# Başur–Tatika ancient-DNA pilot: final report

**Analysis date:** 15–16 July 2026 (Australia/Adelaide)
**Status:** complete

## Executive conclusion

No individual meets the preregistered criteria for unexplained anomalous ancestry, and the pilot does not find evidence requiring an unknown ancestry introduction near Cudi Dağı around 2900 BCE.

The full rotating-source qpAdm panel does distinguish both usable Başur individuals from the sparse three-person strict Tatika-pre panel, so simple local genetic identity is not supported. The difference is nevertheless compatible with multiple sampled Late Chalcolithic/Early Bronze Age regional proxies, and the selected known-regional models remain valid across every full-panel leave-one-right-out test and all eligible processing sensitivities. This is the preregistered pattern of an ordinary regional shift, not an unexplained anomaly. Tatika post is compatible with Tatika pre in the primary qpAdm grid, while its nominal qpWave discontinuity is borderline and changes with sample definition. The negative result remains inconclusive for small or unsampled contributions because local sample sizes and damage-restricted coverage are weak.

This is a test of genetic continuity and regional ancestry modelling around a proposed chronological boundary. It is not a general test of biblical reliability. No genetic result can identify Noah, Cain or any other named person, and the data cannot reveal spoken language.

## Data and prospective design

Four paired-end 1240K capture runs from [ENA project PRJEB83032](https://www.ebi.ac.uk/ena/browser/view/PRJEB83032) were processed: ERR14088882 (SK1096/SB710), ERR14088884 (SK8/SB714), ERR14088885 (SK36/SB726) and ERR14088886 (SK1080/SB731). Published regional genotypes came from AADR v66.p1 ([AADR DOI](https://doi.org/10.7910/DVN/FFIDCW)) and the [Southern Arc study](https://doi.org/10.1126/science.abm4247) and [data release](https://doi.org/10.7910/DVN/3AR0CD). Archaeological context was checked against the [Başur Höyük publication](https://doi.org/10.1017/S0959774324000398).

The dated analysis plan in `preregistered-analysis-plan.md` was frozen before any newly called Başur ancestry result was opened. It defined the local chronology, sample thresholds, comparison populations, primary statistics and stringent seven-part anomaly rule. The unavoidable BWA mismatch-cap change and other implementation details are recorded in `analysis-deviations.md`.

## Independent Tatika chronology verification

The current AADR annotation and Southern Arc records independently reproduce the previous count:

| Chronological class | Published individuals | Primary use |
|---|---:|---|
| Securely pre-2900 BCE | 5 | Three AADR `Pass` individuals with at least 50,000 SNPs form the strict local pre panel; two `Questionable` individuals enter QC sensitivity only |
| Crosses or touches 2900 BCE | 2 | Excluded from the primary pre/post comparison; assigned in both directions in sensitivity analyses |
| Securely post-2900 BCE | 5 | Two `Pass` individuals exceed 50,000 SNPs and also exceed 100,000 SNPs; the other three are too sparse for primary genome-wide tests |

The strict local panel is I4480, I4614 and I4615. I4483 and I4622 are the QC-relaxed secure-pre sensitivity samples. I4481 and I4619 are the usable secure-post samples. I4478 is the high-coverage boundary sample; I4617 is boundary-dated but low coverage. Full date ranges and methods are in `sample-metadata.csv` and `metadata/tatika-verification.md`.

![Başur, Tatika, Cudi Dağı and the frozen regional comparison panel](figures/regional-sample-map.png)

## Sequence processing and authenticity

FASTQ byte sizes, repository MD5 checksums and gzip integrity were verified before processing. Reads were adapter-trimmed and merged, mapped to hs37d5, filtered, coordinate-deduplicated and assessed with mapDamage and a PMD≥3 sensitivity. Seeded pseudo-haploid calls were made at AADR-compatible autosomal 1240K sites. The workflow and resolved environment are in `scripts/`, `environment.yml`, `environment-details.md` and `data-manifest.tsv`. The pipeline follows standard ancient-DNA authentication principles; damage-filtering uses [PMDTools](https://jakobssonlab.iob.uu.se/index.php/software/pmdtools/), and mitochondrial contamination is attempted with [Schmutzi](https://pmc.ncbi.nlm.nih.gov/articles/PMC4601135/).

