Madjedbebe (Malakunanja II)

What happened

1. 1973J. Kamminga and H. Allen test-excavate the shelter, then called Malakunanja II.
2. 1990Roberts, Jones and Smith report TL dates suggesting people arrived 50,000–60,000 years ago, beyond the then-oldest 40,000-year date.
3. 1990Hiscock publishes 'How old are the artefacts in Malakunanja II?', questioning the TL dates.
4. 2015Re-analysis of the 1989 finds supports occupation 50,000–60,000 years ago and argues disturbance claims were overstated.
5. 2017After 2012 and 2015 excavations, Clarkson et al. date the start of the lowest dense artefact band to ~65,000 years (Nature).
6. 2018The excavators answer forum critics, correct a supplementary table and post the reply's data and code openly on OSF.
7. 2018O'Connell et al. argue that an age over 50,000 years for the site is 'unlikely to be valid' (PNAS).
8. 2020Smith, Ward and Moffat find no termite stone lines at Madjedbebe or Nauwalabila.
9. 2021Williams et al. reject those criteria and say the early dates must be rejected; letters follow.
10. 2025Allen and O'Connell argue Neanderthal-DNA timing rules out 65,000-year ancestors; Veth and colleagues dispute the clock assumptions.
11. 2025A study of 2,456 mitogenomes supports settlement of Sahul by ~60,000 years ago, via two routes.
12. 2026A Sulawesi hand stencil with a minimum age of 67,800 years is read as consistent with arrival in Sahul ~65,000 years ago.

Objections 0 of 7 proved wrong

• ?
  Termites, rain-splash, creep and dripline erosion in the sand could have moved artefacts down into older sand, so OSL dates the sand, not the tools; such disturbance can leave little trace.

  Layers or association · James O'Connell, Jim Allen et al. · 2018

  Unresolved — Excavators cite refits, tight single-grain OSL spreads, micromorphology and a trampling test as showing only small-scale mixing; no direct test has settled it.
• ?
  Phase 2 has no infinite (>50 ka) radiocarbon result: charcoal from Phase 2 depth in the 1989 dig (ANUA-9915, Bird et al.) gave 12.8–13.0 ka cal BP, at odds with the OSL ages.

  Dating or samples · James O'Connell, Jim Allen et al. · 2018

  Unresolved — Excavators: Phase 2 hearth charcoal failed pretreatment (16 of 40 samples did) and fine charcoal is mobile; radiocarbon and OSL agree higher up.
• ?
  Artefacts below the dense band (Phase 1, OSL-dated 65–87 ka) are treated as displaced while Phase 2 finds are treated as in place, which is inconsistent.

  Layers or association · James O'Connell, Jim Allen et al. · 2018

  Unresolved — Excavators (2018): Phase 1 finds are sparse and probably displaced from Phase 2, though an earlier occupation pulse is possible; analysis ongoing.
• ?
  The age is 15–20 kyr older than the next-oldest secure Sahul sites, puts axes, seed grinding and ochre use far earlier than elsewhere, and rests on a single site.

  Contradicts the accepted model · James O'Connell, Jim Allen et al. · 2018

  Unresolved — Critics (2025) still place every other Sahul site at 43–54 ka; a 2026 Sulawesi hand stencil at least 67,800 years old is consistent with early arrival, though its own Sahul-arrival estimate draws on Madjedbebe's date.
• ?
  Neanderthal DNA entered all non-African populations after ~50 ka, so Sahul's ancestors must have arrived later; genetics gives no support for a 65 ka presence.

  Contradicts the accepted model · Jim Allen, James O'Connell · 2025

  Unresolved — O'Connell (S12): this would not show the date is wrong, only that the occupants were not significant ancestors of later Sahul people. Veth et al. (2025) also question the clock assumptions.
• ?
  A conservative reading takes the band's modelled end age (53 ± 4 ka), not its start (65 ± 6 ka), as the safe age for Phase 2 artefacts.

  Dating or samples · Joe Dortch, Anna-Sapfo Malaspinas · 2017

  Unresolved — Excavators: no statistical basis to reject the start age (59–71 ka at 95.4%). A 2025 genetic study still judged 53–60 ka likely for the site.
• ?
  The criteria used to rule out termite stone lines at the site are invalid, and its radiocarbon and luminescence data fit termite disruption; the early dates should be rejected.

  Layers or association · Martin Williams, Nigel Spooner et al. · 2021

  Unresolved — Smith et al. (2020), including an excavation co-author, found no termite stone lines at Madjedbebe; the two sides exchanged letters in 2021.

