Foreign-Material Incorporation as a Cross-Chronometer Accuracy Limit

This white paper belongs to the methods branch of the Vacuum-Physics research program: rather than deriving a physical law, it supplies a measurement-accuracy discipline that the program's empirical claims rely on, and it therefore supports the convergence between the foundational vacuum theory and its applications, including the rapid-break-up mechanism developed in Jamming Geodynamics, whose deep-time ages depend on exactly the chronometers examined here. The subject is narrow and deliberately so. It is not whether radiocarbon or uranium–lead decay clocks are correct—both are accepted and independently validated—but whether a particular dated sample records the event one wishes to date. That is the accuracy question, and it is set by contamination and event attribution, not by the decay physics.

The argument opens by separating two claims that are routinely conflated in critiques of geochronology. The first is that a decay clock is unreliable; the second is that a particular sample has been dated inaccurately. Only the second is at issue. Decay constants are measured directly in the laboratory by counting activity, with no geological input and no circular reference to a rock's age, and they are environment-independent over the relevant range—natural-reactor and astrophysical constraints bound any drift in the fine-structure constant to below one part in ten million over two billion years. With the clocks taken as given, what remains genuinely uncertain is whether the analysed material belongs to the target event and whether it has stayed a closed system. This is the same problem the companion radiocarbon paper addresses for biogenic carbon, and the contribution of the present work is to show that radiocarbon and zircon U–Pb fail in the same way, to make that shared failure quantitative, and to meet it with one transparent protocol.

A radiometric age is decomposed along two independent axes. Axis A is internal integrity: given a closed system and a known initial state, does the parent–daughter ratio yield the right elapsed time? For uranium–lead this axis is unusually strong, because two independent clocks—uranium-238 to lead-206 and uranium-235 to lead-207—must agree, a concordance constraint that radiocarbon lacks. Axis B is event attribution: does the dated material correspond to the event, and has it remained closed? Essentially all of the accuracy risk lives on Axis B. It is not a precision limit and not a technology limit; it is an interpretive and geochemical one, and no improvement in instrument quality removes it.

The central observation is that the dominant Axis-B failure is identical across the two chronometers in both mechanism and sign: incorporation of foreign, older material biases a measured age toward greater age. In radiocarbon this is the reservoir, hard-water and dead-carbon effect together with old-carbon contamination; in zircon it is inheritance, antecrysts, xenocrysts and detrital recycling. The mechanism is bidirectional, because loss of the daughter product biases ages younger—episodic lead loss in radiation-damaged zircon, and modern-carbon contamination in radiocarbon—so a complete accounting must guard both directions at once. Where the two methods genuinely differ is not in the mechanism but in its detectability. In radiocarbon the foreign carbon is mixed atom-for-atom; it is homogeneous, inseparable and self-concealing within a single date, and can only be corrected by an externally estimated offset that varies with locality. In zircon the foreign material arrives as discrete crystals carrying their own concordant ages; it appears as separable age populations that can be excluded from the data itself. This single distinction is why zircon is the more self-revealing chronometer, and why a single zircon date is robust for the grain analysed yet requires context to be trusted as the age of a rock-forming event: the date is reliable, but the attribution is the weak link.

The two methods are not merely analogous; they share physics. Matrix exchange is characterised by a Fourier-type diffusion number—the ratio of a characteristic diffusion length to the system size—and the same dimensionless group appears inside the logarithm of Dodson's closure-temperature formulation. Dodson observed that a frozen chemical system in contact with a reservoir is formally identical to the closure-temperature problem, which is exactly what unifies the radiocarbon reservoir effect with diffusive isotopic closure and, by extension, with the retention that preserves a zircon's crystallisation age. The two treatments are methodological siblings rather than convenient metaphors.

From this the paper defines a single protocol that runs on both clocks, with only the implementation of each step differing: screen, then classify, then correct where the foreign component can be characterised, then cross-check against an independent constraint, then report honest uncertainty including the transferability of any correction. The decisive design choice is in the classification step for zircon: grains are sorted on age-independent grounds—crystal position, cathodoluminescence texture, common-lead content—and never by selecting the youngest age. That is what keeps the procedure non-circular, and it is the discipline the whole framework turns on.

Three demonstrations on real data follow. The radiocarbon case is the homogeneous one: a reservoir correction tested by leave-one-out cross-validation on published, independently constrained pairs reduces the root-mean-square error from 418 to 141 years on one shell– charcoal series and from 739 to 453 years on a leaf-wax/macrofossil series. The smaller second gain is itself informative, signalling the limited transferability of a locally estimated offset—precisely the caveat the protocol requires reporting, and exactly what an out-of-sample test, rather than a within-sample fit, is able to expose. The canonical zircon case applies the protocol to the published grain-level dataset for the Lava Creek Tuff supereruption, classifying analyses strictly by crystal position. Crystal faces, the last-grown material, yield a weighted mean of 626.5 plus or minus 2.9 thousand years and reproduce the published rim age exactly; cores grew earlier and sit at 668.8 thousand years, older and excluded as pre-eruptive; an intermediate inter-zone population and a separate volcanic unit are likewise separated out. Two independent anchors—the published eruption age near 631 thousand years and a tephra tie to a marine-isotope-stage transition near 630 thousand years—agree at the one-percent level without any appeal to the uranium– lead ages themselves. The general case uses fully open toolbox datasets to show common-lead screening and correction—a lead-204-bearing suite that is two-thirds discordant before correction and concordant near 312 million years after it, alongside a clean suite that needs no correction—and detrital inheritance, where single-grain ages span a wide range and only the youngest coherent cluster bounds deposition as a maximum depositional age.

