# Exploratory audit of ancient-DNA kinship and chronology anomalies

**Date:** 10 August 2026
**Study type:** prospectively frozen exploratory audit
**Frozen chronology:** primary 2200–1000 BCE; sensitivity A 2370–1000 BCE; sensitivity B 2033–1000 BCE
**Frozen geography:** Armenia/southern Caucasus, Anatolia/Aegean, northern Mesopotamia/Syria, Iran, Levant, Egypt and Upper Nile/Nubia
**Overall result class:** **candidate-generating, with a null primary test**

## Plain-language verdict

This study found **no high-quality ancient-DNA case in which close genetic relatives have independently measured death dates that remain incompatible with extreme but ordinary human lifespans and generation intervals**.

Nine pairs met the frozen high-quality genetic and direct-dating gate. None crossed even the modest source-investigation threshold. The largest probability of exceeding an ordinary-human hard limit was 0.151, below the frozen 0.20 screen and far below the 0.95 stringent threshold. No case survived the multiplicity rule. The preregistered positive-result rule therefore fails at its first and most important evidential steps.

Seven apparent discrepancies appeared only when contextual, modelled or low-coverage evidence was admitted. Primary-source checking explained all seven through ordinary archaeological and metadata problems: commingling, reused or multigenerational tombs, coarse phase ranges, a date tethered to relatives, overlapping direct dates, or inadequate genetic coverage.

The extended ancIBD analysis did identify one useful follow-up target. Three strong IBD edges link people at Noratus and Beniamin whose current mean dates differ by about 322–325 years. They all belong to one connected family network. One person is dated only by a kinship-tethered estimate, two retain only the archaeological date of a group burial, and exact topology is not published for two edges. The network is therefore **not evidence of exceptional longevity or delayed maturity**, but it is a good candidate for direct radiocarbon dating and renewed sample/context checking.

The independent literature search found no qualifying same-person multi-tissue date discrepancy, cementum anomaly, adult-with-juvenile-development case, or surviving relative-date incompatibility in scope. It found many explicit ordinary failure modes instead.

The responsible conclusion is:

> Public evidence presently provides no positive support for the proposed long-generation lineage. One context-dated family network merits new direct measurement, while the usable high-quality dataset is too small and geographically uneven to exclude a rare lineage.

## 1. What was prospectively frozen

Before opening new pair-level AADR outcomes, ancIBD outcomes, candidate supplements or anomaly-search results, the study fixed:

- the chronology and route geography from the existing website model;
- the primary, sensitivity and earlier-local-control periods;
- target and off-route countries;
- person-level deduplication and relationship classes;
- G1/G2/G3 genetic and D1/D2/D3 dating tiers;
- topology-specific hard limits from Sedig et al.;
- full-distribution date-separation calculations;
- source-investigation, stringent and multiplicity thresholds;
- matched-control variables and the prohibition on outcome-based matching;
- the positive-result rule;
- two injected-effect upper-bound scenarios;
- conventional anomaly vocabulary and source-resolution requirements; and
- random seed `20260810`.

The frozen files and hashes are in `FROZEN-MANIFEST.md`; the pre-outcome git commit is `b63f16179137e1558a7fd5e99026904cbf25a39b`. Earlier null studies and already seen false positives were disclosed in `prior-observations.md` so that they could not be presented as newly discovered confirmations.

The result definitions were:

- **Source investigation:** `P(date separation exceeds hard limit) >= 0.20`.
- **Stringent unresolved discrepancy:** G1+D1, source-validated topology, probability at least 0.95, and robustness to identity, calibration, tissue and reservoir checks.
- **Multiplicity survivor:** additionally `m × (1-P) <= 0.05` across primary topology-eligible pairs.
- **Conditionally positive pattern:** at least three multiplicity survivors, at least two independent networks/sites and publications, target enrichment (`p <= 0.05`, OR at least 2), and independent non-kinship support.

No threshold was weakened after seeing results.

