We used carbonate reference materials: the ca. 3 Ma ASH15 (D + K) speleothem (Nuriel et al., 2021) and the ca. 322 Ma RA138 marine calcite (Guillong et al., 2024). While they were not specifically analysed to obtain ²⁰⁷Pb / ²³⁵U dates, the existing ID-TIMS data for these samples could be reprocessed to provide ²⁰⁷Pb / ²³⁵U isochrons and ²⁰⁸Pb / ²⁰⁴Pb ratios. We also newly analysed a South African speleothem from the Hoogland paleocave (X4). This sample, previously dated by ID-TIMS by Hopley et al. (2019), comes from the same basal flowstone as sample HL1 dated by solution MC-ICPMS by Pickering et al. (2019) and should be directly comparable (Fig. 1). For laser ablation analyses, we additionally used the ca. 43 Ma calcite breccia B6 (Pagel et al., 2018).

Hoogland speleothem X4 was cut to yield a 2 cm wide section perpendicular to layering, which was polished and scanned on a desktop scanner. LA-ICPMS analyses were then conducted to obtain compositional profiles across the polished sample surface (Fig. 3). The layer with the highest U / Pb was isolated from the section with a diamond wire saw to yield a 2 mm thick slice (Fig. 3). The slice was then subdivided into sub-mm pieces with a scalpel. The focus was on internal domains of the slice, avoiding outside surfaces previously exposed to saws and the related contamination. Pieces suitable for analysis were chosen in the mass range of 1.5–6 mg and were, as much as possible, free of visible surface impurities of non-carbonate material. Before washing and dissolution, they were weighed on a microbalance.

Single carbonate pieces were sonicated in acetone and ultrapure water, then transferred to individual 3 mL screw-top PFA vials. After an additional rinse in ultrapure water, they were dissolved in ca. 150 µL 6 M HCl. Following dissolution, 5–9 mg of the EARTHTIME ²⁰⁵Pb–²³³U–²³⁵U (ET535) tracer solution (Condon et al., 2015; McLean et al., 2015) was added to each vial. Samples were equilibrated with the tracer for 2 h on a 100 °C hotplate and subsequently dried to chloride salts. For U–Pb separation, samples were redissolved in 1 M HBr and loaded onto 50 µL columns filled with AG1-X8 ion exchange resin. Pb and U were isolated from the solutions in a two-step, HBr and HCl-based ion exchange procedure. Uranium and Pb fractions were collected in separate 7 mL PFA vials and dried down with trace H₃PO₄.

Samples were then dissolved in a silica-gel emitter (modified from Gerstenberger and Haase, 1997) and loaded on single, outgassed, zone-refined Re filaments for analysis on a Thermo Triton Plus thermal ionisation mass spectrometer (TIMS). Both Pb and UO₂ analyses were performed using static multicollection in Faraday cups connected to 10¹³ Ω resistors (von Quadt et al., 2016) except cases of low Pb intensity where ²⁰⁴Pb was collected in the axial secondary electron multiplier. Electronic Faraday baselines were acquired for 1 h twice daily. ¹⁸O / ¹⁶O in the UO₂ analyses was determined online by monitoring mass 272 (Wotzlaw et al., 2017) when sufficient intensity was available, otherwise a value of ¹⁸O / ¹⁶O = 0.00205 ± 2 (1σ) was applied to correct for interferences.

