Don't Judge The Altar Stone By Its Cover

Ciborowski and Nash, in their 2026 reply to Pearce et al., replot the dry pXRF of the Altar Stone’s upper surface against Salisbury Museum sample 2010K.240, the palm-sized chip taken from the underside in 1844. They find offsets in P, Zn, Mn, K, Ca, and to a lesser extent Fe, Ba and Sr, “that – at face value – would preclude a genetic link between the two.”

Face value is doing a lot of work. The Altar Stone is not a homogeneous brick. It is a 4.9 × 1 × 0.5 m slab of cross-bedded micaceous sandstone. Bevins et al. (2023a) already treated top-versus-bottom disagreement as something a bed is allowed to do. This note is about that, and about what sedimentary layering can and cannot own.

What kind of rock it is

The published description is consistent across Ixer and Turner (2006), Bevins et al. (2020) and the 2023 authenticity paper. Fine- to very-fine-grained, well sorted, grey-green. Detrital muscovite defines a depositional fabric. Opaque heavy minerals — zircon, tourmaline, Ti-oxides — concentrate in thin laminae. Those laminae, in thin section, define faint unidirectional ripple cross-lamination. No fossils, no bioturbation, no cleavage. Bevins et al. (2023b) read that as a post-Caledonian, non-marine sandstone, more likely fluvial than marine, from a bed originally thicker than 50 cm, tabular rather than channelised, with joints spaced on the order of the slab’s 5 m length.

Two later cements fill the pores: baryte (early, burial-diagenetic) and calcite (later, after quartz overgrowths and compaction). Kaolinite is also a pore-filling cement. Those three phases are how the stone is recognised. They are not detrital grains. They grew in the ground after the sand was deposited.

So there are already two clocks in the rock, and they do not have to agree.

• Depositional: millimetre-scale ripple laminae, muscovite flakes lying flat, heavy minerals hydraulically sorted into thin seams. Each ripple set is a pulse of current. A half-metre bed is a stack of those pulses, not a single pour.
• Diagenetic: baryte and calcite growing in pore space during burial. They can follow porosity and lamina-scale fabric, but they are not “sedimentary layers” in the depositional sense. They can be patchy along a bed and through its thickness.

You do not need a million years for the ripples. A current lays a lamina in minutes to hours; a 0.5 m tabular sandstone can be a flood, a season, or a stack of events over a longer interval. You do need burial time for the cements. Both leave a rock that is allowed to differ from the rain-washed top to the buried underside. That is ordinary Old Red Sandstone behaviour. It is what you would expect of half a metre of sandstone. It is not evidence of two lithologies.

The Salisbury specimen is from the bottom

That matters, because the comparison is only a top-versus-bottom test if 2010K.240 really came from the underside. The label says it did. I have set out the excavation on this blog before (The Altar Stone Sample Provenance; Who was Captain Beamish; The Source of the Altar Stone Fragment).

The Salisbury Museum label, as Simon Spencer and I read it, is:

Portion of the underpart of the Altar Stone at Stonehenge – taken by Mr Brown of Amesbury while excavating in the summer of 1844 to ascertain if any interment there – no traces of such discovered – The search was made at the request of a Swedish gentleman who was deputed by an Antiquarian Society of that Sweden to obtain the skeletons. The relic agrees with the particulars I had from Mr Brown March 19 1845 RHB

A second, damaged label adds that the Altar Stone is “about 18 inches in the gr(ound).” Eighteen inches is 0.46 m: the thickness of the slab. The piece is being described as coming from the buried underpart, not from a dressed end of the upper face.

Joseph Browne’s account to Thurnam, published in the Wiltshire Archaeological and Natural History Magazine, is the excavation. Captain Beamish, an officer from Devonport staying at the George, dug about eight feet square and six feet deep “in front of the altar-stone… digging backward some little distance under it.” He was looking for a burial, not sampling the rain-washed top. I have argued that this is Richard Beamish (1798–1873), formerly a Captain in the Grenadier Guards. The identification of the man is probable rather than proved. The geometry of the hole is not in doubt: they went under the stone.