| Sample | Called 1240K SNPs | Q30 deduplicated mapping fraction | Duplicate fraction | PMD≥3 reads | Terminal 5′ C→T / 3′ G→A | Authenticity | Primary ancestry use |
|---|---:|---:|---:|---:|---:|---|---|
| SK1096/SB710 | 249,866 | 2.91% | 33.4% | 50,465 | 14.4% / 14.1% | Pass | Yes |
| SK1080/SB731 | 108,370 | 1.31% | 35.8% | 25,578 | 15.8% / 14.7% | Pass | Yes |
| SK8/SB714 | 21,116 | 0.17% | 78.5% | 6,600 | 17.6% / 17.7% | Pass | Exploratory only: below 50,000-SNP primary threshold |
| SK36/SB726 | 34,239 | 0.33% | 45.4% | 366 | 0.31% / 0.17% | Fail | Excluded from ancestry inference |

All four individuals are genetically female. None has sufficient mitochondrial depth for a reliable contamination or mtDNA haplogroup estimate; X-chromosome contamination is therefore also unavailable under the preregistered sex-specific procedure. These are missing estimates, not evidence of zero contamination. SK36's negligible PMD retention and lack of terminal damage make its ancestry calls unusable even though its raw SNP count exceeds 20,000.

![Called 1240K sites and frozen analysis thresholds for the four Başur libraries](figures/basur-1240k-coverage.png)

Because these are post-capture data, mapping fractions are capture-derived proxies rather than unbiased pre-capture endogenous-DNA estimates. Mitochondrial or X-chromosome contamination is reported only when coverage makes it estimable; missing estimates are never treated as zero.

## Ancestry results

### PCA and contemporaneous outlier screen

Neither primary Başur individual is an empirical outlier among the frozen contemporaneous regional controls. For SK1096, empirical p-values are 0.727 in regional PC1–PC4 space and 0.545 in the four-direction f4 space; for SK1080 they are 0.182 and 0.545. The low-coverage exploratory SK8 is the most distant Başur point (PCA p=0.091; f4-space p=0.182), but does not meet the frozen criterion and is not primary-eligible. Only ten controls are available, so the minimum attainable leave-one-out empirical p-value is 1/11=0.091: the preregistered p<0.01 outlier rule cannot be resolved with this panel.

![Projected PCA using present-day West Eurasian references](figures/pca-present_day.png)

![Projected PCA using the ancient regional basis, with a focal-sample inset](figures/pca-ancient_regional.png)

PCA is descriptive only. The outlier screen uses frozen 3200–2500 BCE regional controls and empirical leave-one-out distances in both ancient-regional PC space and four regional f4 directions. If fewer than 100 controls are available, the preregistered empirical `p < 0.01` outlier threshold is mathematically unresolvable and this is stated rather than replaced after seeing the targets.

### qpWave, outgroup-f3 and f4

Using the shared 40-population covariance cache, the primary rank-0 qpWave null is not rejected for SK1096 against Tatika pre (p=0.183), for SK1080 (p=0.543), or for Tatika pre plus both eligible Başur individuals as a set (p=0.325). Model-specific direct covariance gives the same decisions (p=0.269, 0.608 and 0.433 respectively). The pooled Tatika post/pre comparison gives p=0.0447 with the shared cache and p=0.0475 directly: nominally below 0.05 but above the preregistered primary rejection threshold of 0.01, and not stable to the frozen sample-definition sensitivities described below.