All isotopic ratios were processed in Tripoli (Bowring et al., 2011) and a spreadsheet following the equations of Schmitz and Schoene (2007) modified to include ²⁰⁸Pb / ²⁰⁴Pb quantification. Instrumental mass fractionation in Pb isotope analyses was corrected with a value of 0.072 ± 0.025 (1σ) %/amu based on a long-term compilation of Faraday Pb analyses performed with a double-Pb spike. For U, instrumental mass fractionation was determined using the known ²³³U / ²³⁵U of the tracer and assuming that the samples had ²³⁸U / ²³⁵U of 137.818 ± 0.045 (2σ; Hiess et al., 2012). Departures in ²³⁸U / ²³⁵U of up to 0.5 ‰ have been documented in some speleothems (Pollard et al., 2025) but such values have no discernible effect on U–Pb sample age at the current precision level. Measured isotope ratios were corrected for laboratory blank where mass and isotopic composition were estimated based on analyses of 12 total procedural blanks run alongside unknowns, yielding a Pb blank mass of 0.74 ± 0.32 (1σ) pg with a composition of ²⁰⁶Pb / ²⁰⁴Pb = 18.28 ± 1.3 %, ²⁰⁷Pb / ²⁰⁴Pb = 15.66 ± 0.8 %, and ²⁰⁸Pb / ²⁰⁴Pb = 38.11 ± 1.2 %, (all 1σ), and U blank of 0.39 ± 0.07 (1σ) pg (assuming isotopic composition identical to sample). Data pertinent to ²⁰⁷Pb / ²³⁵U geochronology are shown in Table 1 while all sample isotopic compositions are reported in the Supplement. Isochron ages were calculated using decay constants of Jaffey et al. (1971) for ²³⁵U and Audi et al. (2003) for ²³¹Pa. Corrections for ²³¹Pa disequilibrium assumed initial [²³¹Pa / ²³⁵U] = 0.

Laser ablation analyses were performed using an ASI RESOlution LR 193 nm excimer ArF laser ablation system coupled to a Thermo Element XR single-collector sector field ICP-MS following a workflow modified from Guillong et al. (2024) collecting ²⁰²Hg, ²⁰⁴Pb, ²⁰⁶Pb, ²⁰⁷Pb, ²⁰⁸Pb, ²³²Th, ²³⁵U, and ²³⁸U. For LA-ICPMS data, we use the inverse ²³⁵U / ²⁰⁷Pb–²⁰⁸Pb / ²⁰⁷Pb isochron construction (Fig. 2) so the primary focus is on accurate analyses of ²⁰⁷Pb, ²⁰⁸Pb, and ²³⁵U. Preliminary analyses showed that the standard 100 µm spot size leads to unacceptably large uncertainties for the youngest samples (e.g. ASH15) so we adopted the largest available spot size of 380 µm to maximise ion yield. We used static spot ablation at 5 Hz and an energy density of ∼ 2 J cm⁻². To improve precision for small ion beams, the counting times were extended to 100 ms for ²⁰⁷Pb, to 50 ms for ²⁰⁸Pb, and to 20 ms for ²³⁵U, while other isotopes were analysed for 11 ms, resulting in a sweep time of 0.243 s.

All isotopic ratios and intensities were corrected following an adaptation of the standard procedure used in TW carbonate geochronology (Roberts et al., 2017; Guillong et al., 2020). Initial processing was done in Iolite 4 (Paton et al., 2011) and included selection of integration intervals, gas blank subtraction, and the correction of the ²³⁵U / ²⁰⁷Pb ratio for laser-induced elemental fractionation using the U–Pb data reduction scheme (Paton et al., 2010) with NIST SRM 614 glass (Jochum et al., 2011) as the primary reference material. Further treatment was implemented in an Excel spreadsheet. First, fractionation-corrected ²³⁵U / ²⁰⁷Pb ratios were corrected for drift using NIST614. Raw ²⁰⁸Pb / ²⁰⁷Pb was calculated from the two intensities (ratio of intensity) and corrected to the known value for NIST614 (²⁰⁸Pb / ²⁰⁷Pb = 2.4125 estimated from values available in GeoReM, Jochum et al., 2005). Corrected ratios for RA138 were then used to calculate a ²⁰⁷Pb / ²³⁵U isochron date; the comparison of this date with the isochron date obtained by ID-TIMS gave a correction factor applied to all unknown ²³⁵U / ²⁰⁷Pb ratios throughout the sequence. Propagated uncertainties include those resulting from counting statistics and from the RA138 correction, but do not include systematic uncertainties which are likely > 1.5 % as determined for TW dates on RA138 (Guillong et al., 2024). Detailed description of the instrumental setup and methodology used, and all results are included in the Supplement.