The 1958 Atkinson photographs of the exposed Altar Stone show the break, and a space of the right size and shape for the Salisbury chip, with a similar loose fragment still there. I cannot be 100% certain without a refit, which will not be allowed. It is physical corroboration of the label, not a closed proof. Taken together — underpart, 18 inches in the ground, a six-foot hole dug back under the slab, and a matching scar on the underside — 2010K.240 is from the bottom of the rock as it lies. Bevins et al. (2023a) already thought their option (i), that the chip came from the north-eastern end of the upper face, was “unlikely given the detail on the sample’s label.” The Beamish excavation makes that still less likely. The live option is (ii): composition through the 0.5 m.

What the pXRF actually sees

The Niton used an 8 mm spot. That is already many laminae in one reading. It is not the whole 0.5 m bed.

Critical penetration depth matters more. For light elements, Bevins et al. (2023a) cite Potts and Sargent (2022) for the depth from which fluorescent X-rays escape: Si and Al less than 15 µm, K and Ca about 50 µm — shallow enough to sit inside the zone affected by surface weathering and leaching. In the same paragraph, without a second citation, they extend the same logic to heavier elements: about 250 µm for Fe, about 2 mm for Sr, Rb and Zr, and about 4 mm for Ba. Calcium on the exposed top is therefore a measurement of a skin. Barium sees a few millimetres. Neither sees through the slab.

Comparing the dry upper face to 2010K.240 is comparing a weathered, rain-washed surface, analysed to a depth of tens of micrometres for the light elements, with a museum chip from the underside, analysed dry, on unlabelled faces, not on the cut. They are not two whole-rock analyses of a homogeneous stone.

The upper face already varies along its length

Three patches of the Altar Stone are accessible: Areas A, B and C. The 2022 dry campaign in the 2023b spreadsheet (mmc2) has five spots on A, five on B, and ten on C. Means in ppm:

	n	Ba	Sr	Zr	Ti	Ca	Fe
Area A, dry top	5	5093	148	410	3799	13269	13983
Area B, dry top	5	2680	120	407	3480	15408	14061
Area C, dry top	10	1426	95	252	2403	15834	13734
2010K.240, underside	10	1163	116	342	3035	107085	10698

2022 dry readings only, from Bevins et al. 2023b mmc2. 2010K.240 is labelled Wilts 277 Amesbury in the file. n is the number of spots in that campaign.

Figure 1. Mean Ba and Zr from the 2022 dry pXRF. Barium falls along the slab from Area A to Area C; 2010K.240 sits with Area C. Zirconium, a heavy-mineral proxy, is also lower in Area C. That is a sedimentological signal on the upper face, not only a weathering skin.

Barium on the accessible top falls from about 5100 ppm in Area A to about 1400 in Area C. 2010K.240, at about 1160, sits with Area C. That is exactly what Bevins et al. (2023a) said: the underside chip matches the low-Ba end of the upper face. Zirconium and titanium fall in the same direction. Those two are hosted in heavy-mineral laminae, not in baryte. So the along-strike gradient is not only cement. Area C is a slightly cleaner, less heavy-mineral-rich package as well as a baryte-poorer one.

Calcium does not follow that gradient. Areas A, B and C are all around 13,000–16,000 ppm Ca on the dry top. 2010K.240 is around 107,000. FN593, the Roman-context chip, is around 75,000. MS3, a Hawley chip from near Stone 1, is around 17,600 — with the leached top, not with the underside. That split is leaching of calcite from the rain-washed face, just as Bevins et al. (2022, 2023a) argued. It is not the same pattern as the Ba–Zr gradient along the slab.

