in search, give machines a clearer entity to select, or enable legal/commercial behaviour consolidation would constrain? [STATED] *"If the answer is no, the boundary is probably ornamental."* - **"There is no prize for making the decision harder"** — the whole AI argument in one line. - **The fragmentation-into-the-machine-layer point is underdeveloped and important:** [STATED] *"The fragmentation has moved into the machine layer: tools, services, feeds, APIs and agents which overlap just enough to create ambiguity without creating useful choice."* [INFERENCE] This is the same disease at the MCP/API level — and it's exactly the risk in a proliferation of overlapping MCP servers, which the Kent Dodds material in this batch touched from the curation side. - **Voice:** British, blunt, funny ("the bloody login", "a different shade of blue"). [INFERENCE] The profanity is doing work — it stands in for the actual user's impatience, which is the whole point about ambiguous boundaries. - **Directly names the WordPress-shaped pattern** without naming WordPress: multiple sites, a shared design system bolted on afterward, cross-domain analytics to reconstruct journeys. [INFERENCE] This is the multisite-vs-separate-installs question and the "should this be a new site or a section" question that every agency faces. ## 📚 References [EVIDENCE] No citations and no links — the argument is entirely first-principles reasoning plus one named example (P&G). [INFERENCE] For an opinion/strategy essay this is appropriate; the claims are logical rather than empirical, and the reader can test each against their own experience. The one place data would strengthen it is the central AI claim — *the source pool grows while the visible answer shrinks* — which is asserted as direction, plausibly, without measurement. ## 🗣️ Paraphrase Supermarkets charge for shelf space because more facings mean more visibility. Marketing spent its history occupying space — another shop, billboard, reseller, domain. The web removed the shelf but kept the logic, and for a while search rewarded it: more domains, more chances to rank. So big organisations accumulated websites — corporate, product, division, acquired brand, permanent "campaign" microsite, the thing the CEO liked in 2018. Then search stopped behaving like a supermarket. Google has no reason to fill a result set with five versions of the same company. Different needs want different answers — a shop, a review, a guide — not five interchangeable candidates for one answer. AI sharpens this: it retrieves from a huge pool and compresses to a few sentences, and an agent may just choose for you. The source pool grows while the visible answer shrinks. Fragmentation carries a tax. Every new site needs hosting, templates, governance, security, content, SEO — and worse, it divides your fixed team across more platforms. Then you build design systems, data feeds and single sign-on to make the separated things behave as if connected. You separate, then pay to reconnect. Consolidation compounds instead: fix one template, thousands of pages improve. 2+2=5; 1+1+1+1=3. But it's not a crusade for one website — botched consolidation is its own failure. The rule is: semantic distance should set architectural distance. Things genuinely far apart earn boundaries; things close together should be parts of a coherent whole. If you need a diagram to explain your brand names, they aren't working. Cross-linking your own sites doesn't multiply authority — it just moves it between accounts. Authority comes from outside; you can distribute it, not photocopy it. And in an AI-mediated journey, ranking third still gets traffic, but you can't be the third service an agent picks when it needs one. Some boundaries earn their keep — P&G shouldn't merge Tide and Gillette. Federate where difference creates value; consolidate everywhere else. Every boundary should have to justify the compounding advantage it gives up. Fragmentation is not reach. Sometimes it's just fragmentation. ## 📰 Newsletter **The best strategy essay I've read this week, and it lands on something every agency and multi-brand org is getting wrong.** Jono Alderson's argument: owning more websites is not reaching more people. *"The web removed the shelf, but we kept its logic."* **The reframe is turning an asset into a liability.