Directional f4 tests show one clear Başur asymmetry: `f4(Mbuti, SK1096; Tatika_Pre, Iran_GanjDareh_N)` is negative (estimate −0.00343, SE 0.000577, Z=−5.95, 178,743 SNPs). Under this stated population order, a negative sign favours the target–Tatika-pre side of the quartet over target–Iran-Ganj-Dareh affinity; it is therefore not evidence for an added Iran-Neolithic source. The sign repeats against each of the three strict Tatika-pre individuals (Z=−5.65, −4.26 and −3.91), but only for this one comparator and therefore not across the two nonredundant comparator sets required by the anomaly rule. SK1080 has no contrast at the frozen |Z|≥3.3 threshold. The pooled Tatika-post target has negative asymmetries relative to CHG (Z=−5.98), Iran Ganj Dareh Neolithic (Z=−9.93), Iran Seh Gabi Late Neolithic (Z=−7.09), Armenia Chalcolithic (Z=−3.63), Iran Seh Gabi Chalcolithic (Z=−3.36) and Natufian (Z=−3.52), but its pooled qpWave result is borderline and sample-sensitive. Outgroup-f3 values are reported in full as affinity rankings and are not treated as admixture tests.

![Directional f4 Z scores for the primary local-pre comparison](figures/f4-heatmap.png)

### qpAdm regional models

With every unused frozen candidate source rotated to the right, a one-source Tatika-pre model is rejected for SK1080 (p=0.000192) and SK1096 (p=0.00661), but passes for Tatika post (p=0.238). The richer panel therefore detects differentiation between the two Başur individuals and the very small three-person strict local-pre panel that the 11-outgroup qpWave test does not resolve.

Known regional proxy models do pass the frozen basic rules. SK1080 is compatible as a one-source clade with Azerbaijan Alkhantepe Late Chalcolithic (p=0.116) or Iran Hajji Firuz Neolithic (p=0.106), and with a two-source Tatika-pre + Armenia Karnut Kura-Araxes model (p=0.0637; coefficients 0.329/0.671, SE 0.229/0.229). SK1096 is compatible with one-source proxies from Alkhantepe (p=0.235), Titriş EBA (p=0.0902), Kuriki Chalcolithic (p=0.0616) or Armenia Talin Kura-Araxes (p=0.0609); two-source Tatika-pre + Armenia Talin (p=0.233) and Tatika-pre + Armenia Chalcolithic (p=0.0520; coefficients 0.668/0.332, SE 0.116/0.116) also pass. Tatika post is itself compatible with Tatika pre and four other one-source regional proxies.

These are cladal/model-compatibility results, not literal source assignments. Several nominally passing multi-source models have very large coefficient uncertainty and are not interpreted. The Alkhantepe, Titriş, Kuriki and Kura-Araxes labels identify sampled statistical proxies, not a unique migration, people or historical identity.

Passing qpAdm models are compatibility statements, not unique historical reconstructions. A higher p-value does not make one passing model historically true, and no ADMIXTURE-like component is treated as a discrete people.

### Kinship and IBD

SK1096 and SK1080 share 26,205 callable SNPs and have a normalized mismatch score of 1.000, an unrelated-like point classification. The other Başur pairs have fewer than 20,000 overlapping SNPs and are not estimable by this screen. At Tatika, boundary-dated I4478 and secure-post I4481 form a first-degree-candidate pair (516,807 overlapping SNPs; normalized mismatch 0.738); all other eligible Tatika pairs are unrelated-like. This is a single-estimator screen, not a definitive pedigree. Although several Tatika pairs exceed the nominal 400,000-site coverage threshold, validated segment-level IBD is not estimable from the unphased pseudo-haploid panel used here.

The available panel is unphased and pseudo-haploid. Segment-level IBD is therefore not fabricated even if a nominal site-count threshold is crossed; the project reports coverage eligibility and a READ-style mismatch screen, following the logic of [READ](https://doi.org/10.1371/journal.pone.0195491).

## Sensitivity and power

The strict q30 random-call qpWave comparison does not reject rank 0 for either primary Başur individual: shared-cache p=0.183 for SK1096 and 0.543 for SK1080; model-specific direct p=0.269 and 0.608. SK1096 remains non-rejected under majority calling (p=0.177), the q25 filter (p=0.405), the QC-relaxed Tatika-pre definition (p=0.151) and assignment of the boundary individual to pre (p=0.293). Its transversion-only analysis is eligible only at the 20,000-SNP threshold and is also non-rejected (p=0.855); its PMD≥3 call set is too sparse for formal testing. SK1080 is likewise stable under majority calling (p=0.590), q25 filtering (p=0.538), QC relaxation (p=0.348) and boundary-to-pre assignment (p=0.441), while its transversion and PMD≥3 sets are ineligible.

