Dated, in-progress research notes. Not finished essays — the open work behind the research on post-cognition and the validation of language model outputs.
A new interpretability result shows that language models can report on concepts held in their own internal states. It did not refute the framework. It forced a flat premise — no metacognition — to be replaced by a narrow one: no epistemic self-classification. The narrow one is what the argument actually needed.
A spot-validation that failed at 77.8% did not turn up an arithmetic error. It turned up a hidden ontological decision inside the extractor — the same category collapse the framework is built to catch in other people’s outputs, found this time in mine.
The protocol catches up with the study it governs: six claims codified, AR-lex and AR-sem formally defined, quality control turned into a documentable record. And the record’s most prominent entry is the failure of the classifier that produced the headline number — which is the point.
Where the project stands as of July 2026 — the starting point for this log.
The paper’s eight-section architecture is finalized, with Peirce as methodological foundation and the framework situated as a direct continuation of Quattrociocchi et al.'s work on epistemia.
180 API runs across six claims reveal a 74-point gap between lexical and semantic agreement.
31 claims spanning eleven D1 categories demonstrate taxonomic coverage without category collapse.
Fifteen variables (V1–V15) operationalize the taxonomy into a reproducible annotation protocol.
An Austinian check on illocutionary force, added as a mandatory gate before any D1 classification.
A real-world causal claim exposes the asymmetry of causal validation, formalized as a three-outcome procedure for D4.3.
The seven-dimensional taxonomy (D1–D7) takes shape; eleven categories under D1, version 2.2.
The 2018 essay that first raised the ontological question later formalized in the taxonomy.