One of the three missions I’ve given myself in life is to make it to 1,000 years old. Chronologically. I’m aware of how that sounds. But it’s less about the number and more about the orientation: if you’re aiming for a millennium, you take the long view on everything. You stop optimizing for next quarter and start optimizing for decades. I’m 31. My body thinks I’m 26. My brain scan says 46. All three numbers are “correct,” and none of them mean what you think. Chronological age is the laziest metric in medicine. It tells you when you were born. It doesn’t tell you how fast you’re deteriorating, which organs are ahead or behind, or whether anything you’re doing is working. For most of human history, it was all we had. That changed in 2013, when Steve Horvath (the godfather of epigenetic aging research) published a paper that quietly broke the field open. I’ve been tracking my health data seriously for four years now. Epigenetic clocks from TruDiagnostics. Blood panels from Function Health. DEXA and VO2 max from DexaFit. A brain age scan from Kernel. Continuous wearable data from Whoop, Oura, and EightSleep. The results don’t agree with each other. None of them use the same underlying metrics. And that disagreement turned out to be the most useful thing I learned. Every cell in your body carries the same DNA, but which genes are active depends on chemical tags attached to specific positions along the genome. These tags, called methyl groups, shift predictably as you age. Horvath figured out that if you measure the right ones, you can estimate how old a tissue sample is without knowing anything about the person it came from. His model looked at 353 of these positions across 51 tissue types and nailed chronological age with startling accuracy. That was the first [[epigenetic clock::Epigenetic clock: a statistical model trained on DNA methylation patterns to estimate biological age. Different clocks are trained on different outcomes, chronological age, mortality risk, or pace of aging.]]. It was also immediately insufficient. Horvath’s clock and the Hannum clock that followed were trained to match your birthday. They’re good at that. But they’re weak at the thing that actually matters: predicting whether you’ll get sick or die. A 2025 study comparing 14 clocks across 174 disease outcomes confirmed what researchers had suspected. First-generation clocks don’t tell you much about your health trajectory. The second generation changed what the clocks were aiming at. GrimAge, published in 2019, wasn’t trained on chronological age at all. It was trained on time-to-death, using methylation patterns that track proteins linked to inflammation, kidney function, and smoking exposure. The question shifted from “how old do your cells look?” to “how fast are things breaking down?” Then came what I think of as the real breakthrough. [[DunedinPACE::DunedinPACE: a pace-of-aging score derived from DNA methylation, trained on longitudinal biomarker data from the Dunedin Study. A score of 1.0 is average; below 1.0 means aging slower than the norm.]], published by Daniel Belsky et al. in 2022, doesn’t estimate your biological age. It estimates your pace of aging. Every prior clock is an odometer. DunedinPACE is a speedometer. What makes it credible is the training data. The Dunedin Study tracked over a thousand people born in New Zealand in the early 1970s, measuring biomarkers across heart, metabolism, kidneys, liver, immune system, lungs, and even dental health at four points over two decades. That gave researchers actual longitudinal aging trajectories, not cross-sectional snapshots. Belsky’s team then trained DunedinPACE to predict those trajectories from a single blood sample. The ground truth isn’t a statistical guess. It’s twenty years of measured change. A score of 1.0 means you’re aging at the average rate. Below 1.0, slower. My most recent reading: 0.76. Roughly nine biological months per calendar year. The reason I trust this number more than any other is its test-retest reliability. If you draw my blood twice in a week, DunedinPACE gives you nearly the same answer both times. Most other clocks don’t. Four quarterly TruDiagnostics tests over the past year. My OMICm age readings: 28.2, 20.4, 23.8, 26.2. Chronological age at the last draw: 31.2. That 20.4 was not a real ten-year reversal. I didn’t discover the fountain of youth between my first and second test. The swing across four readings spans nearly eight years, and the published literature says that’s normal. Technical noise alone, running the same blood sample through the same machine twice, can produce deviations of 3 to 9 years on major clocks. On top of that, most clocks shift with your circadian rhythm. Blood drawn at midnight looks roughly two years younger than blood drawn at noon. My draws weren’t at the same time of day. That explains a lot. OMICm age is a third-generation clock developed at Harvard’s Brigham and Women’s Hospital in collaboration with TruDiagnostics. During training, it pulls from multiple data layers: epigenetic markers, proteins, metabolites, and clinical records. But the test itself only needs a methylation array. Richer signal in, cheaper test out. It was published in Nature Aging in 2026, though TruDiagnostics was selling it before peer review completed. That gap between commercial launch and published validation is common in this field. Worth knowing. My current reading of 26.2 at chronological 31.2 puts me in the 31st percentile among 31-year-old males in TruDiagnostics’ database. But that database is made up of people who pay $499 for an epigenetic test. The denominator isn’t the general population. It’s the health-optimization crowd. Being in the 31st percentile of that group is a different statement than it sounds. The number I actually track is the DunedinPACE trend. Not a position. A rate. Less sensitive to single-draw noise. Consistent across readings. 