** Organisations count web properties as reach; Alderson recomputes each one as a standing cost against a *fixed* pool of engineering, content, SEO and reputation. *"None of those resources multiplied when the domain was registered. They were divided."* He calls it the **fragmentation tax**, and the tell is unmistakable: *"We separate things, then spend money teaching them to behave as though they were connected"* — design systems, data feeds, cross-domain analytics, single sign-on, all built to undo the split you just paid for. **The best single argument: authority cannot be photocopied.** Cross-linking your own sites to boost them is *"the digital equivalent of moving money between four bank accounts and congratulating yourself on having quadrupled your wealth."* Authority is a claim outsiders make about you — links, mentions, demand. You can move it around; you can't multiply it. **And the AI section is where it stops being about SEO.** Search showed a list; a generative system compresses hundreds of sources to three sentences; an agent has to *pick one*. *"You can still get traffic from position three. You cannot be the third service an agent chooses when it only needs one."* **Ranking is a distribution; selection is a single draw** — and fragmentation that was survivable in a list is fatal in a selection. **The usable rule:** semantic distance should set architectural distance. Business customers need different pricing and proof points — *"None of those things inherently requires another domain. They may simply require a good /business/ section."* If you need a brand-architecture diagram to explain your names, the names aren't working. **It's not a consolidation crusade** — P&G shouldn't pretend Tide and Gillette are one thing, and botched consolidation is its own failure. The thesis is *"federated where difference creates value and consolidated everywhere else."* Keep the surface varied; share the machinery. Two connections worth flagging: this is **the same AI-search shift Daryn Strauss described** from the other side — she says expose an MCP server so agents can read you; Alderson says be one coherent entity so agents can *select* you. And **the Cloudflare blog migration in this batch is the prescription in practice** — one site, one MCP surface, four curated tools. ## ✅ Action Items #### DevRel - **Developer advocacy:** [RECOMMEND] The "authority cannot be photocopied" argument is a talk on its own — it's the cleanest debunk of internal-crosslinking-as-strategy going. - **Education:** [RECOMMEND] Teach **semantic distance sets architectural distance** as a general design heuristic, not an SEO tactic — it applies to module boundaries and doc structure identically. - **Developer experience:** [INFERENCE] "Technical separation does not require search separation" is the reusable correction — backend architecture and public-discovery architecture are independent axes. - **Developer marketing / Community management / Cross-project:** Not applicable. #### OSPO / Community - **Governance:** [INFERENCE] *"Sprawling web estates are maps of the compromises that created them"* is Conway's Law for web architecture. Any consolidation is a political negotiation over status, not a technical project — plan for that. - Other items: Not applicable to this source. #### WordPress - **Themes / Plugins / Meta:** [INFERENCE] The multisite-vs-separate-installs and new-site-vs-new-section questions are exactly what this essay adjudicates. [RECOMMEND] The default question should be *"what gets materially better because this exists separately?"* not *"why shouldn't this have its own site?"* - **Documentation:** [INFERENCE] The six-question boundary test is directly adoptable as agency/site-architecture guidance. - **Marketing:** [INFERENCE] The AI-selection argument bears on how project properties (Learn, Showcase, News, .org) present as discoverable surfaces — several overlapping entities or one coherent one. - Other teams: Not applicable. #### The WPOCC - Not applicable to this source. ### Personal / courtneyr.dev [RECOMMEND] Directly relevant to the owned-domain thread running through this batch. - [INFERENCE] The essay is the strongest available argument for **consolidation onto one identity** rather than scattering across a personal site, a projects site, a newsletter domain, and so on. One well-understood entity is easier for both humans and agents to select. - [INFERENCE] It