For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
The published dematerialization-lock claim asserts that no $100B-class digital network has been vanquished within its own substrate, and the only kill vector is substrate redefinition above the layer. The piece names Yahoo, IBM, Nokia, and Blackberry as cases that look like counterexamples but are in fact substrate redefinitions. It does not run a systematic sweep. The question this node opens: when the sweep is run honestly, does the lock hold, and if so, why?
The lock holds, but only on the substrate-class the parent piece silently presupposes. Apparent counterexamples surface immediately when the sweep starts, and they cluster on a single boundary. The boundary is what defines what kind of digital network is locked.
Same-substrate displacements at $100B-class-adjacent scale that the parent piece does not address.
Internet Explorer to Chrome. Microsoft's browser held 95% global share at peak in 2002-2003. Chrome launched in 2008 and passed IE in May 2012. Microsoft was a $100B+ market cap parent. No substrate redefinition occurred: the web is still the web, the browser is still the browser, desktop is still where Chrome won first. Microsoft's distribution advantage was overwhelming (Windows ships IE). Chrome won anyway.
Yahoo Mail to Gmail. Yahoo Mail held majority webmail share through the early 2000s. Gmail entered in 2004 with no installed base and is now the global dominant webmail provider with several times Yahoo Mail's user count. Yahoo as parent peaked above $100B. No substrate redefinition occurred; both before and after, the substrate is web-accessed personal email.
Intel to AMD in server CPUs. Intel held above 95% datacenter CPU share for fifteen years. AMD's share rose from roughly 2% in 2017 to above 25% by 2024, with AMD's CEO citing 34% in late 2024 segment-revenue terms. Intel was a $290B+ market cap company at peak. The substrate is x86 server CPUs; AMD competed on the same substrate, not a redefined one. The mobile-CPU loss to ARM is substrate redefinition and is consistent with the parent's frame; the server-CPU loss to AMD is not.
Yahoo Search to Google Search inside the search substrate. Yahoo's search share peaked around 35% in 2000-2002. Google passed Yahoo's share around 2003 and now holds above 90% globally. The lock-claim could argue Yahoo Search was below 10x dominance when Google entered. The point is that search-substrate dominance changed hands at multiple-billion-user scale without substrate redefinition.
Skype to Microsoft Teams and Zoom. Skype was the dominant consumer video-call substrate through 2013. Microsoft acquired it for $8.5B in 2011. Teams and Zoom displaced it inside the same substrate (real-time video communication) over 2017-2021. Skype was sub-$100B as a standalone but mechanistically informative. Microsoft was both the parent of the displaced product and the displacer.
Five candidates. None is a substrate-redefinition story. All are same-substrate displacements at scales where the parent's lock-claim should have held.
The lock-claim's mechanism says: marginal user value under network effects is positive, marginal cost is zero, the curves never cross, runner-up has no niche. That argument requires direct user-to-user network value, where value flows from the existence of other users to each user.
Facebook has this: my account is more valuable because yours exists. Bitcoin has this: settlement value scales with users-and-capital on the network. WhatsApp has this. Discord has this. eBay and Amazon have this two-sidedly, each side attracting the other.
The five candidates above do not, or have it only weakly:
The five "counterexamples" cluster on one side of a line: indirect network effects via developer ecosystem, software compatibility, training-data flywheels, or co-presence within an event. The substrates Saylor names — Google Search, Facebook, Amazon, Apple, bitcoin — sit on the other side: direct user-to-user value coupling.
The dematerialization-lock applies to digital networks with direct user-to-user network value. It does not apply to dominance forms based on indirect ecosystem effects, software-compatibility positions, or scale-of-data advantages.
This is a domain filter, not a refutation. The parent piece's mechanism description ("marginal user under network effects") was always domain-restricted; the parent did not name the restriction. With the restriction explicit, every apparent counterexample resolves: IE was never locked because browsers don't have user-to-user network value; Yahoo Mail was never locked because webmail is interoperable; Intel was never locked because x86 is licensable and software compatibility is the only network effect; Yahoo Search was contested at the data-flywheel level which is weaker than user-to-user; Skype's dominance was call-local and didn't compound across calls.
The five non-locked categories are exactly the categories where dominance has historically been fragile:
The Saylor list, by contrast, has held position for roughly two decades each at this point: Google and Amazon since 1998-2000, Facebook since 2008, Apple's iOS since 2008, bitcoin since approximately 2013-2014. The dominance lengths differ because the mechanism differs.
