The shape that does not return

This argues that treating systems as collections of optimized parts misses where behavior actually lives: in the interactions. Optimizing components in isolation narrows the system’s state space and erodes variety, and when variety is consumed the organization loses adaptive capacity. Reversibility is presented as the key criterion: safe optimizations are those that can be undone without catastrophic loss, preserving the capacity to be wrong and to learn. Stafford Beer’s notion of internal slack (relaxation) is invoked to show that resilience comes from restraint, not from a second act of optimization; deliberately irreversible commitments can be justified only when they protect variety elsewhere, while engineering redundancy is itself a fragile, fixed bet.

Two complications sharpen the point. First, hysteresis: some changes leave a permanent mark (the stress-ball versus wet-clay example), so theoretical reversibility is often unavailable in relevant timeframes. Second, reversibility is observer-relative: capital redeployment can be quick for investors yet permanently destructive for communities, and national-scale optimizations impose asymmetric costs across populations. The central test becomes whose choices are preserved by an intervention and over what timeframe. Because optimization inevitably creates losers, the practical response is not to avoid commitment but to design decisions that minimize the worst harms and keep as many options open for those who bear the costs.