Evidence at the time within ~5 years of the claim

• Unambiguous evidenceYes — Over 10,000 flaked tools plus grinding stones, ground ochre and edge-ground axe pieces; critics treat them as artefacts and dispute only their age.
• Documented contextPartly — Finds plotted in 3D; refits (median vertical gap 0.10 m) and micromorphology show small-scale mixing, but the sand is bioturbated: frequent small moves, rare large ones.
• Independent dating methodsPartly — The Phase 2 age rests on single-grain OSL (older TL ages agree); radiocarbon matches OSL only in the upper ~1.6 m, as Phase 2 lies beyond its range.
• More than one labYes — A blind OSL study at Adelaide matched Wollongong results on four samples. Radiocarbon dating is credited to a single laboratory (Q. Hua); the OSL cross-check alone supports 'yes'.
• Professional recoveryYes — Excavated in 2012 and 2015 by university archaeologists and dating specialists, with the Gundjeihmi Aboriginal Corporation as research partner.
• Peer-reviewed reportYes — Full report in Nature (2017) with a 10-section supplement; the 1989 assemblage had been described in J. Hum. Evol. (2015).
• Open to criticsPartly — Raw data/code for the 2018 reply's own analyses are on OSF; the full 2017 dataset isn't separately confirmed as posted. Access is also bounded by Mirarr Traditional Owner permission.
• Independent teamsNo — All site evidence in 2017–2022 came from one collaboration; critics reinterpreted published data and called the site 'the single data point' for >50 ka.
• Outside stakeNo — No documented outside motive; authors declared no competing financial interests; funded by Australian Research Council grants.

What would settle it for a field partner

1. Search archived Phase 2 material for identified charred Pandanus endocarp or intact hearth charcoal, then AMS-date it (ABOx-SC) at two labs; a 1989 hearth sample there already failed pretreatment at ANSTO.Medium cost
   Would show: A background result (>~50 ka) would show only an age above ~50 ka, not confirm 65 ka; finite ages on securely in-place, pretreated material would point to intrusion.
2. Refit and raw-material study linking the Phase 1 artefacts (below 2.6 m, OSL-dated 65–87 ka) to Phase 2 pieces in the archived 2012–2015 collections.Low cost
   Would show: Many cross-phase refits would show downward artefact movement; none, with distinct technology, would support an earlier occupation pulse.
3. With Traditional Owners' consent, blind single-grain OSL by a lab outside the original team on a fresh Phase 2 section, plus micromorphology targeting termite reworking, protocol agreed with critics.Medium cost
   Would show: Matching ages and no reworking fabrics would strengthen the 65 ka reading; mixed grain-age populations or termite fabrics would support displacement.
4. Excavate and independently date a second northern Australian site with a dense lower artefact band (e.g. a fresh Nauwalabila I investigation).High cost
   Would show: A second site at 60 ka or more would support early arrival; repeated absence of pre-55 ka occupation would weaken it.

What would change this record

Finite, in-place Phase 2 radiocarbon ages (micromorphology-confirmed, ABOx-SC, two labs), or large-scale downward artefact movement, would refute the 65 ka reading; the existing 12.8–13.0 ka date alone does not. A second dated Sahul site at ≥60 ka would favour it.

Notes

• Status. Unresolved. Critics argue the lowest tools sank into older sand, and that Neanderthal-DNA timing rules out ~65 ka arrival; excavators reject the first, Veth et al. (2025) the second; a 2025 mitogenome study supports arrival by ~60 ka.
• Accepted view then. Short chronology: Sahul settled about 47,000–50,000 years ago; no site securely dated older than ~50,000 years (earlier TL claims from this site disputed).
• Dating basis. Bayesian model of single-grain OSL ages: start of the lowest dense artefact band 65.0 ± 5.7 ka (95.4%, incl. systematic error). Calendar years; OSL counts from ~2015, not 1950.
• Correction. Common accounts call the 1990 claim 'TL/OSL'; the 1990 Nature paper used thermoluminescence only. Single-grain OSL confirming those TL ages followed in 1998 (Roberts, Yoshida et al.).
• Correction. The CLAUDE.md starter list's '~65,000-year claim' is, precisely, the modelled start of the lowest dense artefact band (65.0 ± 5.7 ka at 95.4%); the same band continues to ~53 ka.
• Correction. Trampling was tested by the excavation team itself (Marwick et al. 2017). The critics' main mechanism is termite bioturbation, with rain-splash, creep and dripline erosion.
• Correction. Allen and O'Connell's latest critique (2025) is genetic (Neanderthal-admixture timing), not about sediment movement.
• This record covers the 2017 ~65 ka claim (features coded 2017–2022). The 1990 Malakunanja II TL claim (~50–60 ka vs a 40 ka limit) is queued by the lead as a separate candidate record, cross-referenced rather than merged here. independent_methods_count=3 = single-grain OSL, concordant TL/1998 OSL, and refit/micromorphology integrity work; radiocarbon does not reach Phase 2. limit_bp uses the upper bound of the pre-claim short chronology (~47–50 ka). 2025 mitogenome data favour ~60 ka but still rate Madjedbebe itself at 53–60 ka. Site work is by Mirarr Traditional Owner permission (Gundjeihmi Aboriginal Corporation). No Commons photos found.

People and evidence

• Proponents. Chris Clarkson, Zenobia Jacobs, Richard G. Roberts, Ben Marwick, Mike A. Smith, Richard Fullagar
• Critics. James F. O'Connell, Jim Allen, Martin A. J. Williams, Nigel A. Spooner
• Evidence types. osl, tl, radiocarbon, stratigraphy, geology, lithics, artifacts, features, plant remains · 3 independent lines
• Record confidence. medium — Key figures checked in the 2017 supplement and 2018 reply; several critiques read only as abstracts or via news; treating 2017 as the claim year is a judgment call.