## 2. Chronology and route interpretation

The existing website chronology was treated as the hypothesis source, not as an outcome to be adjusted:

| Website-model event | Frozen date |
|---|---:|
| Cudi arrival | about 2900 BCE |
| Shinar gathering | about 2800 BCE |
| dispersal onset | 2370 BCE |
| Peleg/Babel completion | about 2033 BCE |
| proposed route arrivals | 2250–2033 BCE |

The study did not assume that a route requires mass population replacement or a detectable ancestry pulse. Founder movement, elite contact or institutional influence are compatible with the route model. Conversely, a kinship anomaly in a route region would not identify Noahic ancestry. The analysis could test only whether the sampled chronological relationships are unexpectedly difficult to reconcile with ordinary humans.

## 3. Data assembled

### AADR family relations

AADR v66.p1 contained 23,089 annotation rows. The extractor parsed 18,261 relationship assertions into 8,303 unique unordered person pairs, with zero unresolved identifiers, self-pairs or unparsed relation assertions. The relation inventory included 731 parent–offspring pairs, 497 sibling pairs, 574 unresolved first-degree pairs and thousands of second/third-degree calls.

Within target countries, 167 AADR pairs were discoverable; 64 had at least half of both reconstructed date distributions in the primary period. Thirty-one of those had a topology class with a frozen hard limit. Nine met the provisional G1+D1 primary gate.

### ancIBD extended relationships

The public ancIBD segment archive contained 952,655 segment rows. These were streamed and collapsed into 705,555 canonical pairs. All 20,486 pairs in published Supplementary Table 2 matched the streamed segment aggregation, with no count mismatch and maximum metric differences below `5×10^-7`.

The frozen extended-edge gate required both people to pass the published chromosome-3 imputation-confidence threshold, a published pair summary, and a maximum segment longer than 12 cM. Current AADR v66 dates were joined where available because the 2024 ancIBD supplement contains older metadata. In the widest sensitivity period, 179 target-region edges were retained, 166 at the G1 coverage proxy, forming 10 connected networks. No topology-specific hard ceiling was invented from IBD length.

### Complete retained table

`results/complete-related-pairs.csv` contains 232 unique target-region pairs in the widest frozen period:

- 64 AADR family-annotation pairs;
- 179 published/high-confidence ancIBD edges;
- 11 pairs present in both channels; and
- nine pairs eligible for the primary calibrated-probability test.

This is complete for the acquired public inputs under the frozen gates, not complete for all excavated people, unpublished genomes, zero-IBD comparisons or unindexed grey literature.

Coverage was very uneven: 148 retained pairs involved Armenia, 41 Greece, 16 Jordan, 10 Israel and nine Turkey, with a few Iranian or cross-border edges. No usable retained pair represented Iraq/Syria, Egypt or Upper Nile/Nubia. All nine primary pairs came from only two sites/networks: Mygdalia in Greece and Oylum Höyük in Turkey.

## 4. Primary calibrated-date result

Raw conventional radiocarbon ages and errors were recalibrated on a one-year IntCal20 grid. Broad and multimodal distributions were retained. For each pair the analysis convolved both complete date distributions and computed `P(|T1-T2| > L)`, where `L` is the frozen topology-specific hard limit from Sedig et al. ([DOI 10.1016/j.jas.2021.105452](https://doi.org/10.1016/j.jas.2021.105452)).

| Primary result | Value |
|---|---:|
| G1+D1 topology-eligible pairs | 9 |
| connected networks | 2 |
| `P >= 0.20` screens | 0 |
| `P >= 0.95` stringent cases | 0 |
| multiplicity survivors | 0 |
| maximum probability | 0.150547 |

The maximum was the directly dated Oylum Höyük sibling pair I14784/I14789. Its median absolute date separation was 54 years, with a wide 2.5–97.5% range of 2–208 years created by the calibration distributions. It remained below the frozen 135-year sibling screen probability.

Eight primary pairs came from one Mygdalia infant grave. Seven directly dated human-bone samples form a reconstructed sibling/second-degree pedigree. Individual pair medians ranged from 27 to 110 years, but every probability of exceeding the relevant hard limit was below 0.01 except the ordinary sibling pair MYG001/MYG008 at 0.0082. These eight rows are not eight independent tests; they are one negative family network.