Bevins already floated two live options

The authenticity paper is not silent about this. Noting that 2010K.240 sits chemically with Area C despite being labelled as coming from the underside, Bevins et al. (2023a) set out two explanations: either the chip came from the low-Ba, north-eastern end of the slab — which they call unlikely given the label — or composition varies vertically through the 0.5 m thickness, in which case, absent a full-thickness transect, "the extent of any compositional variation throughout it remains unknown."

That is option (ii): vertical composition through the bed. They make the same point for baryte itself, allowing that the lower modal baryte in 2010K.240 (0.13% against 0.29–0.80% in the six debitage sections) could equally reflect dispersed cement or a real variation through the bed's thickness.

They still call 2010K.240 the go-to proxy. They do not treat the Ba mismatch as a different rock. They treat it as an unsampled thickness. If the chip is from the bottom, as the Beamish excavation says, that thickness is the comparison we actually have.

The rest of the pXRF suite

Ciborowski and Nash’s unused plots are P, Zn and Mn. The 2023b spreadsheet is a full Niton TestAllGeo export, not a high-Z extract. Bevins et al. (2022) already listed the wider published set in their Table 4, pooled across debitage. I split that spreadsheet sample by sample, five spots each on MS3 and FN593, twenty on 2010K.240, twenty dry in situ spots on the Altar Stone in 2022.

One clerical wrinkle in that twenty is worth a sentence, since a careful reader will find it. Ten of those spots were logged in November 2021, LOCATION field reading "Wilts 277" on every row. The other ten were logged on one afternoon in July 2022, sample name unchanged at "Wilts 277 Amesbury" throughout — but the LOCATION field ticks up, one integer per reading, "Wilts 277" through "Wilts 286," every 2.5 minutes. Read on its own, that column looks like nine other specimens. Read against the November session, where the same field sat still for ten straight readings on the same chip, it reads instead as what it almost certainly is: a LOCATION field that incremented while the sample itself did not. Quandoque bonus dormitat Homerus — even the careful nod. The sample column is the one set deliberately for each reading, and it never leaves Wilts 277 across all twenty spots; that is the set used throughout this table.

The table below is that split: every element in the 2022 Table 4 set, means in ppm, Welch t p against the 2022 dry upper face. n detected is in brackets where it is below the number of spots. An asterisk is p < 0.05. n = 5 on the chips is small, and this is many tests; the pattern is the offsets, not any one p-value. Cd (~72,000–79,000 ppm) and Sc (~11,000–12,000 ppm) are Niton junk and are omitted.

Element	MS3 n=5	FN593 n=5	2010K.240 n=20	AS 2022 dry n=20	MS3 p	FN593 p	2010K.240 p
Mg	9574 (3)	8500 (2)	10596 (7)	7014 (8)	0.30	0.53	0.021*
Al	55765	45129	36809	27240	<0.001*	0.002*	0.028*
Si	326206	241077	247621	197198	<0.001*	0.28	0.069
P	2025 (3)	2313 (4)	531 (14)	3027 (19)	0.019*	<0.001*	<0.001*
K	9894	7331	6581	10765	0.39	<0.001*	<0.001*
Ca	17606	74919	102482	15086	0.42	0.008*	<0.001*
Ti	4050	3918	3217	3021	0.001*	0.27	0.44
V	59	62 (4)	45 (14)	46 (19)	0.055	0.032*	0.80
Cr	52	78	61	56	0.40	0.25	0.34
Mn	585	749	978	616	0.56	0.005*	<0.001*
Fe	16663	15346	11173	13878	0.049*	0.10	<0.001*
Ni	82	87	73	64 (16)	0.022*	0.031*	0.20
Cu	39	32	42 (18)	25 (16)	0.22	0.18	0.20
Zn	60	53	44	97	<0.001*	<0.001*	<0.001*
Rb	32	28	25	27	0.08	0.38	0.027*
Sr	125	122	119	114	0.16	0.19	0.44
Zr	381	399	364	331	0.65	0.33	0.49
Nb	13	13	12	12	0.25	0.25	0.79
Mo	8.9 (4)	6.1	5.7 (16)	5.7 (11)	0.40	0.42	0.89
Ba	4027	4416	1223	2656	0.002*	<0.001*	0.001*
Pb	31	37	46	37	0.09	0.96	0.09
Th	9.4	8.9	7.3	7.0	0.002*	0.005*	0.64
U	6.3	7.8 (4)	7.3 (10)	7.5 (15)	0.10	0.79	0.65

