Everything wears, unless it is cared for

Everything ages. A person ages, and so does an agent, a robot, a bridge, a forest and a program. Machines wear at their bearings and cells, trees lose to drought and beetles, animals and whole ecosystems shift as the conditions around them change. Left alone, each of these lives runs down along its own path.

Care changes that path. Treatment changes a person's life. Servicing and repair change a machine's, forest management changes a stand's, maintenance changes a program's, and an agent kept up to date in its memory and skills stays useful far longer than one frozen at release. Care costs money, time and attention, and it is always spent before its return is visible. So the useful question for anything that lasts is what keeping it well is worth, asked before the spending starts.

The same pattern, everywhere

The evidence from very different fields reads the same way.

Infrastructure. The ASCE 2025 Report Card counts 623,218 bridges in the United States, with an average age of about 47 years, and grades them C; 6.8% are in poor condition. A bridge's life is set as much by the inspection and repair it receives as by the steel and concrete it was built from.

Machines. Remaining life can now be forecast from a machine's own signals. A 2025 study in Energy & Environmental Science predicted the lifetime trajectory of lithium-ion batteries with physics-informed learning, 25 times faster than the standard capacity calibration test, at 95.1% average accuracy across temperatures. Knowing how a battery will age is what lets its care be timed.

Software. Programs decay through what they depend on. Sonatype reported in December 2025 that about 95% of vulnerable components downloaded already had a safer version available, and that roughly 40 million Log4j downloads in 2025 still carried Log4Shell, with fixed versions available for almost four years. The repair existed; the care did not reach the program.

Models and agents. A model's world moves after it ships. A 2025 study of deployed learning-success models found that models trained on learning-platform activity data significantly degraded over time as that data drifted, while models on more stable student profiles held with regular retraining. What a deployed model learned has a shelf life, and the shelf life depends on how it is kept.

Forests. The life of a forest is still hard to see ahead. A November 2025 preprint on drought and insect mortality across the western United States concluded that forecasts from remote sensing are not yet accurate enough to support forest management, and that their skill came mostly from where trees had already died. Forest digital twins are being built to close that gap; a Mediterranean forest twin presented at EGU 2025 names the lack of long growth records as one of the limits on how far such twins have come.

Ecosystems. The EU's BioDT project (2022-2025) built ten prototype digital twins of biodiversity, including grassland and forest dynamics, bird monitoring, ecosystem services and crop wild relatives, described in npj Biodiversity in 2026 together with the integration work still ahead: fragmented data, shared meaning across sources, and the operational effort of keeping twins running.

Six domains, one shape. Each life wears along a path, care bends that path, and the value of the bend is rarely known in advance.

A whole life, with and without the care

Living Worlds lets anything live its whole life, with and without the care it is given. One simulated life runs as it would if left alone; the same life runs again with the treatment, the servicing, the thinning, the patch or the update. The difference between them is the worth of the care, read before a single cost is committed.

That comparison is only as good as the simulated life is true to the real one. Living Worlds checks that before it is trusted: every model is registered, scored once on real data it has not seen, and marked Met or Not met in the public register.

Lifetimes are where this already holds. Frozen on UN data up to 2005, Living Worlds forecast men's life expectancy twenty years ahead in six countries within 1.09 years on average, inside the registered bar of 1.2, and its 80% intervals held all six; the result is Met. Women's life expectancy in four countries came within 0.92 years, Met, with all four forecasts running high. Both were scored on 2025 figures nobody here had read.

Places and the systems running through them hold too. Living Worlds predicted the power flowing every fifteen minutes through 40 real low-voltage feeders that had no meter, for a year it had never seen, ahead of calibrated physics on 38 of 40 feeders and of a learned model on 30 of 40; Met.

Agents carry their life with them. An agent that remembers what a person said predicted that person's answers four years later better than a persona, in every panel and on every question, across 1,911 people in panels never used before; Met. Memory kept current is a form of care, and here it is measured.

And machines share space with people. Living Worlds put a robot into its crowd and predicted how close real people came to a real robot in a room where visitors carried loads, nearer the real closest approach than a crowd blind to the robot and people who ignore it in 7 of 8 recordings; Met.

The lifecycles of machines, of nature and of programs are the next lives being run. They have no results in the register yet. They will arrive the way every domain does: built from measurements, checked on data the model has not seen, and published whichever way the verdict falls.

Beside the live data

A simulated life earns a second use once it runs alongside the real one. Fed the same live data, it keeps saying what should happen next, and the real life either follows or begins to part from it. That parting is the signal. It can be a bug in a program, a fault developing in a machine, a gap in care, such as a service skipped or a dependency left stale, or a gap in the model itself. Each of those is worth finding while it is still a divergence, well before it becomes a failure.

This is the discipline Living Worlds already applies to every life it simulates: check it against the one we live in. Run beside live data, the same check runs continuously, and the record of where the two lives agreed and where they parted becomes part of the life itself.

Where Living Worlds fits

In Living Worlds everything in a world has a life of its own: every person, agent, robot, bridge, tree and program ages, wears and remembers on its own clock, and each one's life shapes the others. Much of the field, even at the frontier, still runs loops: episodes that reset, agents that start each task afresh, scenes stepped and thrown away. Living Worlds runs lives.

Living Worlds simulates how something behaves and changes over time, and its lives: one, many or billions. People and how they live, how they answer, form families and change over a lifetime, are in the register today, alongside places and what moves through them: sound, light, crowds and power, with water next. Robots trained among simulated people come next, building on crowds shown closer to real ones than standard crowd models. Machines, forests and programs follow as lives of their own, each living its whole life with and without the care it is given, so the worth of that care is known before it is spent.