The Rigidity Trap: History-dependent loss and maintenance of future evidence and correction routes

Abstract

Version 2 proposed that sustained success raises rigidity, suppresses alarms and ends in silent failure. This version tests the links. Beyond Paper E, conditional results specify rigidity accumulation, positive never-alarm probability, closure of a paid recovery route and affordable maintenance. A finite reference, F-M1, closes an observation route by construction after successful work. An exact planner certifies 308 of 576 cells; optional maintenance adds two, a protected audit adds 48, and an exchange-rate rule identifies the price boundary. Among tested controllers, sufficient lookahead, a route-aware leaf or the exchange-rate rule matches the planner. Version 2's synthetic passes are built into their programs; its financial program lacks inputs and is excluded from the executable supplement. A trained-network test, fixed before its confirmatory run but not publicly preregistered, measures rectified units closed on a new task. Across 30 seeds in four conditions, fifteen further blocks after initial learning changed the closed fraction and learning deficit by under one point on average. The same update count on changing tasks raised the fraction by up to 56 points and, with higher-rate ReLU, left an 8.4-point deficit. The frozen continued-success prediction is rejected; initial learning had already closed some units. Post hoc interventions reduced the change deficit without identifying closure as its mediator. An unexecuted origination-cohort test specifies contractual routes, competing explanations and measurable rejection conditions in corporate loans. Note on Version 3.0. This version replaces Version 2 (May 2026; then subtitled "Success-Induced Brittleness and Silent Mismatch in Adaptive Systems") and is a full rewrite (about 11,100 to 17,200 words). Version 2's motivating sequence (success, rigidity, alarm suppression, undetected mismatch and failure) is kept as a hypothesis whose links must each be shown; Appendix B lists what is kept, narrowed, corrected and withdrawn. A finite reference (F-M1) and a trained-network test (30 seeds in four conditions) are added, and the frozen continued-success prediction is rejected. Version 2's synthetic passes are built into their programs, and its financial program lacks its inputs and is excluded. Files: the manuscript as PDF and a supplement archive (132 files) with the four unchanged Version 2 synthetic programs and their outputs, the finite model and its checks, an independently implemented menu comparison, the fixed neural design with all 120 per-seed results, and provenance and replay records. The Version 2 files remain available in the previous version of this record. Series. Paper F of the Affective Gain Module (AGM) programme. The integrative flagship, Affective Governance in Adaptive Systems, and the other companion papers (A to E) are archived separately. AI use disclosure. Generative AI (GPT-6.0, OpenAI; Claude Opus 5.5, Anthropic) was used substantively in preparing this work, including source comparison, drafting and editing, and, where applicable, mathematical and counterexample checks and the writing and running of supplementary code. The research questions, framework and final claims were directed and reviewed by the author, who takes full responsibility for the content, including the accuracy of all references and reported numbers. Repository metadata were prepared with assistance from Claude (Anthropic).