The Meta-Adaptive Regime: No Structural Defensive Advantage from Closure Adaptation in Collapse, and the Conditions of its Maintenance under Co-Adaptation
This paper takes up the meta-adaptive regime the series has named and set aside at every turn: an adapting defender and an adapting adversary coupled through the observable trajectory they write together. It argues that whether the defender can win the race of adaptation is a question posed at the wrong level, the asymmetry the race hides being one of acquisition. The trace of a moving defense is open toward the adversary and, on the defensive variable, closed toward the defender: the framework posits the first and a stated independence the second, under which the adversary's emissions bear no mark of the value of that variable (Lemma ε.0, an assembly of these hypotheses). Where the defense mediates, the adversary's search of what the specification admits and the intent never licensed advances with no rejection to mark it, priced by emission alone (Lemma ε.1). The paper proves two results, in two directions. First, closure adaptation, adaptation of the decision function over a fixed observation process, manufactures no robust separation from a collapsed transcript (Theorem 5.1): it may deploy a margin already present, never create one by post-processing. That the absence of a uniformly maintained anchor entails the absence of a uniform defensive advantage is conditional, on a mission-completeness module and a diagonal hypothesis. Second, maintenance under co-adaptation, reversal itself being the defendability layer's, is anchor-relative: it is certified, on the instantaneous risk, where the anchor persists and, for a loop-mediated anchor, where the loop that deploys it satisfies the detection and landed-response floors of Theorem 6.1b. On the defender's certification state, membership in a viability kernel suffices for maintenance under a policy witnessing it. A capture basin on that state certifies that the anchored set can be reached, not that it is reached within the horizon the adversary leaves: a mobility gap, named and left open. Co-adaptation is read through the observable trajectory alone. An adversarial write into the defensive law is typed as a rupture and read by segments, where segments can be posed and the rewritten rule reads only the transcript. The asymptotics are a taxonomy, not theorems: the regime is a frontier, mapped with its direction.
Authors
- Franck Rouxel (ORCID: https://orcid.org/0009-0002-4868-1725)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-14
- DOI
- https://doi.org/10.5281/zenodo.24223521
- Primary Topic
- Information and Cyber Security
- Type
- preprint