every number traces to a script; every dataset to a license and a retrieval date
The audit below covers every cassiterite Sn-isotope dataset cited in the Uluburun dispute. The column that matters most is the failure mode: how data is unavailable is the finding. The authoritative, maintained version lives in the repository (paper/fair_audit.md); this table is a snapshot.
| Source | Measured / claimed | Deposited machine-readable | Failure mode |
|---|---|---|---|
| Powell et al. 2022 (ingots) | 105 ingots | 105 — supplement, GitHub, and Zenodo all carry this one table | none |
| Powell et al. 2022 (ores) | 17 new cassiterite stated in methods | 0 | measured-but-never-deposited; 6 Taurus values recoverable only from a figure |
| Powell et al. 2024 (Kültepe) | 2 new Hisarcık values; 8 of 9 Taurus values plotted | 0 | no retrievable supplementary files |
| Powell et al. 2026 | “a global dataset of 857 cassiterite Sn isotope analyses” | 0 (supplement is artefact-only) | claimed-but-never-deposited (~570 rows beyond the Mathur subset; formally requested 2026-08-01) |
| Mathur et al. 2025 | 272 stated (159 new) | 0 — PDF tables only; 286 physical rows recovered by our parser | PDF-table-only; row-count discrepancy vs stated total |
| Mason 2020 (dissertation) | 75 ore analyses | 0 — PDF tables | near-total overlap with Mathur 2025: double-counting risk, not independent data |
| Berger et al. 2019 | 29 ingot analyses | 0 in SM | tables rendered as images by the journal; recovered via PMC HTML |
| Berger et al. 2022 (Mushiston slag) | 45 ore analyses, same-sample Sn+Pb+Cu+chemistry | all six supplements | none — the model deposit in this literature |
| Berger et al. 2023 / 2023b | 33 ingot + 102 ore analyses | all | download links hostile to scripted access; fine for humans |
| Williams et al. 2025 | 6 data sheets incl. 17 paired Sn+Pb ores | all | none |
| GlobaLID | ~6,400 Pb rows | all (versioned DOI) | none |
Taurus: no tabulated cassiterite Sn dataset exists — nine values total, eight recoverable only by digitizing a published figure. Karnab: the Uzbek half of the Central Asia attribution has no ore Sn-isotope data at all; the attribution is currently unfalsifiable and unconfirmable. The undeposited compilation: ~570 of the ~857 analyses underpinning the dispute's largest claimed evidence base cannot be audited as data by anyone outside the producing group.
The harmonized eleven-source cassiterite reference library (δ124/116Sn, ‰, NIST 3161a scale; derived summary statistics — specimen-level data ships per-source licenses, see downloads).
| Source | Region | n | mean ‰ | sd ‰ |
|---|---|---|---|---|
| Cornwall/Devon (GB) | Europe | 17 | +0.632 | 0.420 |
| Erzgebirge s.s. (DE/CZ) | Europe | 32 | +0.584 | 0.489 |
| Taurus (TR; figure-digitized) | Anatolia | 8 | +0.479 | 0.441 |
| West Serbia | Europe | 17 | +0.415 | 0.217 |
| Portugal (corrected) | Europe | 8 | +0.251 | 0.439 |
| Slavkov Forest (CZ) | Europe | 14 | +0.148 | 0.308 |
| Paprok (AF) | C. Asia | 10 | −0.039 | 0.383 |
| Spain (corrected) | Europe | 9 | −0.191 | 0.228 |
| Mushiston (TJ) | C. Asia | 28 | −0.270 | 0.129 |
| Dara-i-Pech (AF) | C. Asia | 22 | −0.864 | 1.226 |
| Kunar (AF) | C. Asia | 9 | −1.038 | 1.011 |
Everything below is CC BY (attribute this project and the underlying sources). Model outputs are this project's own; the trace database ships as its CC-BY subset (rows extracted from Springer-copyright material are excluded from the download and available only via the scripts in the repository).
