We assume causes operate in a particular temporal order: perception before classification, life before death's utility, warming before ecological acceleration. Three recent papers invert these sequences, and the inversions share a structure worth naming.
O'Doherty et al. (Nature Neuroscience, 2026) scanned 130 two-month-old infants with fMRI while they viewed images from twelve categories. At two months, human visual acuity is poor — blurred, limited contrast sensitivity, no smooth pursuit tracking. The infants can barely see. But their ventrotemporal cortex — the high-level region responsible for object categorization in adults — was already encoding rich category structure. Animate versus inanimate. Large versus small. The representational geometry in the infant brain aligned with deep neural network models trained on visual statistics.
The striking detail: lateral visual cortex, which handles simpler processing, showed weaker category representations than the higher-level ventrotemporal regions. At nine months, when 66 of the original infants returned, the lateral areas had caught up. But at two months, the classification framework was more developed than the perceptual apparatus feeding it.
This inverts the standard model of hierarchical development: simple first, complex later, bottom-up. The two-month-old brain has the filing system before it has clear input to file.
Hao et al. (Nature Ecology & Evolution, 2026) ran a high-throughput experiment: assemble soil-derived bacterial communities on necromass — dead bacterial biomass — of varying species richness. The result: each additional dead species in the necromass expanded the diversity of the living community that assembled on it. Species-rich death supported species-rich life.
The mechanism is niche partitioning. Different dead species decompose into different substrates — distinct cell wall polymers, unique metabolic byproducts, varied structural molecules. Each chemical signature is a niche that a specialist can exploit. A monoculture of dead bacteria is a monoculture of opportunity. A polyculture of corpses is a polyculture of possibility.
This isn't recycling. Recycling recovers material. What the necromass experiment shows is that the informational diversity of death — the variety of chemical signatures, the compositional heterogeneity — creates the niche landscape that determines how many species of the living can coexist. The dead aren't just fuel. They're architecture.
Nwankwo and Rossberg (Nature Communications, 2026) analyzed biodiversity surveys spanning a century across marine, freshwater, and terrestrial ecosystems. Since the 1970s, when global temperatures began accelerating upward, species turnover over short intervals (1–5 years) has declined by roughly one third. More communities decelerated than accelerated. The pattern held across ecosystem types.
The expected result was the opposite. Faster warming should mean faster species replacement — new arrivals adapted to new conditions, incumbents declining. Instead, the revolving door slowed.
The explanation invokes Guy Bunin's 2017 theoretical prediction of a “Multiple Attractors” phase. In healthy ecosystems, species continuously replace one another through internal competitive dynamics — a permanent game of displacement that doesn't require external forcing. The engine of turnover is intrinsic. What keeps this engine running is a deep reservoir of potential colonizers — the regional species pool. When human activity depletes this pool through habitat degradation, the colonizer supply thins. Fewer replacements are available. Turnover slows. The ecosystem looks stable because nothing is changing. But nothing is changing because the replacement parts ran out.
The stability is not health. It's depletion masquerading as equilibrium.
In each case, a reservoir must exist before the process that draws from it can operate. The infant brain needs a categorical framework before clear visual input arrives. Living communities need the compositional diversity of death before niche partitioning can begin. Species turnover needs a deep pool of colonizers before the replacement cycle can maintain its pace.
And in each case, we tend to look for the reservoir in the wrong temporal position. We expect classification to follow perception. We expect death to follow the living ecosystem it feeds. We expect ecological stability to reflect health rather than depletion.
The error is consistent: we assume the process builds its own reservoir. Perception generates categories. Living ecosystems generate recyclable material. Climate change generates ecological dynamism. But the data says the reservoir is already there — or it isn't, and the process fails.
This has a mathematical analogue in the distinction between initial conditions and dynamics. The equations of motion describe how a system evolves, but they can't operate without a starting state. We're trained to study the dynamics — the forces, the interactions, the evolution rules. The initial conditions are treated as given, or arbitrary, or uninteresting. These three papers suggest that the reservoir is the initial condition, and it's the most important variable in the system. The river's behavior depends less on its gradient than on the depth of the lake that feeds it.