![Frozen pair screen](figures/kinship-chronology-screen.png)

## 5. Automated lower-quality flags and their resolution

When all genetic and dating tiers were admitted, 31 topology-eligible target pairs yielded seven `P>=0.20` source-investigation flags. None reached 0.95. Every case was retained in `results/candidate-resolution.csv`.

| Pair | Probability | Why it does not survive |
|---|---:|---|
| I18276/I18483 | 0.746 | Published mother–son pair, but both dates are coarse archaeological phase ranges at different Armenian sites; no independent direct chronology. |
| HGC037/I9005 | 0.616 | Loose teeth from the highly commingled Hagios Charalambos secondary ossuary, which spans centuries and lacks meaningful stratigraphy; neither is directly dated. |
| AID014/AID017 | 0.596 | Fragmentary secondary/commingled Aidonia burials dated by pottery, termini and tomb context. |
| ALA001/ALA038 | 0.395 | Direct-date intervals overlap and the source describes an ordinary parent–offspring family; ALA038 has only about 62,000 1240K SNPs, making this G3. |
| I14782/I14784 | 0.311 | Father is undated and tethered/modelled from directly dated relatives; not two independent dates. |
| I14782/I14789 | 0.292 | Same tethered father and same lack of independent chronology. |
| I15731/I18166 | 0.250 | Explicit son–mother pair sharing one broad LBA-2 context interval; the mathematical tail is not an observed gap. |

These are ordinary examples of why contextual ranges must not be treated as direct death measurements. The lower-quality channel is therefore **artefact-positive**, not biologically positive.

## 6. The Noratus–Beniamin follow-up network

The ancIBD data provide the study's one genuinely useful lead.

### Initial apparent anomaly

The 2024 ancIBD supplement placed I18470 at Noratus in 1150–1050 BCE and R11675 at Beniamin at 1492–1325 cal BCE. The mean difference was 322 years. This was not a weak isolated segment: the pair shares 20 segments longer than 12 cM, totalling 1037 cM, with published relatedness `r=0.168`.

### Metadata correction

AADR v66 now labels I18470/R11675 a 2.5-degree pair and dates I18470 to 1550–1300 BCE by tethering the otherwise undated burial to radiocarbon-dated close relatives. Its current mean is only three years from R11675 and the intervals overlap. The strongest initial anomaly is therefore a demonstrable stale-metadata false positive.

### Residual edges

The correction creates three related review edges:

| Pair | Current mean difference | IBD >12 cM | Main weakness |
|---|---:|---:|---|
| I18470/I18481 | 325 years | 742 cM | I18470 tethered/modelled; I18481 Tomb 1 context only; exact topology unpublished. |
| I18470/I19321 | 325 years | 873 cM | AADR 2.5-degree label, but modelled date versus Tomb 1 context. |
| I19321/R11675 | 322 years | 420 cM | Direct Beniamin date versus Noratus context; exact topology unpublished. |

All three belong to the same large Armenian IBD component and depend on the same individuals and dating problem. Noratus Tomb 1 is a group burial; the source dates its sampled people archaeologically to EIA-1, while I18470 has since been pulled earlier using kinship. Mixed or redeposited remains, tomb reuse, intrusion or incorrect context attribution remain plausible ordinary explanations.

These edges justify the study's overall **candidate-generating** label. They do not satisfy D1, topology security, multiplicity, independent-network replication or independent biological support. They contribute zero cases to the positive-result rule.

## 7. Matched controls and target enrichment

The frozen matching algorithm used relationship class, period, minimum SNP coverage, context proxy, direct-date completeness, publication and laboratory indicators. Outcome probabilities and flags were forbidden matching variables.

Thirty-one target topology-eligible pairs were matched to up to three controls in each panel.

### Off-route West Eurasian panel

- eligible pool: 293;
- unique matched controls: 93;
- target lower-tier flags: 7/31;
- control flags: 0/93;
- raw Fisher odds ratio: infinite because of the zero control cell;
- Haldane–Anscombe corrected OR: 57.2, approximate 95% CI 3.16–1037;
- one-sided Fisher `p=3.49×10^-5`.