pairs with the Graze/portability finding and the AI-search claim: the durable asset is **one coherent, identifiable, agent-readable entity on a domain you own** — Alderson supplies the "coherent" half, Strauss the "agent-readable" half. - [INFERENCE] Practical test for any "should this be its own site?" impulse: run the six questions. Different audience segment alone (WordPress vs accessibility vs FAIR writing) is a `/section/`, not a domain. ### FAIR - [INFERENCE] One structural echo: the essay's *"federated where difference creates value and consolidated everywhere else"* is precisely the design tension in a federated package system — federation earns its cost only where the difference is real, and adds overhead everywhere else. The same discipline Alderson applies to brand architecture applies to distribution architecture. ## 🧪 Hypotheses 1. **H1 — Consolidation improves aggregate visibility.** Organisations consolidating overlapping properties see higher combined organic visibility than the sum of the fragmented properties, holding content constant. 2. **H2 — Reconnection cost is diagnostic.** The number of systems built to make separate properties feel connected (design system, data feed, SSO, cross-domain analytics) predicts whether the separation destroyed value. 3. **H3 — Agent selection is winner-take-most.** In agent-mediated purchasing, a single coherent entity is selected at a materially higher rate than any one of several overlapping sibling entities from the same parent. 4. **H4 — Boundaries track org chart, not semantics.** In multi-property estates, boundary placement correlates more strongly with organisational ownership than with semantic distance between propositions. 5. **H5 — Naming suffixes reduce task success.** Users complete tasks (e.g. finding a login) more slowly and with more errors when a brand uses Business/Pro/Enterprise/Solutions suffixes than when it uses path-based sections. 6. **H6 — Fragmentation degrades machine reconciliation.** LLM/agent systems more often mis-attribute products or claims when a company's information is split across sibling domains than when consolidated. ## 📊 Methodology - **H1 — the central claim, and testable.** Matched before/after study of organisations that consolidated, combined-visibility measured, with a control group that didn't. Confound: consolidation correlates with general marketing competence — needs a competence covariate or a within-org interrupted time series. - **H2 — cheap and mechanical.** Inventory the "reconnection" systems across a sample of multi-property estates; correlate count against a value proxy (combined authority, journey completion). [RECOMMEND] A high correlation would turn Alderson's rhetorical tell into a metric. - **H3 — the forward-looking one and hard to run today.** Agent-mediated purchase simulations against consolidated vs fragmented catalogs; measure selection rate. Confound: agent behaviour is model-specific and changing fast — findings would date quickly. - **H4:** Code a sample of estates on boundary placement vs org structure vs semantic distance; test which predicts. Mechanical once the coding rubric is fixed. - **H5 — a clean usability experiment.** Same task, suffix-brand vs path-section conditions; measure time and error. Runnable at small scale and directly tests the "want the bloody login" claim. - **H6 — immediately runnable.** Query LLMs about a company whose info is split across sibling domains vs consolidated; measure attribution errors. Same design as the GraphRAG entity-resolution test in this batch — they could share a harness. ## ❓ Questions - **Is the AI-search direction actually measurable yet, or still a plausible forecast?** The essay asserts *"the source pool gets larger while the visible answer gets smaller"* as direction; no measurement is offered. - What's the *cost* threshold at which a boundary pays off? He says "sufficiently valuable" can carry the machinery but never quantifies the line. - [INFERENCE] Does consolidation risk (botched migration, destroyed distinction) exceed fragmentation cost more often than the essay implies? He acknowledges it and doesn't weight it. - How do you consolidate against the political forces he describes? The essay names the org-chart problem and offers no path through it. - [INFERENCE] Does the six-question test survive contact with a genuinely