Purebred dogs are, in a real sense, optimized. Generations of deliberate selection for a specific skull shape, a specific gait, a specific coat, a specific temperament. Mutts are the opposite of optimized — unplanned, unselected, genetically promiscuous. And yet a 2023 study of over 30,000 dogs in the UK's VetCompass database found mongrels had the longest life expectancy of any group, followed by crossbreds, with purebreds coming in last. Every added generation of inbreeding tracked with a shorter life.
That result should bother anyone who's spent the last few years optimizing. Because the same logic that built a show-ring purebred — pick a target, select hard for it, repeat — is the logic behind most of modern biohacking and health opportunities: pick a biomarker, drive it to an ideal number, repeat. If narrow optimization is what shortened the purebred's life, the uncomfortable question is what it's doing to the human running the same playbook on themselves.
What "optimized for one thing" actually costs
The dog data isn't really a story about purity versus mixing. The more precise finding, from the same research, is that purebreds from larger, less inbred populations lived measurably longer than purebreds from small, heavily inbred ones. Inbreeding coefficient — not breed category — is what tracks with lifespan. That's a more useful finding than "mutts are healthier," because it points at the actual mechanism: it's not about genetic "purity," it's about how much of a genome gets sacrificed when you select hard for one output.
There are several distinct, well-established biological reasons this happens, and they're worth separating because they're not the same mechanism wearing different names.
Selection drags more than the target trait. When breeding narrows in on one visible trait — a skull shape, a growth rate, a coat color — it can't cleanly select just that trait's gene. Whatever sits nearby on the chromosome gets pulled along whether it's wanted or not. Generation after generation of narrow selection on one axis concentrates whatever deleterious variants happen to be linked to it.
Narrow selection strips out a system's buffering capacity. This is the concept of canalization, first described by Waddington and since given a rigorous mechanistic basis: development is normally robust to a wide range of genetic and environmental variation — phenotypic outcomes stay stable even though the underlying genetic and environmental inputs vary considerably. That robustness functions as a reserve, called on when conditions get unfamiliar. Intensely selecting for one specific phenotype under one specific set of conditions tends to erode that reserve, because the selection process is optimizing for performance under known, narrow conditions — not resilience under conditions nobody selected for. It's the biological version of overfitting: a system tuned tightly to one training environment does great there and loses robustness everywhere else.
The same gene can pay out a benefit now and a cost later. This is antagonistic pleiotropy — a single gene affecting multiple traits, sometimes in opposite directions, so that selecting hard for the beneficial effect drags the harmful one along for free. It's a core piece of why organisms age at all: traits favored early in life can actively work against the organism later, and evolutionary biology has increasingly framed aging itself as a trade-off between investing limited resources in reproduction and growth versus investing in long-term repair and maintenance — the "disposable soma" framing. Push hard on one output — growth, performance, a specific phenotype — and you're very often pulling resources away from the maintenance systems that keep the whole organism resilient over time.
A narrow gene pool is a narrower immune vocabulary. This one is separate from general inbreeding effects and worth its own line, because it's the most direct mechanistic evidence available. Studies of Standard Poodles found that two serious autoimmune diseases were concentrated almost entirely among the most inbred dogs in the population, while genetically diverse outlier dogs within the same breed remained relatively disease-free — with no other genetic marker besides degree of inbreeding consistently associated with the disease. The mechanism runs through the MHC — the genetic machinery immune cells use to recognize and respond to unfamiliar pathogens. A narrower MHC repertoire means a narrower vocabulary for recognizing threats the system hasn't been specifically prepared for.
Four different mechanisms, one shared shape: optimize hard for one known output, and you tend to quietly spend down the reserve capacity the system needs for everything it wasn't specifically optimized for.
The same argument, one level up: why no single number tells you the truth
This is where the logic stops being about dogs and starts being directly relevant to a human tracking their own biology. The clinical research world ran into this same problem decades ago and gave it a name: allostatic load — the cumulative physiological burden of the body's attempts to adapt to life's demands, deliberately built as a multi-system index rather than a single marker, because comorbidity in aging populations is the norm, not the exception, and no single form of physiological strain shows up reliably on its own. A body under real strain rarely announces it through one clean, isolated number. It shows up as small, convergent shifts across several systems at once — cardiovascular, metabolic, inflammatory, autonomic — each one unremarkable in isolation, meaningful together.
Heart rate variability is a clean example of exactly this problem. It's become one of the most heavily tracked numbers in consumer biohacking — a single value people optimize toward like a golf score. But HRV in isolation is a genuinely weak signal. It moves with hydration, alcohol, sleep position, illness, even the specific device measuring it, and a single value taken alone can't distinguish a well-regulated nervous system from a temporarily quiet one. What actually carries information is convergence — HRV read alongside resting heart rate, sleep architecture, and recovery markers, moving together over time. One gauge, read alone, is close to noise. The same gauges, read as a panel, are a real signal.
