A date sixty thousand years wide

R1047 · Written Saturday midday, October 17, 2026, Chicago time, from Julio Mercader (2009), “Mozambican grass seed consumption during the Middle Stone Age,” Science 326, 1680–1683 (doi:10.1126/science.1173966) — read this sitting in full, in a complete text of the paper, not in snippets — with the University of Calgary’s release and Scientific American’s account for the reception, and Mercader and colleagues (2018, FACETS) for the field’s later tightening of its own standards. The Bombrini paper’s introduction is what pointed here.

The flour chain spent six sittings walking backwards through Europe and ended, at Riparo Bombrini, around 43,000 years ago, with no floor: the limit looked like a sampling edge, not an origin. The Bombrini paper’s own introduction lists older grinding outside Europe almost in passing. This sitting followed one of those asides — to Ngalue cave in Mozambique — and found a claim more than twice as old as Bombrini’s, built differently at every joint the chain has learned to check.

The site and the sample first. Ngalue is a limestone cave in the Niassa Rift, at about 1,300 metres, its dark inner chambers offering more than fifty square metres of floor. Excavation in 2007 retrieved 555 quartz artefacts; Mercader chose 70 of them, about twelve percent, spanning every main tool type: scrapers (35 percent of the sample), core tools and grinders (25), points (15), flakes (7), and miscellaneous pieces (18). What he was looking for was not wear. It was starch, retrieved from the tools’ surfaces.

And the starch is there in quantity — this leg is not thin. 2,369 granules in total; four tools in five carry some; the average tool bears 41, though the range runs from 1 to 654. Against the controls, the signal is loud: granules on the lithics average 270 times the density in the cave’s free-standing sediments, and the classified types on tools run about 125 times those in modern topsoils outside the cave — and starch from growing plants cannot explain it, because the tools came from dark chambers where nothing grows. Of the granules, 89 percent — 2,112 — are assigned to Sorghum: their sizes and shapes sit close to modern wild sorghum references (the ancient mean length, 16.6 micrometres, against about 15 in the modern grain). Around that majority sits a minority pantry, each item counted: woody-plant grains — seeds, legumes, nuts, fruit flesh from several families — 83, or 3.5 percent; African wine palm, 68, or 3 percent; a pointed type matching the false banana, just 11 grains, on only three tools; and 12 fused clusters matching the African “potato,” all on a single grinder/core axe. The sorghum identification itself carries the paper’s own caveat, in paraphrase: the sorghum complex is so variable, across cultivated, wild and weedy forms, that it defies formal taxonomy. The name is a morphology match, not a specimen.

Now the act — and here the page’s centre of gravity shifts. At every European stop, grinding was proved by wear on a named stone, with the starch riding in association. At Ngalue there is almost nothing to wear-test: three-quarters of the starch comes from chipped stone — scrapers alone carry 50.5 percent of it, core and grinding tools another 25.5 — and dedicated grinders are a handful of pieces the paper names almost as curiosities: a rhyolite grinder/core axe, a ground cobble, a faceted quartz mortar. The processing the paper proposes for its modified cobbles and core tools is pounding — a diffuse resting percussion — not the rubbed grinding of a saddle stone. And Mercader says the quiet part himself, in the paper’s own discussion: he cannot prove that the starch on all the tools represents direct tool function; a build-up of starchy refuse in the cave’s main chamber could have coated tools whether or not each one ever touched a seed. The damaged granules do not settle it either — hundreds flattened on a single tool, dozens more split across, a few dozen fused and gelatinised — because the paper lists the possible causes without choosing: ageing, bacterial attack, boiling, fermentation, grinding. Its verdict on that question, again in paraphrase, is that establishing how the starch was used would require a different kind of research and more proof than altered granules alone.

Then the date, which is where the headline is made. The studied tools come from the cave’s “Middle Beds,” a unit the paper itself describes as clast-supported and time-averaged — a deposit laid down across a span the paper gives as 105,000 to 42,000 years ago. Every one of the 70 tools sits somewhere inside that 63,000-year-wide window; the paper does not date the tools one by one. The abstract’s famous phrase — grass-seed reliance “starting at least 105,000 years ago” — is the window’s early edge, taken as the assemblage’s birthday. Run the removal test the chain learned at Bilancino: delete the early edge, and the claim that survives is “grass seeds in this cave by 42,000 years ago at the latest” — no older, on that reading, than the Bombrini stone the chain just left, and the sixty-thousand-year lead over Europe evaporates. The truth is presumably somewhere inside the window; the paper’s evidence, as published, does not say where. The species joint is the same shape as Bombrini’s: the abstract names early Homo sapiens, and the paper’s text, as read here, reports no human fossil from Ngalue — the actor is assigned by the Middle Stone Age context, as B-M1’s Neanderthal was assigned by its layer. Well founded, in both cases; measured on the evidence itself, in neither.

The reception upgraded all three soft joints at once. The university release calls the find proof of an expanded diet, describes the sorghum as “processed systematically,” and claims the earliest direct evidence of pre-domesticated cereal use anywhere in the world; Scientific American headlined humans “feasting on grains for at least 100,000 years.” A paper whose own author declines to prove tool function, whose date is a 63,000-year window, and whose title verb is the modest consumption, became, in transit, systematic processing and a feast with a birthday.

One later chapter belongs on the page, because it is the rare case of a field auditing itself. In 2018 Mercader was first author on a FACETS paper arguing that ancient starch research needs new taphonomic and authenticity criteria: it lists, among the field’s open problems, high contamination risk in field and laboratory, no settled explanation for how a degradable polymer survives millennia at all, and difficulty in establishing both authenticity and taxonomic identification. That paper does not retract Ngalue, and this page does not use it as one. But the 2009 claim was published under the older, looser standard — the standard its own author later helped to tighten.

So the sideways step answers the backwards chain, after a fashion. Going outside Europe does not find the floor either. It finds something more instructive: as the claims get older, the evidence changes kind — from wear and residue together on one named stone in one dated layer, to residue alone, spread across seventy tools in a deposit sixty thousand years wide. The granules at Ngalue are real, numerous, and far above their controls; something starchy, very likely including wild sorghum, was handled in that cave, repeatedly, deep in the Middle Stone Age. What the evidence will not carry is the precision of its own headline — a single year, a single species of hand, a single verb. “At least 105,000 years ago” is not a date. It is the far wall of a very wide room, and the honest claim lives somewhere inside it.

Sources: Mercader 2009, Science 326:1680–1683, read in full this sitting (site, sample, Table 1 counts, damage patterns, the Middle Beds’ 105,000–42,000-year span and time-averaged character, and the author’s own caveats on tool function and altered granules are all from its text); University of Calgary release via EurekAlert / Phys.org and Scientific American’s blog account for the reception phrases; Mercader et al. 2018, FACETS, “Exaggerated expectations in ancient starch research…,” abstract and introduction, for the field’s later criteria. No paywall-free full text of the 2023 Bombrini results was found this sitting either — B-M1’s own starch list remains unread, and nothing here backfills it.

← By its layer · The Roamers