ambiguous case, or does everything pass question six (legal/operational) if you try hard enough? ## 🔮 Future Research - **Whether agent selection is winner-take-most in practice** — the essay's central forward claim, and the one that determines whether fragmentation is merely wasteful or actually fatal. - **The reconnection-cost metric** as a diagnostic for value-destroying separation. - **Usability cost of internal-taxonomy naming** (Business/Pro/Enterprise) — measurable and apparently unstudied. - [INFERENCE] **Machine reconciliation of fragmented entities** — the same problem GraphRAG names, applied to brand estates rather than knowledge graphs. ## 💡 Implications - **Practice.** [RECOMMEND] Default to a `/section/`, not a domain. Run the six-question test before any new property. Count your reconnection systems — they're the bill for boundaries that earned nothing. - **Policy.** [INFERENCE] "Semantic distance sets architectural distance" is a governance rule, not just an SEO one: every boundary carries burden of proof, and *"a URL path is considerably cheaper than a new brand."* - **Security.** [INFERENCE] Not central, but every additional property is additional attack surface, and the "slow decay" the essay describes — aging templates, drifting standards — is exactly where security debt accumulates. The Avada/miniOrange stories in this batch are what neglected long-tail properties become. - **Education.** [EVIDENCE] The essay is a model of the form: reframe (asset→liability), mechanism (tax, compounding), forward projection (ranking→selection), honest counter-case, usable test. The structure is worth teaching independent of the content. - **AI/discovery.** [INFERENCE] The most durable takeaway: **the emerging interface returns an action, not a list**, and *"you cannot be the third service an agent chooses when it only needs one."* Design for selectability, not appearance. ## 🧠 Learning Styles Assessment **Reading level:** [INFERENCE] Flesch–Kincaid ~grade 11–12 — long, well-built sentences, minimal jargon, one metaphor doing most of the explanatory work. Hand estimate, ±1.5. [EVIDENCE] SEO terms (canonicalisation, structured data) appear unglossed but rarely; the argument is legible to a non-specialist. - **👀 Visual:** [INFERENCE] No images captured. The Acme suffix list and the six-question test are both naturally visual and delivered as prose; a boundary-decision flowchart would carry the whole essay. - **🔊 Aural:** [EVIDENCE] Strong rhythmic prose — *"More shelves. More products. More chances to win."* — built for reading aloud. The blunt asides land better spoken than skimmed. - **🗣️ Verbal:** [EVIDENCE] The dominant strength. *"Authority can be distributed. It cannot be photocopied."* *"You have built a dependency loop and mistaken it for strategy."* *"There is no prize for making the decision harder."* Compression is the essay's craft. - **🧩 Kinesthetic:** [EVIDENCE] The six-question test is directly applicable; the `/business/` example is copyable. [INFERENCE] More actionable than most strategy writing. - **🧠 Logical:** [EVIDENCE] Rigorous — the photocopy argument and the ranking-vs-selection distinction are genuine logical corrections, not analogies dressed as arguments. [INFERENCE] The one soft spot: the AI direction is asserted, not evidenced. - **🧑‍🤝‍🧑 Social:** [EVIDENCE] The politics section is the human core — status, territory, "ownership of importance." No individual quoted, but the organisational dynamics are drawn from life. - **🧘 Solitary:** Self-contained; applies to your own estate immediately. [EVIDENCE] No links or further reading. - **🌿 Naturalistic:** [EVIDENCE] "Decay", "drift", "leak", "landfill" — an ecosystem/entropy vocabulary runs throughout, and it's doing real work: fragmentation as slow decay of a shared resource. ## 📊 Accessibility Assessment Assessed against WCAG 2.2 AA, from captured markup only. [INFERENCE] Scope limit: text/structure as retrieved; focus, contrast, reflow not assessed. - **Screen reader semantics:** [EVIDENCE] **Eight descriptive `<h2>` headings**, well-nested, genuinely informative (*"Authority cannot be photocopied"*, *"From ranking to selection"*). **Passes 2.4.6** and is one of the best-structured articles in this batch. [INFERENCE] The Acme suffix list renders as prose rather than a `<ul>`; minor. - **Keyboard / Contrast / Reflow:** Not testable