Chapter 21. The Result
The audit table, completed: Chapter 17 fills one page, every cell carries a chapter’s score, and one column records no failures. That table contains the whole of the book’s argument. This chapter states what follows from it.
Web 3.0, defined: on this web, read is transparent all the way down (S1–S4 at every layer, R1–R3 at the substrate), for humans and machines alike. An agent is only another reader, as Chapter 23 will show. Every earlier use of the term outside this book named no particular property, and this one names the properties the audit table scores.
In Web 1.0, read was transparent over documents, and those documents were declarative, addressable, and indexable, the properties that beat every contemporary in Chapter 1’s history. In Web 2.0, write was added, and the fused term came with it: selection, arrangement, and presentation collapsed into one component, the application stayed on the web only at its rendered surface, and read no longer reached the state behind it. Web 3.0, on this definition, adds no third invention. It extends the transparency Web 1.0 gave documents to the state that lies behind the fused term. Chapter 10’s lateral churn was two decades spent inside Web 2.0. The vertical direction, building new layers on top, was open the whole time, and Part II proved that under the three requirements the next layer had exactly one shape: the state model (5.3). The table now shows what the preface could only claim: the JSON APIs, the JavaScript frameworks, and the compile-to-browser toolchains each have a column, and each one reads as a partial rediscovery of the derived architecture or a detour from it.
One cost remains: the advanced web forfeits the lowest common denominator. Part II derived what advancing requires, and Part IV computed what refusing it costs; Chapter 17 tabulates both. The choice is the reader’s.
Beneath the particular scores is the claim about method the preface promised to return to. The web has been treated as software engineering (frameworks, taste, iteration) and Chapter 10’s lateral churn is what that treatment costs: motion with no fixed point to measure it against. An object with forced structure should be treated as a science instead: derived, proved, and pinned to claims that an observation could refute. The stack is the finding, and the genre of science is the method that found it.
The audit applies to this book as strictly as to the paradigms it scores, so the register below lists the book’s own open claims, each paired with the observation that would refute it:
| claim | where | falsified by |
|---|---|---|
| the SPA paradigm caps at Web 2.0 | Ch 11 | a fused-architecture deployment whose state is machine-consumable at web scale without a compensating adapter layer |
| the convergence recovers neither R3 nor S4 without new incentives | Ch 14 | a mainstream framework shipping addressable intermediates and global references as defaults |
| the agent economy converges on generic systems with domains as data | Ch 23 | agent infrastructure stabilizing permanently on per-application protocol servers, adapter counts growing linearly |
| attribution pressure keeps selecting the fourth position | Prop. 9.2 | a successor standard that discards named graphs |
| a requirement-failure and a compensating industry always coincide | Ch 13 | a paradigm that fails a derived requirement with no compensating market at its web boundary, or such a market around a paradigm that fails none |
Retrodictions (claims history had already graded, like quads and the JS convergence) are marked as such where they occur. This table lists only what is still open. Registered July 2026.
The canonical edition of this book (under construction) is a dataspace on Chapter 19’s machine: its propositions are resources, its dependencies are typed links, and its figures are live queries. Reading it there is itself a demonstration of the architecture it argues for.
The book rests on one claim. Strip any page and the same skeleton appears. Three requirements determine what the state behind the pages must be, and the web already meets all three. That model was standardized by 2014. The next web needs no inventing; it needs only to be put to use.