Our new (Table 1) and reprocessed ID-TIMS results demonstrate the ability of this method to generate usefully precise ²⁰⁷Pb / ²³⁵U dates. For ID-TIMS data, we use the inverse ²³⁵U / ²⁰⁷Pb–²⁰⁴Pb / ²⁰⁷Pb isochron construction (Figs. 3, 4). The Hoogland X4 sample returned ²⁰⁷Pb / ²³⁵U isochron dates of 5.29 ± 0.15 Ma (filtered, n=5/8, details of filtering in Sect. 4.3 below) and 5.35 ± 0.13 Ma (n=8/8). Both dates are consistent with the ID-TIMS ²⁰⁷Pb / ²³⁵U isochron date of 4.9 ± 1.9 Ma of Hopley et al. (2019) obtained on the same sample X4, and are indistinguishable from our uncorrected ²⁰⁶Pb / ²³⁸U isochron date of 5.349 ± 0.085 Ma (Fig. 3c, n=5/8), indicating that initial ²³⁴U / ²³⁸U disequilibrium was likely negligibly small. Our X4 dates are also consistent with the Tera-Wasserburg intercept date of Pickering et al. (2019) for the same speleothem when corrected using the Vermeesch et al. (2025) algorithm (Fig. 1c), but not when the original correction is applied.

Reprocessed RA138 data (Fig. 4) gave ²⁰⁷Pb / ²³⁵U isochron dates of 324.0 ± 2.3 Ma (n=15/21) and 323.0 ± 3.9 Ma (n=21/21), both indistinguishable from the uncorrected Tera-Wasserburg intercept dates of 321.99 ± 0.65 Ma (n=17/21) and 320.8 ± 1.3 Ma (n=21/21) presented by Guillong et al. (2024). This result is also consistent with the uncorrected ²⁰⁶Pb / ²³⁸U isochron dates for the same data of 321.4 ± 1.4 Ma (n=15/21) and 321.4 ± 1.1 Ma (n=21/21). Reprocessed ASH15 data returned ²⁰⁷Pb / ²³⁵U isochron dates of 3.084 ± 0.056 Ma (n=35/36) and 3.086 ± 0.056 Ma (n=36/36), which is older than the uncorrected Tera-Wasserburg intercept date of 2.965 ± 0.011 Ma (n=35/36, Nuriel et al., 2021) and the uncorrected ²⁰⁶Pb / ²³⁸U isochron date of 2.974 ± 0.015 Ma (n=35/36) but consistent with minor amounts of initial disequilibrium determined in Ashalim speleothems by Vaks et al. (2010). In the three presented cases, the uncertainty of ²⁰⁷Pb / ²³⁵U isochron dates is a factor of 1.5–6 larger than that of the respective uncorrected ²⁰⁶Pb / ²³⁸U isochron or Tera-Wasserburg concordia intercept date, but in each case the dispersion of ²⁰⁷Pb / ²³⁵U data is consistent with a single age population, while ²⁰⁶Pb / ²³⁸U data are overdispersed.