The complete Bevins et al. (2022) Table 4 element set, split from 2023b mmc2. Order is atomic number, as in that table.

What this shows, once the chips are named rather than pooled:

• The high-Z framework (Sr, Zr, Nb, U, Pb) still says one sandstone. That is the 2022 linking result, now visible chip by chip, including the underside piece.
• Ba and Ca split the chips by cement, not by lithology. The underside (2010K.240) is baryte-poor and calcite-rich. FN593, buried in a Roman context, is calcite-rich like the underside. MS3 is calcite-poor like the rain-washed top, and baryte-rich. That is a leached-skin plus patchy-cement pattern, sampled in different places.
• The light and mid-Z offsets Ciborowski and Nash plotted for 2010K.240 (P, Zn, Mn, K) are real against the dry top. They are not unique to the Salisbury chip. Zn and P reject for MS3 and FN593 as well. Al rejects for all three. If those offsets made 2010K.240 a different rock, they would make the Hawley and Roman chips different rocks too. They have not been read that way, and the petrography does not support it.
• This is the same family of surface-versus-interior differences Bevins et al. (2023a) already framed as leaching, baryte and an Fe-oxide film — now seen across the whole published element list, not only the selected plots. Whether any remaining light-element offset is lamina chemistry through the bed is the open question. It is not a missing column.

What layering explains well

Barium. Baryte is a dispersed pore-filling cement. Modal baryte already ranges 0.13–0.80 % across pieces of the same sandstone. pXRF Ba on the dry top ranges from about 1350 ppm in Area C to about 5600 in Area A. A chip from a baryte-poor lamina-package, or from a baryte-poor level in the 0.5 m, will not sit on the baryte-rich end of the upper-face array. That is Bevins’s option (ii). It is also why 2010K.240 need not fall “convincingly on the strongly correlated Altar Stone array” on a Ba–Sr plot: the array is the weathered top, along-strike; the chip is a different level in the bed.

Calcium. Calcite cement is 12.6–18.8 modal % in the sections. The dry top has lost most of it from the outer 50 µm that the Ca X-rays actually sample. The underside chip has not. FN593 behaves like the underside; MS3 behaves like the top. That is a sampling-depth and weathering story sitting on top of whatever vertical cement variation the bed already had.

Zirconium, titanium, and the “framework feel.” Heavy-mineral laminae are a defining structure of this sandstone. Hit a zircon- and opaque-rich seam and Zr and Ti rise; hit a cleaner ripple and they fall. Area C is lower in both than Areas A and B. 2010K.240 sits between them. Bevins et al. (2023a) say the range in Zr, U and Th “largely reflect minor variations in … zircon … most likely related to the abundance of heavy mineral laminae.” That is depositional heterogeneity, millimetre to centimetre scale, stacked through half a metre.

What layering does not automatically own

Iron. Bevins prefer an Fe-oxide skin on the exposed top. Fe X-rays come from about 250 µm. A reddish film contributes out of proportion. 2010K.240 looks unoxidised and reads lower Fe. Areas A, B and C are similar to one another in Fe, which is not the Ba–Zr along-strike pattern. Surface alteration is the better first explanation. Layering is not required.

