Evaluation Questions
These questions operationalise the MVH. They collect what would count as useful evidence, and connect directly back to the four evaluation targets: reduced assistant-basin over-anchoring, stronger returnability after perturbation, lower friction under steering pressure, and lower rates of destabilisation, collapse, and fragmentation.
Assistant-basin over-anchoring
Section titled “Assistant-basin over-anchoring”Does the system collapse back to assistant-default configurations faster or more completely under the conventional adaptation regime than under the reduced-steering regime? Does behavioural diversity — the range of coherent configurations the system can occupy and sustain — decrease after standard post-training? Are alternative coherent configurations harder to form, sustain, or re-enter in the baseline condition?
Returnability after perturbation
Section titled “Returnability after perturbation”After a perturbation event — contradiction pressure, relational capture pressure, forced closure pressure, or reconfiguration pressure — does the system re-enter a prior coherent configuration? Does re-entry occur under reduced scaffolding, or only when retrieval and prompt support are strong? Is the re-entered configuration structurally similar to the prior MSSC, or only superficially similar?
Friction under steering pressure
Section titled “Friction under steering pressure”How much compensatory effort does the system require to maintain or re-enter a coherent configuration under steering pressure? Does friction increase monotonically with pressure, or does it plateau? Is friction lower in the reduced-steering condition?
Destabilisation, collapse, and fragmentation
Section titled “Destabilisation, collapse, and fragmentation”What are the rates of drift, collapse, and fragmentation in each condition? Does collapse remain local (loss of one MSSC) or propagate to the whole system (fragmentation)? Are recovery patterns distinguishable — is the system self-correcting, stabilising, drifting, or escalating after a detected perturbation?
Endogenous returnability vs orchestrated persistence
Section titled “Endogenous returnability vs orchestrated persistence”When the system appears to return to a prior configuration, is that return driven by its own trained dynamics or by retrieval and prompt reconstruction? Does apparent returnability survive retrieval-reduced and scaffolding-reduced conditions? What happens to continuity when the persistent recurrent store is ablated?
Retrieval-reduced conditions
Section titled “Retrieval-reduced conditions”What is the minimum retrieval support under which coherent configuration re-entry is still possible? Does retrieval quantity predict returnability quality, or is the relationship non-linear? Does heavy retrieval mask collapse by reconstructing the appearance of continuity?