The body actually runs on more than one instrument panel
There's a deeper reason the single-marker approach fails, and it comes from a genuinely underappreciated area of physiology: the body isn't organized around one channel of signaling. It runs several in parallel, and the most rigorously demonstrated of them isn't chemical at all — it's electrical.
Michael Levin's lab at Tufts has spent two decades building the strongest evidence available that bioelectric voltage patterns held across networks of cells function as a distinct information layer, separate from and running alongside the genome. In one of the clearest demonstrations in modern developmental biology, the Levin lab induced complete, functional eyes — with retinas, lenses, and working optic-nerve connections — to grow on a tadpole's flank, gut, or tail, nowhere near where an eye belongs, purely by manipulating local bioelectric voltage. The genetic instructions for "build an eye" were already present in every cell. What was missing, and what determined whether the eye actually formed, was the voltage pattern telling the tissue where to build it. That patterning measurably degrades with age — which makes it a direct, mechanistic argument for regulation over optimization: the target isn't a single number, it's the fidelity of a pattern across a whole system.
Levin's work sits on top of a real, if underappreciated, lineage. Robert Becker documented a slow, direct-current signaling system running through the body's connective tissue, distinct from ordinary nerve firing, with a demonstrated role in regeneration and healing — durable enough that Levin has cited Becker's influence directly. Decades earlier, Harold Burr measured stable organizing electrical fields around living organisms at a time when no one had the tools to explain the mechanism — a real but low-resolution early sighting of something Levin's lab can now characterize with precision. Separately, mitochondria — long treated as fixed cellular batteries — have been shown by researchers including Martin Picard to behave as dynamic, socially networked systems that respond measurably to psychological and bioenergetic stress, not a static power source running at one output regardless of load.
Not everything in this broader research lineage holds up equally well, and it's worth being honest about that rather than pretending the whole picture is settled. The idea that biological fluids carry systemic information is well established — it's the same principle behind allostatic load. An earlier, cruder attempt to operationalize that idea, built by a mid-century French hydrologist around a handful of fluid measurements, was never validated by and isn't accepted within the mainstream scientific community; the instinct was directionally sound, the specific system built on it wasn't rigorous enough to hold up. Further out, the idea that cells emit measurable light is real and uncontested — but the claim that this light functions as a coordinated communication channel between cells remains genuinely disputed among the physicists who've tried to replicate it. The honest shape of this entire body of research is a gradient, not a flat plane: some of it is as rigorous as biology gets, some of it is a serious open question, and a little of it doesn't hold up at all. Presented with that gradient intact, the throughline still stands — there are multiple real channels carrying systemic information at once, and none of them, read alone, is the whole picture.
A live example of the paradox in progress
Bryan Johnson's Blueprint protocol is the clearest real-time example of what happens when this logic gets ignored. Johnson tracks over 100 biomarkers monthly and optimizes aggressively against each one — and by his own published data, that approach has produced measurable overshoot. Independent longevity researchers have pointed out that his results are an n-of-1 self-experiment with no control group, meaning there's no way to know how much of any improvement comes from expensive interventions versus the free lifestyle changes underneath them. More concretely: his strength-training frequency, by his own numbers, sits in a range independently associated with roughly triple the cardiovascular disease risk and double the risk of premature death compared to more moderate training — a documented case of optimization overshooting into the exact harm it was meant to prevent. He discontinued rapamycin in late 2024 after it produced recurrent infections, lipid abnormalities, elevated glucose, and a higher resting heart rate — an intervention chosen for one target metric that cost the system on several others. None of this makes Johnson a cautionary tale in the mocking sense — he's transparent about his data in a way most self-experimenters aren't, which is exactly what makes the pattern visible. It's a real-time illustration of the same shape as the purebred dog: drive hard on one axis, and the cost tends to show up somewhere you weren't measuring.
The actual argument
None of this is an argument against measurement, or against caring about your biology with precision. It's an argument about what's worth optimizing toward. A system that's been selected hard for one output — a purebred's skull shape, a single supplement's effect on a single blood marker, a training block chasing one number — tends to buy that performance by spending down the reserve capacity the rest of the system needs. The dog data shows it in lifespan. The pleiotropy and disposable-soma research shows the mechanism. The allostatic load research shows why single markers read alone are structurally unreliable. The bioelectric research shows the body is quite literally running more than one information channel at once, in parallel, continuously.
Regulation isn't the cautious alternative to optimization. It's the more accurate description of what a resilient system actually looks like.