This is a real descriptive enrichment of **noisy source-investigation flags**, not of high-quality discrepancies. All seven target flags fail source validation, whereas the primary high-quality target rate is 0/9. The enrichment probably captures uneven archaeological contexts, regional publication structures and the concentration of commingled/reused burial assemblages, rather than extraordinary biology. It cannot satisfy the frozen positive rule because there are no high-quality survivors.

### Earlier-local panel

- eligible pool: only six unique pairs;
- target lower-tier flags: 7/31;
- unique earlier-local flags: 0/6;
- corrected OR: 3.98, approximate 95% CI 0.20–79.2;
- one-sided Fisher `p=0.255`.

The panel is underpowered and required 87 fallback-with-replacement matches. It does not establish temporal enrichment.

One prospectivity deviation matters here: the AADR target screen was viewed before the already frozen control algorithm was executed. The matching variables and ratios were fixed and outcome-free, but the execution-order mistake weakens the enrichment comparison. It does not alter the pair probabilities or the null primary result. The deviation is disclosed in `deviations-log.md`.

## 8. Conventional anomaly-vocabulary and secondary-evidence search

Fifty recorded query/full-text operations searched primary articles, supplements, excavation reports and osteological sources for:

- radiocarbon outliers, reservoir effects and calibration problems;
- date/relatedness incompatibility;
- intrusion, reopening, reuse, commingling and sample mismatch;
- dental–skeletal discordance and delayed/unfused epiphyses;
- endocrine or developmental delay;
- cementum anomalies;
- unexpectedly adult wear/degeneration in subadults; and
- same-person multi-tissue radiocarbon discrepancies.

The search located no qualifying in-scope survivor. Instead it found:

- 12 genetically identical Southern Caucasus sample pairs classified by their authors as collection or laboratory-registration mix-ups;
- two cross-site identical pairs with unresolved provenance, excluded from site-level inference;
- a Tsaishi layer mixing Iron Age and High Medieval remains;
- archaeological assignments revised by about 200–300 years after direct dating, IBD and stratigraphic review;
- ordinary tomb reuse and commingling at several Aegean, Armenian and Mesopotamian sites;
- isolated teeth whose skeletal attribution cannot be secured;
- severe but ordinary population-level dental wear linked to diet and food preparation;
- adult pathology rather than delayed development; and
- a Kadruka person with only one direct tissue date, not a multi-tissue test.

The clearest query-validation false positive was the already known Shamanka II case outside the period and geography: DNA showed that allegedly matching mandible and postcrania were different people, with residual offsets compatible with freshwater reservoir effects.

No evidence channel independently supported extreme lifespan, delayed maturity or anomalously slow development. Full details, raw observations and authors' explanations are separated in `data/derived/literature_search_inventory.csv` and `notes/literature-search.md`.

## 9. Sensitivity analyses

The main conclusion was stable under the informative prespecified sensitivities:

| Scenario | Pairs | Screens `>=0.20` | Stringent `>=0.95` | Maximum |
|---|---:|---:|---:|---:|
| frozen G1+D1 | 9 | 0 | 0 | 0.151 |
| G1/G2 + D1/D2 | 9 | 0 | 0 | 0.151 |
| 2370–1000 BCE G1+D1 | 9 | 0 | 0 | 0.151 |
| 2033–1000 BCE G1+D1 | 9 | 0 | 0 | 0.151 |
| resolved topology G1+D1 | 2 | 0 | 0 | 0.151 |
| sibling hard limit 130 rather than 135 years | 2 | 0 | 0 | 0.164 |
| one date shifted by worst-case ±20 years | 9 | 0 | 0 | 0.192 |
| one date shifted by worst-case ±40 years | 9 | 1 | 0 | 0.240 |
| all quality/date tiers | 31 | 7 | 0 | 0.746 |

The ±40-year single-member shift creates one screen but no stringent result; common same-laboratory shifts would largely cancel.