The pooled Tatika post/pre result is borderline and definition-sensitive: strict p=0.0475, QC-relaxed pre p=0.0283, boundary-to-pre p=0.0442, boundary-to-post p=0.0842, I4481 alone p=0.124 and I4619 alone p=0.0846. This is not a robust discontinuity under the frozen rule.

For qpAdm, the selected one-source Alkhantepe proxy remains passing for SK1080 and SK1096 under all 35 full-panel leave-one-right-population-out tests (minimum p=0.0937 and 0.1996 respectively). It also passes every eligible majority-call, q25 and local-definition check on the compact frozen deep-right sensitivity panel. Neither Başur target has enough data for a PMD≥3 qpAdm replication; SK1096's transversion set reaches the qpWave threshold but not the 50,000-site qpAdm threshold. Tatika's primary one-source Tatika-pre model passes (p=0.238), but QC-relaxed pre and boundary-to-pre substitutions fail on the compact panel (p=0.0280 and 0.0437), boundary-to-post passes (p=0.0836), and replacing the pooled pre source with I4480 or I4614 fails while I4615 passes. Both individual post targets pass. Removing the complete I4478/I4481/I4619 kin network leaves no eligible post target and is not estimable.

Power is the dominant limitation. In the 100,000-replicate planning simulation, the original five-pre/two-post design has only 22.3% power for a 10% contribution from a highly differentiated source and 6.2% for the modest-source scenario, with an observed false-positive rate of 5.1%. Even under the deliberately optimistic scalar model, 80% power for a 10% shift requires about 16 individuals per side for a highly differentiated source and 252 per side for a modestly differentiated source. These figures describe a simplified design effect, not guaranteed qpAdm performance.

![Simulated power of the current five-pre/two-post Tatika design](figures/power-current-tatika.png)

The seeded 100,000-replicate scalar simulation reproduces the earlier result for the current five-pre/two-post Tatika design and extends it to 1%, 5%, 10%, 25% and 50% contributions under high and modest source differentiation. It also reports false-positive and non-detection rates, the observed eligible Başur design, balanced 10+10, 20+20 and 50+50 designs, and optimistic minimum balanced sample sizes. These are design-level planning calculations, not qpAdm power guarantees.

## Six required answers

### 1. What can these data test, and what can they not test?

The data can test whether the callable Başur individuals and pooled Tatika chronological groups are statistically cladal with the frozen local pre-2900 BCE reference across a specified set of outgroups, whether directional affinity shifts exist toward sampled regional groups, and whether those shifts are compatible with sampled regional qpAdm sources. They can also screen authenticity, gross kinship and design sensitivity. They cannot test a global flood, identify Noah or any named individual, identify a people from an ancestry component, recover spoken language, prove that an unsampled ancestry was absent, or turn a non-rejection into proof of biological continuity.

### 2. Does any ancestry appear absent from the secure pre-event panel?

The richer rotating-source qpAdm panel does detect ancestry differentiation not fully represented by the strict three-person Tatika-pre panel: Tatika-pre-only models fail for both Başur individuals. That is evidence of local-panel underrepresentation, but not of a previously absent or unknown component. The smaller qpWave panel does not reject either individual or their joint set, neither is an empirical contemporaneous outlier, SK1096's only strong f4 comparator favours the SK1096–Tatika-pre side rather than added Iran-directed affinity, and several frozen known-regional proxies fit. The pooled Tatika-post result is sample-sensitive and Tatika post is qpAdm-compatible with Tatika pre. With this power and sampling, the pilot cannot establish literal prior absence; it shows ordinary regional differentiation beyond the sparse local reference.