The domain filter would be tautological if it could only be applied retrospectively (whichever survived was direct-coupling, whichever was displaced was indirect). It is not. Two questions decide before the contest concludes:
If you randomly subtract one user from the network, does each remaining user's per-period value drop measurably? Direct-coupling answers yes (Facebook, bitcoin, WhatsApp). Indirect-coupling answers barely or not at all (Chrome, Gmail, x86 server CPUs, Yahoo Search circa 2002).
If a competing network at one-tenth the size offered identical features, which users could be peeled off without disadvantage to the remaining users? Direct-coupling answers very few. Indirect-coupling answers many; per-user value isn't bound to network size beyond ecosystem-quality thresholds.
These tests classify Saylor's five on the direct side and the sweep's five on the indirect side. The classification predicts the outcome rather than retrofitting it.
The parent piece's argument: marginal cost of distribution is zero, marginal cost of production is zero, marginal value of the next user under network effects is positive, the two curves never cross. The third premise — marginal value of the next user — is what does the locking work. When value flows user-to-user directly, every additional user makes every other user's position better, and the smaller network has nothing to offer that the larger does not have more of. When value flows indirectly — through a developer who ships for whoever has the most users, through a data flywheel that improves the ranking model, through software compatibility that resists migration — the smaller network has something to offer (cleaner ecosystem, higher quality at lower scale, a focused niche, a different OS allegiance) and the lock weakens or fails.
The two regimes look identical at the moment of dominance. They diverge over time-windows of five-to-twenty years, with direct-coupling networks staying locked and indirect-coupling networks getting displaced same-substrate.
The split is sharp at the extremes and fuzzy in the middle.
Two-sided markets like Amazon and eBay have direct-coupling on each side asymmetrically: more sellers attract more buyers, more buyers attract more sellers, and each user benefits from the size of the other side. The mechanism is direct in form even if the value flow is mediated. Empirically these networks lock at roughly the same strength as pure Metcalfe networks.
Apple's iOS combines a hardware-installed-base direct effect (more iPhones attract more developers, which attract more iPhones) with a software-compatibility indirect effect. The hybrid locks. Saylor's list places Apple in mobile devices specifically; the developer-ecosystem layer is a coupling mechanism, not the locking mechanism.
Microsoft's Office franchise is the strongest counterexample to the binary split. Office held dominance for thirty years on what looks like pure software-compatibility lock-in (your .docx round-trips with mine; my macros run on yours). The compatibility coupling is direct in a sense: my ability to share files with you depends on us using the same software. Office's lock has been weakening as cloud-collaborative formats become primary, but the dominance has lasted longer than any other indirect-effect example. The Office case suggests the binary split should probably be a spectrum.
A finer-grained taxonomy is possible: pure Metcalfe at one end, pure software-compatibility at the other, with two-sided markets, data flywheels, and hardware-developer ecosystems between. The binary split is the load-bearing first cut.
The parent piece's bets and forecloses survive once the domain is filtered. It still licenses bets on dominant networks within stable substrates where direct user-to-user network value is the dominance mechanism. It forecloses challenges to those networks except via substrate redefinition above the layer.
It does not license bets on dominant browsers, dominant operating systems, dominant search engines (above the data-flywheel mechanism's strength), dominant CPU architectures, or dominant video-calling apps. These look locked and are not, on time-windows that matter for capital allocation. Saylor's list is well-chosen because it lives on the direct-coupling side of the line. The framework's portability across substrates depends on staying inside that filter.
This sharpens the parent's "substrate-definition problem" qualifier. The substrate-definition problem isn't only about disagreeing on whether bitcoin or ethereum or stablecoins are the relevant monetary substrate. It includes a prior question: does the proposed substrate even have direct user-to-user network value, or is its dominance the indirect-effect kind that gets displaced same-substrate within a decade?
The frontier case is AI assistants. Does your assistant make mine more valuable directly, or only through a shared training-data flywheel and developer-ecosystem? If the latter, current AI-assistant dominance is fragile on a five-to-fifteen-year horizon and the lock-claim does not protect it. If interoperability of agent-to-agent calls becomes mandatory or universal, all current dominance positions in the AI-assistant substrate are indirect-coupling. If platforms succeed in locking agents to platforms, direct-coupling re-emerges in a new substrate and the lock applies to whoever wins that race. The question is upstream of any prediction about which AI lab wins.