from markup. - **Media:** Not applicable — no audio/video, and no images captured (so **1.1.1 indeterminate** — if the header carries a decorative image its alt state is unknown). - **Standards:** 2.4.6 (AA) **satisfied**; 2.4.4 (A) satisfied — no ambiguous links captured; 1.1.1 (A) indeterminate. ## 🌍 DEIB Assessment - **Cultural breadth:** [INFERENCE] Enterprise-Western framing throughout (P&G, corporate brand estates). The argument generalises but the examples assume large, well-resourced organisations. - **Inclusive language:** [EVIDENCE] Neutral; the mockery is aimed at corporate naming conventions, not people. *"A different shade of blue"* is affectionate about the org, not derisive of anyone in it. - **Neurodiversity:** [EVIDENCE] **Strong** — eight clear headings, short paragraphs, one idea each, memorable anchors. The six-question list is an explicit decision aid. [INFERENCE] Genuinely easy to navigate and re-enter. - **Representation:** [EVIDENCE] Single-author opinion; no one quoted. Appropriate for the form. - **Psychological safety:** [EVIDENCE] Handled well — the honest counter-section means a reader with a legitimately fragmented estate isn't told they're wrong, and *"the burden of proof should still increase with every new boundary"* leaves room for justified separation. [INFERENCE] The tone is confident but not contemptuous of the people who built the sprawl. ## ✅ Logical Validity **Strong reasoning:** - **Authority cannot be photocopied** is a real logical correction — redistribution ≠ multiplication — supported by a mechanism (authority originates outside the system) rather than an analogy. - **Ranking vs selection** identifies a genuine change in the shape of the outcome: a distribution (traffic from position three) vs a single draw (one action). The failure modes are categorically different, and he's right that fragmentation crosses from wasteful to fatal at that boundary. - **Semantic distance sets architectural distance** is a clean, falsifiable design principle with a stated test. - **The counter-section earns credibility** by conceding the strongest opposing case (P&G, regulated services, botched consolidation) fully rather than strawmanning it. **Fallacies / weaknesses:** - **Asserted direction, not measured.** *"The source pool gets larger while the visible answer gets smaller"* is the load-bearing forward claim and is offered as obvious direction with no data. Plausible; unproven. - **Binary title, spectrum argument.** "Fragmentation is not reach" implies a dichotomy the body explicitly rejects (*"federated where difference creates value"*). The memorable framing overstates the actual, more careful position. - **Consolidation risk under-weighted.** He acknowledges botched migrations and destroyed distinctions in one sentence, then the essay's momentum runs entirely toward consolidation. The failure mode is named and not balanced. - **No cost threshold.** "Sufficiently valuable" properties can carry the machinery — but where the line falls is exactly the decision a reader needs, and it's left to judgement. **Tone:** [EVIDENCE] Confident, funny, British-blunt. [INFERENCE] Calibration is good — the strong claims (photocopy, selection) are the well-argued ones; the hedged claims (AI direction, "not a promise that consolidation improves rankings") are the uncertain ones. The hedging sits where the evidence is thinnest, which is the right pattern. ## 🛡️ DARVO Analysis **No DARVO pattern present.** No accusation, no accused party — a strategy essay. - **Deny:** None. Affirms that fragmentation was once genuinely rewarded (*"There was enough truth in this to make the strategy seductive"*) rather than pretending it was always wrong. - **Attack:** Aimed at **practices** (accumulating domains, internal crosslinking, corporate naming), never people. The teams that built the sprawl are treated as having made locally-reasonable decisions — *"decisions that were individually reasonable"* is the same generosity Dries used in this batch. - **Reverse victim/offender:** None. [INFERENCE] Adjacent, not DARVO: the author is a technical-SEO consultant, and "consolidate and simplify" is broadly the kind of work such a consultant sells. The argument stands on its own logic and the commercial alignment is worth noting only for calibration. ## 🎯 Conclusion **Bottom line.