The paper then leaves demonstration for the documented record, because the mechanism is recurrent rather than exceptional. Snails living in southern-Nevada springs return apparent ages near 27 thousand years on living animals, purely because the groundwater carries dissolved dead carbon from ancient carbonate; the half-century-old benchmark of living river mollusks dated to roughly 2.3 thousand years is the same effect; and the marine reservoir age of four to five centuries, with regional offsets, is its global form. The East Polynesia settlement chronology is the textbook screening result: a meta-analysis of more than fourteen hundred dates found proposed chronologies varying by over a millennium until the corpus was restricted to reliable short-lived material, which moved initial settlement about four centuries younger and resolved a coherent colonization pattern. Bone shows the opposite, younger-biased failure and its ladder of fixes: for the Vindija Neanderthals, legacy collagen, ultrafiltration and single-amino-acid dating gave successively older and more accurate ages, the last exceeding 48 thousand years and proving that even ultrafiltered dates were still contaminated. In zircon the Bishop Tuff is the sharpest case, where core-targeted ion-microprobe dates of roughly 850 to 892 thousand years stand against a high-precision eruption age near 767 thousand years— the very same older grains being contamination if one wants the eruption age and signal if one wants the magma's storage history, with the discriminator always age-independent. Radiation- damaged grains that have lost lead supply the younger-biased zircon failure, the granular analogue of modern-carbon contamination.

Against this backdrop the paper maps the mitigation methods each field already uses and shows them to be instances of one strategy. The radiocarbon ladder runs from acid– base–acid pretreatment through stepped combustion, ultrafiltration of bone collagen, and compound-specific dating of a single bone-specific amino acid, each rung removing a more tightly defined fraction on chemical or molecular grounds and each with a stated ceiling. The zircon ladder runs from mechanical air abrasion to chemical abrasion that anneals and then dissolves the damaged, lead-loss domains, with imaging-guided microsampling supplying the rim-versus-core classifier and isotopic screening removing the common-lead component. Reservoir databases and calibration curves, and Bayesian age models with explicit outlier detection, add a statistical layer. Every one of these is the same two moves: isolate or avoid the foreign component on grounds independent of the age being measured, then validate out-of-sample or by cross-method consilience. The contribution is not a new reagent but a single accounting that makes those moves explicit, comparable across chronometers, and auditable per claim through the directly-confirmed, inferential and assumption tags carried throughout.

One further layer makes that audit explicit. Because the decay clock passes independent severe tests while the attribution often does not, each result is graded on a single axis—what age-independent test would have failed had the attribution been wrong, and was it performed—on a four-point scale, from an anchored classification that reproduces an independent reference age down to an untested assignment made in a competitive setting. This exposes a structural hazard: the toolkit that flags contamination is asymmetric, catching younger-biasing artefacts cleanly through discordance or percent-modern carbon but older-biasing ones—inheritance, ambiguous antecrysts, old wood—only weakly, so an asymmetric filter on a two-sided error distribution leaves an old-side residual, reinforced in first-appearance priority disputes by an incentive gradient pointing the same way. The claim is falsifiable: in anchor-controlled deep-time cases the screened-minus-reference residual should be sign-symmetric if unbiased, yet the same Lava Creek population already yields a competing near-rim age about thirty-two thousand years older, the disagreement residing entirely in which grains are assigned to the eruption. The scale localises such error rather than overturning chronology, since wholesale failure would require thousands of independent tests to break in coordination.

The honesty is structural. Each method has a working range—uranium–lead degrades at the young end, where trace common lead dominates, and at the extreme-old end, where the shorter-lived parent is depleted—and the protocol states it. There is also one irreducible epistemic asymmetry that the paper refuses to elide: radiocarbon is directly validated in its core interval against historical artefacts and tree-ring counts to 12,593 calendar years, whereas zircon uranium–lead operates in an interval that cannot be checked against any written record, so its accuracy rests on consilience—internal concordance, cross-method agreement and astrochronological tuning. That is strong evidence, but it is not the same kind of evidence as direct ground truth, and no mitigation converts an Axis-B attribution problem into an Axis-A guarantee. Where the foreign component cannot be characterised, the correct output is a wider, explicitly stated uncertainty rather than a precise but unsupported number. The framework is held to three rules throughout: validate against independently confirmed reference ages, test corrections out-of-sample, and tag every quantitative claim so that confirmed facts are never silently merged with inference or assumption. Neither clock is in question; the shared bottleneck is foreign older material, and it yields—never by re-litigating the clock—to age-independent classification and out-of-sample validation.