### 3. Do known regional populations adequately explain any difference?

Yes, at the level qpAdm can test. Both Başur individuals have one-source compatibility models using frozen Late Chalcolithic/EBA Caucasus or Upper Mesopotamian proxies, and each also has a passing two-source model that includes Tatika pre plus a Kura-Araxes/Armenia proxy. The selected Alkhantepe proxy remains valid in all 35 full-panel leave-one-right-out tests and in every eligible processing sensitivity for both targets. Thus no unsampled source is required to obtain a stable regional model. These alternatives are not unique reconstructions: a passing one-source model means cladal compatibility relative to the right panel, not literal descent from the named site, and several multi-source coefficients are too uncertain to interpret historically.

### 4. How sensitive is the result to sample choice and low coverage?

The two usable Başur results are stable to the tested call method, mapping-quality threshold and Tatika-pre definition, and their selected known-regional qpAdm proxy survives every full-panel leave-one-right-out test. Neither has enough PMD≥3 data for the strongest damage-restricted formal replication; only SK1096 reaches the 20,000-overlap transversion qpWave threshold, still below the qpAdm threshold. The Tatika post/pre comparison is not stable at p=0.05 when boundary assignments or individual pre sources are substituted. SK8 is too sparse for the primary analysis and SK36 fails authenticity. The result is therefore robust as a statement of *no unexplained Başur anomaly in the available data*, but weak as a test for small ancestry contributions or a sharp population event.

### 5. What additional samples would provide a decisive test?

A decisive next phase needs directly dated, damage-authenticated, unrelated individuals sampled tightly on both sides of 2900 BCE at both Başur and Tatika, ideally with at least 100,000–200,000 overlapping 1240K SNPs and enough endogenous data for PMD/transversion replication and contamination estimation. The planning model suggests at least 16 per side for 80% power to a 10% contribution from a highly differentiated source; a modestly differentiated source may require hundreds per side. A practical staged design would first target 20+20 high-quality individuals, balance cemeteries and archaeological contexts, predefine date-boundary handling, and expand if the observed source contrast is modest. Denser regional sampling from northern Mesopotamia, the upper Tigris, the Zagros and the southern Caucasus is as important as increasing the local count.

### 6. How could a reproducible signal later be compared with a Cain-window Susiana genome?

Pre-register a second, symmetric analysis before opening the Susiana result: define the archaeological “Cain-window” and dating uncertainty; apply the same authenticity and SNP-overlap gates; place the Susiana genome and the Başur candidate(s) on the same SNP panel; test qpWave rank, f4 symmetry and qpAdm models with identical deep outgroups and a rotating regional-source design; and require replication in independent individuals plus PMD/transversion call sets. Compare effect directions and model residuals, not labels or visual PCA proximity. A shared unexplained residual would still indicate only ancestry not represented by the sampled reference panel; it would not identify Cain, Noah, a family lineage or a language.

## Interpretation boundary

A non-detection is inconclusive under low power. A modest shift with a passing known-regional model is ordinary migration under the preregistered rules. A poorly fitting individual first triggers checks of date, authenticity, coverage, kinship and known regional sources. Even a replicated unexplained outlier would motivate further sampling; it would not identify a named biblical person or a language.

## Deliverable map

- Frozen design: `preregistered-analysis-plan.md`
- Tatika and Başur metadata: `sample-metadata.csv`, `metadata/tatika-verification.md`
- Sequence provenance and QC: `data-manifest.tsv`, `processing-and-quality-control.csv`, `uniparental-and-contamination.csv`
- Formal results: `results/f-statistics.tsv`, `results/qpwave-summary.tsv`, `results/qpadm-summary.tsv`, `results/outlier-screen.tsv`, `results/pairwise-kinship-screen.tsv`
- Sensitivities: `results/sensitivity-summary.tsv`, `results/qpadm-sensitivity-summary.tsv`, `results/qpadm-leave-one-out-summary.tsv`
- Power: `power-analysis-report.md`, `results/power-simulation.tsv`, `results/power-minimum-balanced-sample-size.tsv`
- Figures: `figures/`
- Reproduction: `README.md`, `scripts/`, `environment.yml`, `environment-details.md`