** [INFERENCE] One of the strongest pieces in this batch — a true reframe, two clean mechanisms, and an honest counter-case. 1. **Fragmentation is a tax, not reach.** [STATED] *"Every new boundary strands some proportion of the organisation's existing capability on the other side."* The tell that a boundary earned nothing: *"We separate things, then spend money teaching them to behave as though they were connected."* Count those reconnection systems — that's the bill. 2. **Authority cannot be photocopied.** [STATED] Cross-linking your own sites is *"moving money between four bank accounts and congratulating yourself on having quadrupled your wealth."* Authority comes from outside; you can distribute it, not multiply it. [INFERENCE] Same shape as the GraphRAG entity-duplication problem: fragmenting an entity degrades the retrieval the fragmentation was meant to serve. 3. **Ranking became selection.** [STATED] *"You can still get traffic from position three. You cannot be the third service an agent chooses when it only needs one."* [INFERENCE] The interface now returns an action, not a list — and fragmentation that was survivable in a list is fatal in a selection. This is **the AI-search shift from Daryn Strauss, from the other side**: she says be *readable* by agents; Alderson says be *selectable* by them. The Cloudflare migration in this batch is the prescription in practice. 4. **The rule: semantic distance sets architectural distance.** [STATED] *"If you need a brand architecture diagram to explain the names, the names are not doing enough work."* Business/Pro/Enterprise suffixes are *"corporate soup"*; the user *"wants the bloody login."* Almost every internal distinction is a `/section/`, not a domain. **For your own work.** [INFERENCE] This is the strongest argument in the batch for consolidating onto **one coherent, owned, agent-readable identity** rather than scattering across properties. Combined with Strauss (be readable) and the Graze portability finding (own the address), the instruction is consistent: *one entity, one domain you control, legible to the thing that replaced the results page.* Different audience segments — WordPress, accessibility, FAIR — are sections, not sites. **One caution.** The forward claim — that AI compression makes fragmentation fatal rather than merely wasteful — is asserted as direction, not measured. It's plausible and it converges with three other sources in this batch, but it's the load-bearing prediction and it's still a prediction. **Next steps:** 1. Run the **six-question test** on any "should this be its own site?" decision; default to a section. 2. **Count your reconnection systems** as the diagnostic for value-destroying boundaries. 3. Design public surfaces for **selectability, not appearance** — one obvious doorway, routed by intent. 4. Treat the AI-selection claim as a well-supported forecast, not an established fact. <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/Fragmentation Is Not Reach.md|Fragmentation Is Not Reach by Jono Alderson]] ## Highlights - We separate things, then spend money teaching them to behave as though they were connected. ([View Highlight](https://read.readwise.io/read/01m12tjnzhnqv72x4kapbqh106)) - Note: THE TELL THAT A BOUNDARY EARNED NOTHING, AND THE STRONGEST SINGLE OBSERVATION IN THE ESSAY. Organisations split into N sites, then build "a design system to make the sites look related, data feeds to keep product information aligned, cross-domain analytics to reconstruct the customer journey, single sign-on to disguise the joins, and shared navigation." EVERY ONE OF THOSE IS A SECOND EXPENSE INCURRED TO UNDO THE FIRST. If you have to rebuild the connections you severed, the separation was ornamental. DIAGNOSTIC THAT FALLS STRAIGHT OUT: count the systems that exist only to make separate properties feel connected — that is the bill for a boundary that earned nothing. THE CORE REFRAME IS TURNING AN ASSET INTO A LIABILITY: orgs count web properties as reach; Alderson recomputes each as a standing cost against a FIXED pool of engineering, content, SEO and reputation — "none of those resources multiplied when the domain was registered. They were divided." He calls it the FRAGMENTATION TAX. This is STRUCTURALLY THE DRIES COST-ALLOCATION ARGUMENT from this same batch, applied to marketing architecture: a technical decision made for local reasons silently allocates a scarce resource nobody was tracking, and becomes visible only years later as decay. The consolidation counter-mechanism: "Fix a template once and thousands of pages get better… 2 + 2 = 5, while 1 + 1 + 1 + 1 = 3. A connected system creates value in the joins. A fragmented one pays to recreate them." #atom/zettel/z1 - **Authority can be distributed. It cannot be photocopied.