| File | What it is |
|---|---|
| source_library_params.csv | the table above |
| batch_membership.csv | posterior source membership per cargo batch, 11-source + outlier model (the paper's heatmap) |
| prior_families_pi.csv | posterior region shares under the five pre-registered prior families |
| prior_families_diagnostics.csv | divergences, r-hat, conflict checks, ψ posterior per family |
| prior_knockout_sweep.csv | region shares with each elicitation weight zeroed in turn |
| stress_sigma_inflation.csv | exclusion strengths under reference-dispersion inflation |
| stress_admixture_bounds.csv | upper bounds on Mushiston mass fraction per bulk batch |
| stress_aggregation_curve.csv | identifiability vs batch-aggregation m, out to m=128 |
| stress_positive_controls.csv | synthetic-cargo recovery (the pipeline says yes when warranted) |
| stress_partition_k6.csv / _k15.csv | memberships under k-means repartitions owing nothing to the published clustering |
| stress_orbit_offset.csv | memberships under a differential inter-laboratory offset (CEZA-orbit sources vs the rest) |
| prior_alpha0_sweep.csv | prior→posterior movement vs prior concentration α₀ — how much the likelihood actually says about apportionment |
| stress_penouta_demo.csv | the granularity demonstration: one never-proposed mine entered as its own component |
| stress_source_fractionation.csv | memberships under per-source fractionation δ_k (heavy prior) |
| trace_reference_public.csv | the trace-element reference database, CC-BY subset (long format with method, grain-vs-bulk basis, and censoring limits) |
The joint isotope–chemistry model of paper II, one CSV per claim: the censored-rebuild verdict and its treatment ladder, the full-field membership with every measured alternative excluded, and the sensitivity batteries the review rounds demanded.
| File | What it is |
|---|---|
| p2_censored_membership.csv | P(Cornwall) per bulk batch, imputed vs censored references, element panels |
| p2_fullfield_membership.csv | twelve-component membership with partial trace channels; every measured alternative excluded |
| p2_pairwise_bayes_factors.csv | Cornwall vs each measured alternative — the roster-invariant exclusion numbers |
| p2_kappa_profile.csv | the mass-balance shift profiled over fixed κ ∈ [0,2] plus both prior anchors |
| p2_element_lattice.csv | all fifteen lead-free element subsets — the whole decision surface |
| p2_district_symmetry.csv | Cornwall's reference cut to one granite district at a time |
| p2_eta_systematics_sweep.csv | independent per-source systematics η swept to 3 log-units — the verdict never inverts |
Trace elements are the one discriminator with demonstrated power in this dataset, and the paper's value-of-information analysis prices them as the field's highest-value acquisition. The database now exists — long format, one row per analysis × element, with the two fields that make cross-source use honest: method and basis (bulk digests and grain-level spot analyses are different measurands and are never silently pooled).
| Source | District | Analyses | Basis | In public download |
|---|---|---|---|---|
| Williams et al. 2025 (CC BY) | Cornwall-Devon 100 · Iberia 17 · Erzgebirge 11 · Brittany 7 · Massif Central 4 · other 3 | 142 | bulk digest | yes |
| Berger et al. 2022 OR6 | Mushiston area (TJ) | 36 | bulk EDXRF | no (Springer ©; via repo scripts) |
| Konopelko et al. 2022 ES | Mushiston (TJ) — cassiterite, stannite, mushistonite separates | 45 | grain SEM-EDS | no (Elsevier ©; via repo scripts) |
Mushiston thus carries both bases for the same deposit — the calibration pair the grain-vs-bulk bridging work (Project 2C) needs. The notable negative: Serbia has no trace-element reference data anywhere — confirmed at the source (the most-cited Serbian ore paper is a field survey with no per-analysis tables) — which is why the paper ranks a Serbian trace dataset as the field's single highest-value new measurement.