At the sample sizes employed here, the key variable governing the size and shape of uncertainty envelopes in the ²⁰⁷Pb / ²³⁵U isochron diagram is the Pb blank correction. This is because in some cases sample ²⁰⁷Pb is low enough that a large fraction of the measured ²⁰⁷Pb comes from the laboratory blank. In most cases, the blank Pb isotopic composition is indistinguishable from the composition of the initial Pb in the sample (there may be some exceptions; see Szymanowski et al., 2025 for a compilation of TIMS lab Pb blank compositions). Consequently, the blank correction will shift data along the isochron and away from the common Pb intercept, expanding data-point uncertainty and turning positive error correlations into strongly negative ones. In cases where the two compositions are different, the correction might add scatter to the data, which highlights the importance of correctly accounting for blank mass and isotopic composition. In the dataset presented here, the mass ratio of sample to blank Pb was generally < 100 (and often < 10) in ASH15 and X4, and it was on the order of 100–1000 for RA138. Young or small samples similar to ASH15 or X4 may be very sensitive to the blank correction parameters. For example, X4 has relatively more correlated errors than ASH15 at similar sample sizes and U / Pb ratios (Fig. 4), which is a consequence of a 3× larger relative uncertainty on the Pb blank mass used in reducing the X4 data. In cases of significant sensitivity, it might be worthwhile to increase the sample size to improve data-point uncertainties; however, this may come at the cost of decreased spread in the isochron diagram. Old/radiogenic samples such as RA138 will be less affected by blank corrections but such considerations are an important part of planning how much material to dissolve for ID-TIMS analyses.

In carbonate U–Pb geochronology, sample characteristics dictate that we rely on the isochron approach, where a suite of cogenetic samples of variable U / Pb are used to calculate one isochron date. The variations in (blank-corrected) U / Pb and Pb isotopic ratios, and the position along the isochron, result from variable mixing between U-bearing domains of the sample containing radiogenic Pb, and initial Pb-bearing domains. By definition, an isochron is a mixing line between domains of the same age and the same, single initial Pb component. It follows that for a sample of homogeneous age, a real isochron can only accommodate mixing with one isotopic endmember of initial Pb. A violation of this requirement may manifest in the form of obvious excess scatter of data about the isochron but it may also bias the date in ways that are invisible to common statistics describing the quality of isochron fits such as MSWD. Thus, it would be advantageous to isochron age determination if extraneous initial Pb components could be identified and excluded from calculation, thereby satisfying the basic principle of the isochron and improving the accuracy of age interpretation.

In Fig. 4 we illustrate a method to use the measured (blank-corrected) ²⁰⁸Pb / ²⁰⁴Pb as a provenance index of initial Pb that expands upon the suggestion of Engel et al. (2019). ²⁰⁴Pb is a stable, non-radiogenic isotope, while the measured ²⁰⁸Pb will represent a mix of initial ²⁰⁸Pb and the generally negligible radiogenic ²⁰⁸Pb^*. In Th-free carbonates (as well as any ²⁰⁸Pb^*-corrected Th-rich samples), both ²⁰⁸Pb and ²⁰⁴Pb should remain invariant after the closure of the system and so a sample's ²⁰⁸Pb / ²⁰⁴Pb ratio should only depart from the initial value in cases of mixing with extraneous Pb sources. An anomalous ²⁰⁸Pb / ²⁰⁴Pb of an analysed aliquot is a reason to suspect that the aliquot violates the initial Pb homogeneity requirement of the isochron approach, which may introduce bias to its ²⁰⁴Pb / ²⁰⁷Pb and ²³⁵U / ²⁰⁷Pb systematics.

To apply any exclusion criteria, we need enough analyses to identify any trends in the data, while the data-point uncertainties need to be sufficiently small to distinguish trends with confidence. The simplest approach would be to eliminate statistical outliers in ²⁰⁸Pb / ²⁰⁴Pb (Engel et al., 2019), but ideally one may strengthen the interpretations by developing an understanding of the geochemistry of the possible mixing endmembers. These may include, for example, detrital material included in the sampled volume whose Pb isotopic composition may be known or estimated, younger alteration, contamination at the sampling site or in processing. Detrital components of different age may have very different Pb isotopic compositions that are to some extent predictable.