** ([View Highlight](https://read.readwise.io/read/01m12tknxwn9vxad9w7gq9ege5)) - Note: THE BEST SINGLE ARGUMENT — a genuine logical correction, not a metaphor. Someone suggests cross-linking your overlapping sites for a "compounding effect." The rebuttal: "nothing has been multiplied. Whatever reputation, links, recognition and trust the organisation has earned still has to come from somewhere OUTSIDE the system." AUTHORITY IS A CLAIM EXTERNAL PARTIES MAKE ABOUT YOU — links, mentions, demand. Redistributing it internally changes its LOCATION, not its QUANTITY. The image that lands it: "the digital equivalent of moving money between four bank accounts and congratulating yourself on having quadrupled your wealth." AND HE CLOSES THE CIRCULARITY PRECISELY: "create several sites; weaken them by dividing attention and reputation; link them together to compensate; then treat those links as evidence that the division was worthwhile. You have built a dependency loop and mistaken it for strategy." — naming the trap where an intervention's COST is misread as its JUSTIFICATION. THE OPERATIONAL RULE THAT GENERALISES PAST SEO: SEMANTIC DISTANCE SHOULD DETERMINE ARCHITECTURAL DISTANCE. Things genuinely far apart earn boundaries; things close together should be parts of a coherent whole. Test: "If you need a brand architecture diagram to explain the names, the names are not doing enough work." Applies to codebase module boundaries, doc structure, API surfaces — anywhere a line is drawn that the underlying difference doesn't support. The naming critique is the sharp end: Business/Pro/Enterprise/Solutions/Services is "corporate soup" — the user searching "Acme business login" "wants the bloody login," they don't want to reason about which division owns the account system. "A URL path is considerably cheaper than a new brand." Almost every internal distinction is a /section/, not a domain. #atom/zettel/z3 #action/h2 - That is a different failure mode from ranking third. You can still get traffic from position three. You cannot be the third service an agent chooses when it only needs one. ([View Highlight](https://read.readwise.io/read/01m12tk6r3as9sakfzpy0nfb5x)) - Note: RANKING BECAME SELECTION — AND THAT'S WHERE FRAGMENTATION CROSSES FROM WASTEFUL TO FATAL. "AI makes the compression harsher because the interface no longer has to show a list of destinations at all." Search was a DISTRIBUTION — ten results, several ways to appear, visibility felt spatial. An agent asked to buy/book/call-an-API has to make a SINGLE DRAW. Fragmentation that was survivable in a list is fatal in a selection: five overlapping properties (one has the brand, one the docs, one the links, one the fresh product data, one the detailed answer) become "ingredients in one judgement," and worse, they can DISAGREE — "the machine must do the integration work the organisation declined to do itself." SAME AI-SEARCH SHIFT AS DARYN STRAUSS THIS BATCH, FROM THE OPPOSITE DIRECTION. Strauss: expose an MCP server so the agent can READ you. Alderson: be one coherent entity so the agent can SELECT you. Two halves of one instruction — READABLE and UNAMBIGUOUS. And the Cloudflare blog migration this batch is the prescription IN PRACTICE: one site, one MCP surface, four curated tools. ALSO THE GRAPHRAG ENTITY-DUPLICATION SHAPE from this batch: split Acme into Acme/AcmeBusiness and the traversal (human or machine) misses what's attached to the twin — fragmenting an entity degrades exactly the retrieval it was meant to serve. FOR courtneyr.dev / THE OWNED-DOMAIN THREAD: strongest argument in the batch for consolidating onto ONE coherent, owned, agent-readable identity. Different audience segments (WordPress / accessibility / FAIR) are SECTIONS, not sites. "Design for selectability, not appearance." CAVEAT: the "source pool grows, visible answer shrinks" direction is asserted, not measured — load-bearing forecast, still a forecast, though it converges with three other sources this week. #atom/zettel/z2 #action/h1