The aging clocks were finally graded on whether they move — 51 intervention studies, 16 clocks, and a recommendation to prioritise two of them (Nature Medicine, 21 Aug 2026)

Why it matters (spine). RDKG's longevity shelf is full of biomarkers, and a biomarker's usefulness has two separate properties that get conflated constantly. Prognostic means it predicts who declines. Responsive means it changes when you intervene. Almost all the published validation work — including the paper behind the garden's dunedinpace-aging-clock card, whose result is a prospective dementia prediction — tests the first. This paper tests the second, at scale, and says plainly why that matters: "comparatively little is known about biomarker responsiveness to interventions, despite responsiveness being a prerequisite for surrogate endpoint qualification." Without responsiveness there is no surrogate endpoint, and without a surrogate endpoint every longevity intervention needs a decade-long trial. This is the paper that decides whether the shortcut exists.

What the paper is, and what it is not. It is an Analysis article in Nature Medicine, open access, published 21 August 2026, DOI 10.1038/s41591-026-04562-9, first author Albert Higgins-Chen (Program in Pathology, Yale University School of Medicine). It is a harmonised re-analysis of other people's trials, not a new trial. That is its strength — it removes the "everyone computed a different clock panel" problem the authors name — and its limit: it inherits the design of every study it pools.

What it did. The authors curated TranslAGE, described in the abstract as "a harmonized database of 51 public and private longitudinal interventional studies," and calculated "a consistent set of 16 prominent epigenetic clocks for each study, along with 94 other DNA methylation (DNAm) biomarkers that can help explain the changes observed for each clock." The 16 clocks span generations: "generation 1 biomarkers (Horvath1, Horvath2 and Hannum), generation 2+ biomarkers (PhenoAge, GrimAgeV1 and DunedinPoAm38), updated clocks from earlier generations (PCHorvath1, PCHorvath2, PCHannum, PCPhenoAge, PCGrimAge, GrimAgeV2 and DunedinPACE) and generation X (Explainable) clocks developed during initial testing phases (SystemsAge, OMICmAge and DNAmEMRAge)." As negative and positive controls they "analyzed four control datasets and nine pro-aging event datasets." The sanity check held: "only aging events and interventions showed significant effects, with effect sizes of 0.14 (increase) and −0.06 (decrease) in the one-sample t-test (P < 0.0001)."

The two-rule test, which is the reusable part. The authors propose that an intervention with a bona fide effect on aging biomarkers should satisfy two conditions: "the intervention should modify DNAm biomarkers of a given generation to the same magnitude and direction in a particular study (rule 1), and a second study of the same intervention should modify the same biomarkers (rule 2)." Their reasoning is the boring-redundant-evidence argument the garden already runs on: "it is unlikely that multiple clocks and multiple studies would repeatedly yield the same false positives."

Applying it, from the paper's own text:

• Anti-TNF therapies in arthritis and inflammatory bowel disease "modify nearly all generation 2+ biomarkers (both reliable and otherwise) to very similar magnitude, fulfilling both rules."
• Two different Mediterranean diets in healthy cohorts "decrease similar subsets of generation 2+ DNAm biomarkers, satisfying both rule 1 and rule 2."
• Two SRW supplement studies change the same biomarkers but not in a consistent direction — "PCHorvath1 and PhenoAge decrease, whereas PCHorvath2, PCHannum and DNAmEMRAge increase" — which "generates uncertainty about the effects of such an intervention on aging."
• Senolytics fail both rules. "The five different senolytics studies have multiple biomarkers changing in different direction within a study (GrimAgeV1 and V2 versus others, not satisfying rule 1) and sometimes the same biomarker changing in opposite direction across studies (PCHorvath2, not satisfying rule 2). This example suggests that the interventional effects of senolytics, at least from an epigenetic aging perspective, may be inconsistent." (The paper says "five different senolytics studies". Since TranslAGE is described as a database of interventions "in humans" these are presumably human studies — but the word "human" is my inference, not the paper's word in this sentence. Wherever "human" qualifies this result on this card or in the changelog, read it as that inference.)

The recommendation. The paper's closing paragraph is explicit: "Our findings support the prioritization of generation 2+ clocks, particularly DunedinPACE and PCGrimAge, in clinical trials. They also highlight the utility of system-specific biomarkers and the importance of population health stratification."

A correction to how this result was being summarised. A trade summary of this paper surfaced during the sweep and framed the finding as pharmacology beating lifestyle, naming anti-TNF therapies and metformin as producing strong changes "more so than supplements, diets or lifestyle changes." That summary was read as a lead and is not a source for anything on this card — it is quoted here only in order to be corrected, and it is listed in the provenance paragraph below as read-and-excluded. The paper does not say that. Its own abstract says "pharmacological and lifestyle interventions drive the strongest responses from DNAm biomarkers" — the two together — and the body's worked example of an intervention satisfying both consistency rules includes two Mediterranean diets. The split the paper's evidence actually supports is between consistent and inconsistent interventions, which cuts across the drug/lifestyle line rather than along it. Metformin appears in no passage of the paper opened for this card, so no claim about it is made here in either direction.