0.76, and holding. That I can work with. The most granular view comes from SYMPHONYAge, developed at Yale and licensed by TruDiagnostics. Instead of one composite age, it reports a separate biological age for each of 11 organ systems, each trained against health outcomes specific to that system. The lung clock predicts lung cancer. The heart clock predicts coronary disease. The spread across my organs tells me more than any single number. My brain reads youngest at 21.7, nearly a decade below chronological age. My hormone system is the oldest at 30.0, basically where you’d expect for a 31-year-old. Musculoskeletal sits at 27.8, which I attribute to training volume. I exercise every day: running, strength, rucking, averaging 14,000 to 16,000 steps a day. That keeps cardiovascular and metabolic clocks low but accumulates mechanical stress. What I find most useful is watching the trajectory. My kidney clock dropped from 33.2 to 25.2 across four tests. Liver from 34.2 to 26.2. Both were flagged in my first test and both improved. I adjusted supplements, diet timing, and hydration all at once, so I can’t isolate what worked. But the direction is consistent across multiple readings, and that’s the kind of signal I’m looking for. SYMPHONYAge launched commercially in June 2024 based on a preprint, not a peer-reviewed paper. The validation data looks solid, but the algorithm hasn’t been through the full scrutiny process. I use the data. I hold the conclusions loosely. In October 2025, I sat under a Kernel Flow helmet for a brain age scan. The result: 46.1 years. Fifteen years older than my chronological age. Kernel uses functional near-infrared spectroscopy, which measures blood flow and oxygenation patterns in the brain. It’s reading hemodynamic activity, not methylation. Completely different measurement, completely different construct. There’s a well-known bias in brain age models: they run high for younger people and low for older ones. Cole and Franke (the two researchers who defined the field of brain age prediction) documented this in 2017. The raw number matters less than the bias-corrected gap, and Kernel accounts for this in their methodology. My SYMPHONYAge brain clock reads 21.7. Kernel reads 46.1. That’s a 24-year gap on the same brain. One infers from blood methylation patterns. The other measures real-time neural activity. They can diverge wildly in the same person, and I think that divergence is actually the point. If every tool agreed, you wouldn’t need multiple tools. The disagreements are where you learn something. One data point from an exploratory tool. I’m not panicking. I’m filing it. This is the part that nobody in the biological age space talks about honestly enough: different tools don’t just give you different numbers. They use entirely different inputs, different algorithms, and different definitions of what “biological age” even means. TruDiagnostics reads [[DNA methylation::DNA methylation: the addition of methyl groups to specific positions on the genome, a chemical tag that shifts predictably with age and disease exposure. The raw signal that epigenetic clocks read.]]. Function Health runs blood biomarkers through a variant of Morgan Levine’s PhenoAge formula (one of the most cited biological age algorithms in the literature), which uses nine standard lab values like albumin, C-reactive protein, and glucose. DexaFit combines body composition and fitness data through a proprietary AI layer. Whoop builds a composite from wearable sensor data: heart rate variability, sleep patterns, estimated VO2 max. Each platform is measuring a legitimate dimension of health. None of them are measuring the same thing. Function Health told me I’m 15.5 years younger than my chronological age. Whoop says 9.5 years younger. DexaFit says 5 years. TruDiagnostics says 5 years. Kernel says 15 years older. These aren’t rounding errors. They reflect different tools asking different questions about different biological systems. The 2025 research confirms this isn’t just a consumer-product problem. A study comparing proteomic age and epigenetic age in the same individuals found weak agreement between them. A 2025 opinion piece in a Nature-affiliated aging journal asked the uncomfortable question: do we actually need aging clocks? The authors’ argument is that these molecular layers don’t converge on one “biological age.” They’re each capturing something real. They’re just not capturing the same thing. For me, the takeaway isn’t that the tools are useless. It’s that no single tool gives you the picture. I use TruDiagnostics for pace-of-aging trends and organ-level tracking. Function Health for comprehensive blood biomarker coverage. DexaFit for body composition and VO2 max. Whoop, Oura, and