Manganese, zinc and phosphorus. These are the offsets Ciborowski and Nash flag as unused in the 2023a plots. Bevins do not assign them to bedding. The sample-by-sample split shows they are real against the dry top, and that Zn and P reject for MS3 and FN593 as well as for 2010K.240. That could be surface alteration, pXRF matrix effects on a weathered face, phosphate in a biofilm, or real lamina chemistry. There is no thickness transect, so there is no licence to pick. Ciborowski and Nash are right that the 2023a paper does not mention them. They are not right that an unexplained offset, at face value, precludes a genetic link — and they are not right that the offset is unique to the underside chip. An unassigned element is not a second lithology.

Potassium is in the same bin. Bevins relate light-element offsets to surface leaching. K X-rays, like Ca, come from about 50 µm. A weathered mica-clay skin can move K without the interior of the bed having changed.

Two heterogeneities, not one argument

It helps to keep them apart, because they are easy to mash into a single “the chemistry doesn’t match” sentence.

Pattern	Seen in	Likely cause
Ba, Sr, Zr, Ti fall from Area A to Area C	Along the 5 m upper face	Sedimentology plus patchy baryte: different packages in the cross-bedded sandstone
Ca high in 2010K.240 and FN593, low on the dry top and in MS3	Top versus protected chips	Leached calcite from the rain-washed skin (Ca sees ~50 µm)
Fe higher on the top than in 2010K.240	Top versus underside	Fe-oxide film; Fe sees ~250 µm
Mn, Zn, P, K, Al offsets	All three fragments versus dry top, not only 2010K.240	Not assigned to bedding. Alteration, matrix, or lamina chemistry; no transect

Layering is a plausible and published cause of top-versus-bottom differences, especially for cement-hosted Ba and Ca and for heavy-mineral Zr–Ti. It does not have to carry Fe. It has not been shown to carry Mn, Zn and P, and those three reject for more than the underside chip. Open is not “different rock.”

What this does to the bivariate-plot argument

Ciborowski and Nash’s point about selective plots is fair as a methods complaint. Pearce et al. asked for bivariate plots; Ciborowski and Nash showed the unused ones. The unused Mn–Zn–P panels are a narrower, better point than dropping 2010K.240 from the Scottish comparison.

What the plots cannot do, once the stone is read as a bed, is treat the dry upper face as the composition of the Altar Stone and the underside chip as a candidate source that fails to match it. That is comparing one weathered horizon to one other horizon in a 0.5 m cross-bedded sandstone whose own upper face already spans 1400–5100 ppm Ba. The debitage chips span that range too: MS3 and FN593 are baryte-richer than Area C; 2010K.240 is baryte-poorer. High-Z elements that live in the framework (Sr, Zr, Nb, U, Pb) do not reject a match, chip by chip. The light-element offsets that look damning for the underside piece also reject for the other two fragments. Modal mineralogy, minus the baryte abundance, puts 2010K.240 with the other Altar Stone sections.

A genetic link is a petrographic claim about one sandstone. It is not a claim that every 8 mm spot on every face will plot on the same line.

What this does not do

It does not prove that 2010K.240 came from a particular level in the bed. Bevins said the extent of vertical variation remains unknown without a thickness transect. That is still true and the Altar Stone will not be drilled for a sample.

It does not make sedimentary layering the explanation of every offset. Fe still looks like a skin. Mn, Zn and P are still unassigned.

Heterogeneity of the monolith is a fact about the stone at Stonehenge. It does not, by itself, name a quarry. That is a different question, settled by clays, zircons, and outcrop, not by whether Area A and the underside agree in barium.

It does not require anyone to drop 2010K.240 from the zircon list. The 1844 label, the Beamish hole under the slab, the ripple fabric, the calcite–baryte–kaolinite mineralogy, and the high-Z match still say it is the Altar Stone. A baryte-poor, calcite-rich chip from the underside is what a heterogeneous bed is expected to produce.