We plot ²⁰⁸Pb / ²⁰⁴Pb in our three ID-TIMS datasets against the reciprocal of Pb concentration, which can visualise straight-line mixing trends (Fig. 4). We suggest a possible approach to data filtering that is not exhaustive – it is only meant to illustrate the potential of the method. In the case of Hoogland speleothem X4, we find five indistinguishable main-group analyses and three that trend toward a high-Pb, low-²⁰⁸Pb / ²⁰⁴Pb endmember. The Hoogland cave was developed within the Archean–Proterozoic Malmani Subgroup dolomite. Similar old U-poor dolomites in the Transvaal craton can have ²⁰⁸Pb / ²⁰⁴Pb values as low as 35.0–35.6 (Bau et al., 1999). Low ²⁰⁸Pb / ²⁰⁴Pb values thus likely represent contamination with similar old, non-radiogenic material. In contrast, among the data for RA138 one can identify a low-Pb, high-²⁰⁸Pb / ²⁰⁴Pb endmember that may be more easily explained with a younger (i.e. not Archean) rock contaminant or a modern lab-derived component. The analyses of RA138 were conducted on powders resulting from the drilling of one specific type of marine cement out of a total of four carbonate generations in the rock (Guillong et al., 2024), so it is conceivable that some analyses included a fraction of Pb derived from a different carbonate generation. The ASH15 data come with large uncertainties on ²⁰⁸Pb / ²⁰⁴Pb which are expected from small samples of initial Pb-poor carbonate. Here, we only identify one potential high-Pb, low-²⁰⁸Pb / ²⁰⁴Pb outlier.

Applying our exclusion criteria results in improving the goodness of fit of the isochrons for RA138 and X4, with a negligible effect on ASH15 (Fig. 4). Thus, the suggested approach appears promising for identifying data skewing the ²⁰⁴Pb / ²⁰⁷Pb–²³⁵U / ²⁰⁷Pb systematics that may go otherwise unnoticed, and it provides a geology-based outlier detection tool that is superior to outlier rejection based purely on poor statistical fit, i.e. “cherry picking” for the sake of improving isochron statistics (Vermeesch, 2026). Finally, the variability of ²⁰⁸Pb / ²⁰⁴Pb illustrated here (Fig. 4) has two important consequences for selecting the optimal analytical strategy for ²⁰⁷Pb / ²³⁵U geochronology: (1) variations in ²⁰⁸Pb / ²⁰⁴Pb in Th-poor carbonates imply that ²⁰⁸Pb is not ideal as a normalisation isotope for ²⁰⁷Pb / ²³⁵U isochrons, (2) analytical setups that can produce precise and accurate ²⁰⁴Pb data (ID-TIMS, rarely ICPMS) offer a big advantage over those that do not (most literature solution and laser ablation ICPMS data).

For LA-ICPMS, we use the inverse ²³⁵U / ²⁰⁷Pb–²⁰⁸Pb / ²⁰⁷Pb isochron construction (Fig. 5). In the three investigated samples, data describe clear linear arrays, apart from several anomalous analyses of ASH15 which will be discussed below. We can evaluate the accuracy of the LA-ICPMS data by reference to ²⁰⁷Pb / ²³⁵U ID-TIMS isochrons in Fig. 4. The ²⁰⁷Pb / ²³⁵U data for B6 are slightly overdispersed (MSWD = 2.6) yielding an isochron date of 47.0 ± 3.5 Ma (n=14/15, excluding one low-intensity analysis). B6 does not have an ID-TIMS reference age, but its age is expected to be ca. 43 ± 1 Ma based on LA-ICPMS TW data (Pagel et al., 2018). Results for X4 (5.09 ± 0.26 Ma, MSWD = 1.2, n=30/30) are in excellent agreement with the ID-TIMS ²⁰⁷Pb / ²³⁵U isochron date of 5.29 ± 0.15 Ma. Interestingly, the ²³⁵U / ²⁰⁷Pb ratios returned by LA-ICPMS for X4 (ranging from 45 to 140) are all larger than those obtained for solutions (2–39), which underscores the capacity of laser ablation sampling to avoid visible contamination. Finally, data for ASH15 were slightly overdispersed (MSWD = 3.5, n=50/50) and resulted in an isochron date of 3.04 ± 0.25 Ma, in agreement with the ID-TIMS date (Fig. 4). Here, the range of U / Pb ratios was more restricted than for solution data, suggesting that common Pb in this sample is not hosted in any major “hotspots” (that may be more easily picked up by solution analyses but avoided by laser sampling) but that it is distributed within the lattice. The precision of our LA-ICPMS ²⁰⁷Pb / ²³⁵U isochron dates is 3–6 times lower than the corresponding Tera-Wasserburg concordia intercept dates, which is comparable to the result obtained for ID-TIMS.