EightSleep for daily recovery and sleep patterns. They overlap in places and contradict in others. The contradictions are where I pay the most attention. The evidence for most interventions is thinner than the supplement industry wants you to believe. Only two interventions have large, multi-year, placebo-controlled RCT evidence for slowing epigenetic aging. The first is caloric restriction. The CALERIE-2 trial put 220 non-obese adults on 25% fewer calories for two years and found a modest but real slowing of DunedinPACE. In population data, that effect size translates to roughly 10 to 15 percent lower mortality risk. That’s an extrapolation, not a directly measured outcome, but the direction is consistent. Other clocks like PhenoAge and GrimAge didn’t budge. CR slows the rate. It doesn’t reset the clock. The second is omega-3 supplementation. The DO-HEALTH trial tested omega-3, vitamin D, and exercise in a factorial design across 777 older adults over three years. Omega-3 alone slowed multiple clocks. Vitamin D and exercise alone didn’t. Three to four months of slowed aging over three years. Not dramatic, but it’s real randomized controlled trial data. Exercise has the strongest observational signal of anything on this list. VO2 max, your maximum oxygen uptake during exertion, has the tightest relationship with all-cause mortality of any fitness metric. Moving from low fitness to even moderate fitness cuts mortality risk roughly in half. I train daily for this reason more than any other. But nobody has run a randomized trial measuring epigenetic clocks with exercise as the sole variable. Sleep matters too. A 2025 Mendelian randomization study found insomnia causally accelerates one of the second-generation clocks. Observational data consistently links poor sleep and irregular sleep timing with faster biological aging. But again, no controlled trial has isolated sleep as the variable. The supplement landscape is mostly hype. NMN and NR, two supplements marketed as precursors to a molecule called NAD+ that dominate longevity marketing, have no epigenetic clock data from human trials. A 2025 review in Nature Metabolism concluded the clinical evidence for slowing aging is insufficient. Charles Brenner (the biochemist who actually discovered how NR feeds into NAD metabolism), has consistently argued that the claims outpace the science. Dasatinib plus quercetin, a senolytic combination, produced a troubling result in its first human epigenetic study: the 19 participants showed clock acceleration, not deceleration. Small study, but not the direction anyone predicted. My own protocol isn’t optimized for any single score. One meal a day with occasional two-meal windows. Daily exercise mixing cardio, strength, and rucking. Consistent sleep tracked through Oura and EightSleep. I don’t know which of these is responsible for my DunedinPACE of 0.76, and I probably never will. The quarterly testing tells me whether the aggregate is trending the right way. I started serious blood work four years ago. Before epigenetic clocks, before organ-level data, before any of the tools I’ve described. The most valuable thing I have now isn’t any single number. It’s a baseline. When you’ve been collecting data for years, you stop reacting to individual readings. You stop celebrating the good ones and worrying about the bad ones. You watch the trend. My kidney clock dropping from 33 to 25 over a year matters more to me than whether it reads 25.2 or 24.8 on any given draw. My DunedinPACE holding at 0.76 across multiple draws matters more than whatever my OMICm age does on a Tuesday. The noise in this field is real and well-documented. Technical replicates vary by years. Circadian rhythms shift readings. Platform updates retroactively change your historical scores. TruDiagnostics updated their algorithm in late 2025 and my earlier readings were recalculated. The trend I’d been watching moved. That’s disorienting when you’re paying $499 per test and tracking quarterly. The cost compounds. Quarterly epigenetics alone is $2,000 a year, before blood panels, DEXA scans, or wearables. This level of health surveillance is available to a narrow demographic. I’m aware of that. A 2025 paper in Frontiers in Aging asked the blunt question: are commercial biological age products ahead of the science for individual use? The researchers who build these tools said yes. The clocks work at population scale. For guiding one person’s decisions, the evidence isn’t there yet. I still test quarterly. Not because I think any single reading is definitive. Because four years of data points have given me something no single test can: a sense of direction. I know roughly where my body was when I started paying attention, and I can see, imperfectly and noisily, whether the things I’m doing are moving the needle. Everything in this space should be taken with a grain of salt. The tools are young, the science is evolving, and any number you get on a given day is one data point from one instrument measuring one dimension of an impossibly complex system. But if you’re going to track anything about your health over decades, trend beats snapshot every time. And if someone tells you they’ve reversed their biological age by ten years, ask them which clock, how many draws, and whether the algorithm changed since their last test.