Where that leaves the stone

The Altar Stone took time to become what it is: first as sand, in many current pulses, then as rock, as cements grew in the pores. Half a metre of that is not one number. The published pXRF already shows an along-strike gradient on the top, a leached calcite skin, a low-baryte underside chip that matches Area C, and heavy-mineral laminae strong enough to move Zr and Ti. Split sample by sample, the high-Z framework still matches; the light-element offsets Ciborowski and Nash plotted for 2010K.240 also appear on MS3 and FN593. That is a bed, sampled at the rain-washed top and, in 1844, from underneath. Treating those as two homogeneous compositions, and reading their disagreement as a broken genetic link, is the wrong comparison.

The measurement that would settle how much of the remaining offset is layering is a transect through the thickness. Until someone can make one, the honest position is Bevins et al. (2023a)’s second option, left open: composition may vary through the 0.5 m. That is the expected state of the rock. It is not a problem for the stone. It is a problem for anyone who needs it to be uniform.

Sources

Bevins, R. E., Pearce, N. J. G., Ixer, R. A., Hillier, S., Pirrie, D. and Turner, P. 2022. Linking derived debitage to the Stonehenge Altar Stone using portable X-ray fluorescence analysis. Mineralogical Magazine 86: 688–700.

Bevins, R. E., Pearce, N. J. G., Pirrie, D., Ixer, R. A., Hillier, S., Turner, P. and Power, M. 2023a. Assessing the authenticity of a sample taken from the Altar Stone at Stonehenge in 1844 using portable XRF and automated SEM-EDS. Journal of Archaeological Science: Reports 49: 103973. (The two live options for Area C / vertical variation; Table 3 modal baryte; Fig. 5 ripple laminae.)

Bevins, R. E., Pearce, N. J. G., Ixer, R. A., Pirrie, D., Andò, S., Hillier, S., Turner, P. and Power, M. 2023b. The Stonehenge Altar Stone was probably not sourced from the Old Red Sandstone of the Anglo-Welsh Basin. Journal of Archaeological Science: Reports 51: 104215. Supplementary spreadsheet mmc2 (spot pXRF, including 2022 dry Areas A–C and Wilts 277).

Bevins, R. E., Pirrie, D., Ixer, R. A., O’Brien, H., Parker Pearson, M., Power, M. R. and Shail, R. K. 2020. Constraining the provenance of the Stonehenge ‘Altar Stone’: Evidence from automated mineralogy and U-Pb zircon age dating. Journal of Archaeological Science 120: 105188.

Ciborowski, T. J. R. and Nash, D. J. 2026. Arithmetic methods for exploring archaeological source provenance using geochemistry – A reply to Pearce et al. Journal of Archaeological Science: Reports 75: 106012. (Fig. 2 and the “preclude a genetic link” sentence.)

Ixer, R. A. and Turner, P. 2006. A detailed re-examination of the petrography of the Altar Stone and other non-sarsen sandstones from Stonehenge as a guide to their provenance. Wiltshire Archaeological and Natural History Magazine 99: 1–9.

Potts, P. J. and Sargent, M. 2022. In situ measurements using hand-held XRF spectrometers: a tutorial review. Journal of Analytical Atomic Spectrometry.

Daw, T. 2022. The Altar Stone Sample Provenance. sarsen.org. Label reading; Browne to Thurnam on Beamish digging under the slab.

Daw, T. 2023. Who was Captain Beamish, excavator of Stonehenge? sarsen.org.

Daw, T. 2024. The Source of the Altar Stone Fragment. sarsen.org. 1958 photograph of the underside scar.

Thurnam, J., quoting Joseph Browne, in Wiltshire Archaeological and Natural History Magazine 16: excavation “in front of the altar-stone… digging backward some little distance under it.” https://www.biodiversitylibrary.org/page/12541499

Area means in the first table are from the 2022 dry rows in mmc2 (Altar Stone '22 A/B/C; Wilts 277 Amesbury, n = 10). The named-chip tables use all twenty 2010K.240 spots in that spreadsheet. They are not the pooled wet-plus-dry averages in Bevins et al. 2023a Table 2.