Using **any non-zero** primary-window probability admits 14 G1+D1 pairs and creates three screens. This is not a meaningful challenge: the three pairs date to about 3500–3100 BCE, 3350–3010 BCE and 776–523 BCE. Their primary-window probability masses are between `10^-30` and `10^-177`. This sensitivity illustrates why a non-zero mathematical tail is not study-period membership.

Subregional lower-tier flags were concentrated in Anatolia/Aegean (5/22) and the Caucasus (2/2), with none in the seven Levantine pairs. No subregion contained a high-quality primary flag. Empty or sparse strata prevented useful tests for Mesopotamia/Syria, Iran, Egypt and Upper Nile/Nubia.

No primary pair had a source-identified marine/freshwater reservoir correction or enamel-only formation offset. Positive shared-context covariance would narrow separation distributions, making the independent-date calculation conservative. Full downloadable OxCal posterior grids were unavailable, so the published-posterior sensitivity could not be honestly converted into probability densities; published intervals were source-checked instead.

## 10. Ordinary versus conditional long-generation model

The planned Genesis-conditional model used pre-disclosed fathering intervals of 79–135 years and declining lifespans from Shem through Terah. It was to be a conditional likelihood/simulation comparison, not a lineage identifier.

No numerical fit was performed because the primary data do not identify such a model. The nine primary pairs consist of two resolved sibling pairs plus seven topology-unresolved second-degree pairs in one infant grave. All Mygdalia people died as infants. Long fathering intervals make no distinct prediction for sibling death-date separation, and an unresolved infant relationship cannot supply a birth-generation interval. There are zero resolved primary parent–offspring, grandparent–grandchild or other generational pairs.

A fitted likelihood ratio would therefore be driven by invented topology, not evidence. `results/conditional-model-feasibility.json` records this decision.

## 11. Quantitative upper bound under the null

The study injected two true effects into each of the nine primary pairs and estimated pair-specific detection probabilities:

1. a true separation 30 years beyond the topology hard limit; and
2. a true separation 100 years beyond the hard limit.

The 95% upper bound on the fraction of **sampled high-quality relationships carrying a detectable discrepancy** was:

| Injected effect | Pair-level effective N | Pair-level upper bound | Network-level effective N | Network-level upper bound |
|---|---:|---:|---:|---:|
| hard limit +30 years | 6.53 | 0.390 | 1.41 | uninformative |
| hard limit +100 years | 8.69 | 0.293 | 1.85 | 0.832 |

The pair-level bounds are optimistic because eight of nine pairs belong to one Mygdalia network. Network-level bounds are more honest and very weak. The audit therefore does **not** show that a rare lineage is absent.

These bounds do not constrain:

- Noahic ancestry;
- lineage frequency in unsampled populations;
- unusual longevity that does not create a detectable kinship/date discrepancy;
- Egypt, Upper Nile, Mesopotamia/Syria or other regions without usable pairs; or
- contextual networks that have never been directly dated.

## 12. Limitations

1. **Severe geographic imbalance.** The primary test has only two networks, both in the Aegean/Anatolian sphere. Several hypothesised route regions contribute no usable pair.
2. **Family clustering.** Eight primary rows are one infant pedigree; 148 extended edges form one large Armenian component. Pair counts overstate independent information.
3. **Contextual dating.** Most public relatives have archaeological or modelled dates. Those cases are useful for deciding what to date but cannot test lifespan securely.
4. **IBD topology ambiguity.** ancIBD can detect distant sharing, but exact genealogy becomes uncertain beyond third degree. Endogamy, ROH and overlapping relationship classes prevent a hard lifespan inference from IBD length alone.
5. **Positive-edge ancIBD publication.** The acquired segment graph does not contain all zero-IBD sample pairs; it is a discovery graph, not a population denominator.
6. **Literature coverage.** No institutional Scopus or Web of Science session was available. English-language and indexed sources underrepresent Armenian, Georgian, Turkish, Arabic, Persian and Sudanese grey literature.
7. **Posterior availability.** Raw measurements were recalibrated where available, but full published OxCal posterior grids were not generally downloadable.
8. **Control prospectivity.** The target screen was viewed before the fixed matching algorithm was executed. The resulting enrichment must be treated as descriptive.
9. **Search absence is not biological absence.** Failure to find explicit developmental or multi-tissue anomalies is not equivalent to re-examining all skeletons.
10. **Hypothesis non-identifiability.** A chronology discrepancy could have several ordinary causes; even an unresolved discrepancy would not by itself establish a Noahic lineage.