The consistency of our LA-ICPMS data with the ID-TIMS isochron dates for X4 and ASH15 suggests that at least in these two cases, using ²⁰⁸Pb as the numerator in the y-axis ratio does not have major adverse effects. One explanation is that there was no heterogeneity in ²⁰⁸Pb / ²⁰⁴Pb at the scale of laser sampling, another possibility is that it was masked by lower analytical precision. Notwithstanding the general consistency of isochron dates, the relatively large scatter of the LA-ICPMS ASH15 data suggests that the ASH15 sample analysed by LA-ICPMS may have some amount of heterogeneity in the initial Pb isotopiccomposition that was undetectable in the ID-TIMS ²⁰⁸Pb / ²⁰⁴Pb data due to large uncertainties in this Pb-poor sample (Fig. 4). In fact, this is supported by two distinct arrays in the Tera–Wasserburg diagram with very different apparent common Pb intercepts (Fig. 6a). Some of the data in the lower-²⁰⁷Pb / ²⁰⁶Pb array are also enriched in Th up to 1 ppb, suggesting a detrital component. This is in contrast to previous LA-ICPMS data on ASH15 (Nuriel et al., 2021) which scattered around a single isochron, and may be specific to the newly mounted material prepared specifically for this study utilising large laser spots. Irrespective of the reason, this behaviour is an opportunity which may be used in a fashion similar to ²⁰⁸Pb / ²⁰⁴Pb filtering: identifying outliers in Tera-Wasserburg space or in Th concentration which may contribute to improving the ²⁰⁷Pb / ²³⁵U isochron date interpretation. In the case of ASH15, excluding 12 low-²⁰⁷Pb / ²⁰⁶Pb data points results in no improvement of the scatter on the ²⁰⁷Pb / ²³⁵U isochron (Fig. 6b). However, this does illustrate the benefit of analysing ²⁰⁶Pb and ²³⁸U as part of the LA-ICPMS routine.

The main challenge encountered in implementing LA-ICPMS carbonate ²⁰⁷Pb / ²³⁵U geochronology in young samples was the low signal intensity for all involved isotopes, which may make the analyses particularly sensitive to outlier (“spike”) rejection. Such spikes in the time-resolved signal are often expressed as increased counts for a single mass spectrometer sweep, and they likely reflect the detection of single larger particles mobilised from the ablation cell, sample, or transfer tubing. The spikes are not representative of the sample composition, but they may skew a low (tens to hundreds cps) intensity signal in a major way, creating biases in the resulting ratios. In this case, we performed spike elimination on the ²⁰⁸Pb and ²⁰⁷Pb signals separately (2 SD outlier rejection) before calculating a ratio from the means of the two signals (ratio of intensities) to increase chances of success compared to doing this using the time-resolved ²⁰⁸Pb / ²⁰⁷Pb signal (mean of ratios). However, ten analyses in the ASH15 dataset were particularly spiky, so the Iolite routine could not remove the spikes correctly, leading to large biases to the ²⁰⁸Pb / ²⁰⁷Pb ratio. This was not a problem for X4, which was measured at similar < 1000 cps intensities, suggesting that this behaviour is sample-specific. In ASH15, this affected a fraction of analyses which could be excluded from consideration, but any implementations of this method should carefully monitor for this effect.