Money angle and metacrisis lever. Money angle: this is a specification, published by the field's own methodologists, for which biomarkers a longevity trial should buy. Anyone selling clock assays, running a healthspan trial, or building the measurement layer under one now has a citable basis for choosing DunedinPACE and PCGrimAge over first-generation clocks — and, symmetrically, a citable basis for refusing to pay for a panel that cannot move. Metacrisis lever: decade-long trials are the reason interventions in aging do not get tested; a defensible surrogate endpoint is the difference between a field that can run randomised trials and one that runs on anecdote and supplement marketing.

What would prove it wrong (falsifier). Two limbs.

• The analysis fails if the TranslAGE database cannot be reproduced from its stated sources, if the harmonised clock calculations diverge from the original studies' published values, or if the responsiveness ranking flips when the pooled set is extended — a curated database of 51 studies is a sampling frame, and publication bias in the underlying literature is a defect the authors themselves name.
• The surrogate-endpoint conclusion fails on the barrier the paper states in its own limitations: "the absence of an empirically defined minimal clinically important difference for DNAm clocks remains a key barrier to their use as surrogate endpoints; future work must anchor clock changes to clinically meaningful functional, physiological or disease-specific outcomes." It also warns that generation 2+ clocks are trained on composite risk factors, so "observed responsiveness may reflect the modulation of intermediate health or inflammatory states rather than direct effects on aging biology per se." Both limbs are open. Until a clock change is anchored to an outcome, responsiveness is necessary and not sufficient, and this card should not be cited as though a clock reading were a health result.

A scale note the paper offers in its own defence, included because it cuts against the easy dismissal: "for widely used clocks such as Horvath, Hannum, PhenoAge and GrimAge, where the standard deviation of epigenetic age deviation is typically on the order of ~4–6 years, an effect size of ~0.2 corresponds to an approximate 1-year shift in epigenetic age, indicating that the observed changes are not biologically trivial."

Where this sits against the cabinet, including where it does not reach. dunedinpace-aging-clock is the card this paper most directly promotes, and it is worth recording that that card's own falsifier already named responsiveness as a test — it fails, in its words, if the clock "fails to move with interventions that demonstrably extend healthspan." The garden was therefore not blind to the question; what is new is a systematic, harmonised answer, and the answer favours the clock the garden already carried. Two boundaries. protphenoage-proteomic-aging-clock is a plasma-proteomics clock; TranslAGE covers 16 epigenetic clocks and 94 other DNA-methylation biomarkers, says nothing about proteomic clocks, and is not a grading sheet for it. And senolytics-preclinical-multiorgan-rejuvenation is a mouse preclinical card, already held at signal, already stating that no human RCT shows senolytics extend healthspan — the two-rule test does not test that card's claim, it adds an independent negative from the DNA-methylation side — human, on my inference from the database's description rather than on the paper's word in that sentence — which sits alongside it.

Provenance (what I actually read). The Nature Medicine article page at nature.com/articles/s41591-026-04562-9, fetched from the main thread on 28 August 2026 and read as a saved file. The date is on the face of the page — the visible line "Published: 21 August 2026" — and is corroborated in the page metadata by citation_online_date: 2026/08/21, dc.date: 2026-08-21 and prism.publicationDate: 2026-08-21. Article type "Analysis", marked Open access. Every quotation of the paper above comes from that page: the abstract, the introduction paragraph on responsiveness and surrogate qualification, the database-and-clock-panel paragraph with the control datasets and t-test effect sizes, the two-rule consistency section including the anti-TNF, Mediterranean-diet, SRW-supplement and senolytics examples, the limitations paragraph, and the closing recommendation. One source outside the paper was read and deliberately not used as a source: a trade summary of the paper surfaced during the sweep, whose "more so than supplements, diets or lifestyle changes" framing is quoted in the correction section above solely in order to be corrected against the paper's own words. It contributed no fact to this card. The two cross-referenced cards named above (dunedinpace-aging-clock, senolytics-preclinical-multiorgan-rejuvenation) were opened in the repository and their quoted falsifier and claim language read directly. I read the abstract, introduction, the consistency-of-effects section, the limitations paragraph and the conclusion; I did not read the full Methods, the figures, the supplementary tables, or the PDF, so this card makes no claim about the composition of the 51 studies beyond what the text above states, and no claim about total sample size — a participant count circulated in secondary coverage of this paper and is deliberately not repeated here, because it does not appear in any passage I opened. Author affiliation is taken from the page's citation metadata.