The argument is published two ways: as three self-contained papers aimed at different journals, and as the single monograph they were extracted from.
Each part stands alone and carries its own abstract, references and conclusion. Read in order they build one argument: why the hard problem dissolves, what evidence bears on it, and how the architecture is formalised.
The philosophical argument
The hard problem rests on an assumption that has gone largely unexamined: that phenomenal experience consists of raw qualia, pre-conceptual states existing independently of meaning. This part argues the assumption is false, and that recognising it dissolves the explanatory gap rather than narrowing it.
Read Part IThe empirical case
What the evolutionary, developmental, clinical and comparative evidence says about a three-level architecture, and the falsifiable predictions it generates. Includes the clinical prediction matrix mapping architectural failures onto known disorders.
Read Part IIThe formal treatment
The architecture formalised across hierarchical predictive coding, variational free energy, intrinsically motivated reinforcement learning and cascade control, with proofs that the four are projections of one functional.
Read Part III
The three papers were extracted from a single book, which remains the fullest statement of the argument: eleven sections and nine mathematical appendixes, including the equation reference sheet, the clinical prediction matrix, the theory of intelligence, and the bilevel-optimisation treatment that the papers only gesture at.
It is not a superset in the tidy sense. The split renumbers objections, moves the comparison material into Part II, and condenses throughout, so a claim can sit in a different place in each. Read the book for completeness and the papers for the argument as submitted.
Elsewhere on the site: Interactive playground|Replies and essays|Algorithm 1 source|Poster