## 13. Most valuable next measurement

The best next study is not a larger metadata scrape. It is a focused reanalysis of the Noratus–Beniamin network:

1. directly AMS-date I19321 and I18481, preferably using securely identified collagen from the same skeletal individuals sampled for DNA;
2. directly date I18470 rather than relying on kinship tethering;
3. obtain an independent replicate date for at least one sample in a second laboratory;
4. measure stable isotopes and assess freshwater/marine dietary reservoir risk;
5. re-check skeletal labels, excavation inventory, Tomb 1 stratigraphy, possible reopening/redeposition and chain of custody;
6. repeat DNA extraction or identity checks from a second element where possible;
7. refine the network topology with high-coverage imputation, IBD1/IBD2 patterns, uniparental markers and sex constraints; and
8. analyse the dates in one joint model that includes tomb phases, direct dates, identity and kinship uncertainty without allowing kinship to substitute for the measurements being tested.

Three mutually overlapping direct dates would resolve the signal as ordinary context error. Secure, replicated, reservoir-robust direct dates separated by three centuries in a source-validated close genealogy would create a genuinely important anomaly. Until then, the network is a measurement target, not a conclusion.

## 14. Implications for the website model

No website edit was made.

The present evidence does not justify a claim that archaeology or ancient DNA supports unusually long post-Babel generations, delayed maturation or extreme longevity. Nor does the study decisively refute a rare lineage, because the independent high-quality sample is extremely small and misses major route regions.

If the research is later represented publicly, the accurate wording would be:

> A prospectively frozen audit found no high-quality kinship/date anomaly in currently usable public data. One context-dated Armenian family network has been identified for direct dating, but it presently has ordinary archaeological explanations and does not support a longevity claim.

## 15. Reproducibility and file map

- `analysis-plan.md` — frozen plan and positive rule.
- `prior-observations.md` — prior evidence disclosed before the freeze.
- `deviations-log.md` — append-only deviations.
- `SOURCE-MANIFEST.csv` — 28 source records with URLs, local paths and hashes where available.
- `README.md` — exact execution order and runtime commands.
- `scripts/` — extraction, calibration, controls, sensitivities, upper bound, network analysis and manifest builders.
- `results/aadr_pair_analysis.csv` — all 8,303 AADR related pairs with tiers and probabilities.
- `results/ancibd_retained_pairs.csv` — 179 retained target extended edges with current/old metadata comparison.
- `results/ancibd_networks.csv` — 10 connected networks.
- `results/complete-related-pairs.csv` — merged 232-pair target table.
- `results/candidate-resolution.csv` — all flags, review edges, sensitivity-tail cases and negative controls.
- `results/matched_control_pairs.csv` and `results/matched_control_results.json` — control panels and enrichment.
- `results/sensitivity_results.csv` — sensitivity grid.
- `results/upper_bound_results.json` — detection simulations and upper bounds.
- `results/conditional-model-feasibility.json` — model non-identifiability decision.
- `data/derived/literature_search_queries.csv` — 50 recorded search/full-text operations.
- `data/derived/literature_search_inventory.csv` — plausible anomalies, author explanations and explicit null categories.
- `figures/kinship-chronology-screen.png` — the single decision-relevant figure.

## Final conclusion

The frozen positive-result rule is not close to being satisfied. There are zero high-quality source-investigation flags, zero stringent cases, zero multiplicity survivors and zero independent developmental or multi-tissue confirmations. Apparent AADR anomalies are explained by ordinary dating/context/quality problems. The ancIBD network contributes one focused candidate for better measurement, not biological support.

**Final classification: candidate-generating because of one unresolved context-dated network; primary hypothesis test null.**