The Artifact Problem: Why the Act-Level Criterion Cannot Be Settled by Reading This Corpus's Own Objections Sections, and What the Record of This Corpus's Actual Production Process Shows Instead

The Act-Level Criterion distinguishes P-falsifiability, a property of propositions, from A-falsifiability, a property of communicative acts: an act is A-falsifiable only if the communicator genuinely responds to critical engagement by updating, qualifying, or withdrawing, rather than absorbing every challenge as consistent with the original position. That paper built the distinction to detect sophisticated propaganda — content that is individually accurate while the act deploying it is oriented toward belief-production regardless of evidence. Nothing in the criterion's definition restricts it to propaganda directed at an external audience. This paper applies it reflexively, to this corpus's own practice of writing Objections and Falsification-Conditions sections, and reports a finding the criterion was not built to anticipate. Read as finished artifacts, every Objections section this corpus has produced shows the identical pattern: a real objection is stated, and the section ends with the objection answered rather than the thesis revised. This is, by the Act-Level Criterion's own Question 2, indistinguishable on the page from absorption — the behavioral signature of A-unfalsifiability — regardless of whether the underlying epistemic process that produced the page was honest. The paper names this gap the Artifact Problem: a finished text, by the genre conventions of an Objections section, can only ever report the objection that was answered, never the ones that succeeded in changing something before the text was finalized, which means the artifact is structurally incapable of exhibiting the very behavioral evidence the Act-Level Criterion's own test requires. The paper then does what reading alone cannot do: it audits the actual production process behind three papers in this corpus, documented in a record independent of the papers themselves, and finds three instances in which a stated claim was revised, corrected, or narrowed in direct response to external challenge — a fabricated citation identified and removed, a tone-level argument restructured after a specific complaint, and a structural asymmetry claim about A5-self and A5-other conceded and withdrawn inside the same exchange that raised it. None of these three instances appears, or could appear, inside any single paper's own Objections section, because each occurred during the process that produced a paper's final form rather than after it. The paper further isolates a distinct, upstream-failure point the Act-Level Criterion does not itself test: even a genuinely honest response to a stated objection does not establish that the objection selected was the strongest one available, a selection problem Mercier and Sperber's argumentative theory of reasoning predicts will systematically favor objections the author already has a ready answer for. The paper closes by applying Lakatos's own distinction between novel and merely accommodated facts to its own central piece of evidence, and by stating plainly what would falsify its own optimistic reading.