Ideas are real things. Somebody has to check them.
How do you check what a machine claims, after it has claimed it? A research programme on the post-cognitive validation of language model outputs, built from inside the content industry — where the outputs are already shipping.
Current figures and open questions: Project status, updated September 3, 2026
Language models produce assertions without classifying what kind of assertion they are producing. They can report on some of their own internal states; nothing indicates that content type is among them. The programme builds post-cognition: an external, auditable procedure that first fixes what kind of claim is on the table — historical-factual, empirical-quantitative, ethical-normative, metaphysical-ontological, and seven more — and only then asks whether it is true. Ontology precedes epistemology. From September 2026 the programme publishes on two lines, an ontological one and an empirical one, which advance at different speeds.
α1 · Writing & submission
The ontological line
It carries the principle that gives the programme its name, the seven-dimensional taxonomy with its coding manual, the asymmetric procedure for causal claims, and the argument that a validation protocol is a configuration of rules, model and researcher rather than an instrument one applies to the other. Its corpus is thirty-one claims annotated across ten of the eleven active categories — illustration of what the taxonomy does, not validation of whether it holds. It is close to finished, and it is answerable to other arguments.
β1 · In data collection
The empirical line
It carries the vertical validation across 180 runs and the multi-model control of 31 August, and two things that are still outstanding: the blind coding, on which two per-claim proportions and the second arm’s semantic agreement rate depend, and the uplift experiment, pre-registered and not yet executed. It is not close to finished, and it is answerable to data.
Three claims that can be shown to be wrong
-
β Disagreement between model runs is lexical, not epistemic. Across 180 runs, lexical agreement was 4.44% and semantic agreement 76.67% — a gap of about 72 points.
Wrong if the two rates converge on a larger and more varied corpus, or if the semantic rate collapses under a stricter annotation protocol. It has already met one: a label-first re-extraction in July 2026 moved the semantic rate by 1.66 points.
-
β Classifying the type of a claim before evaluating it changes what the evaluation finds. On four causal claims, a prompt that fixes the claim type first should flag a defective causal premise more often than one that does not.
Wrong if a naive prompt flags the structural defect at the same rate as the framework prompt. Whether the same holds for human coders — whether classifying first changes what people find, and how far they agree — would need a control arm of annotators who skip the step. That study is not designed yet and sits outside the scope of this phase.
-
α A model can report on its own internal states and still not classify the epistemic type of its own claims. That is why the check has to come from outside.
Wrong if a model, with no external scaffolding, assigns claim types to its own outputs at the agreement level of trained human coders.
Lab · Independent research
The changelog of the programme. Dated, in-progress notes — including the ones where the framework had to be corrected.
All notes →
September 3, 2026
Post-cognition
Writing & submission
Running the protocol on a second model broke the extractor, and the reason it broke turned out to be an argument about the protocol itself — an argument with no place in an empirical paper. From here the programme runs on two lines.
Read the note →
The Abstract
Reading the Content Economy from Europe — one essay a month
23 July 2026
Europe’s extension on high-risk AI systems makes validation no less urgent. The Fable 5 case showed, in forty-eight hours, what happens when an organization doesn’t know
18 June 2026
Why the right learned the postmodern lesson better than its teachers — and what comes after
22 May 2026
Why the term explains almost nothing, why that matters politically, and why Karp is right about the premise and wrong about everything else
Subscribe on Substack →
Claudio Cammarano leads marketing at De Agostini Libri and teaches at three Italian universities. Twenty years in publishing — Mondadori, Rizzoli, RCS, GEDI — and an academic background in semiotics and philosophy of language. He wrote Il mercato del libro (Solferino).
The point of the day job is not the credential. It is that the outputs this programme is about are already being shipped, at scale, in the industry he works in.
If you work on validation, epistemics or the economics of content, and any of this is near what you do, write: claudio@claudiocammarano.com.
Episteme Advisory (page in Italian) is the consulting practice that applies the method. The research is independent of it: conducted in his own name, published in full in the Lab, and accountable to no client.