Sumerian King List Paleoclimate Alignment Test

Context

The Sumerian King List starts with eight kings who ruled before the flood. Their reign lengths are enormous and unusually regular. Three examples are 28,800 years, 36,000 years, and 43,200 years. Most are integer multiples of 3,600, and all eight are multiples of 600. Their total is 241,200 years.

The input sequence tested here is the ETCSL composite antediluvian list:

Order	King	Reign length
1	Alulim	28,800 years
2	Alalgar	36,000 years
3	Enmenluana	43,200 years
4	Enmengalana	28,800 years
5	Dumuzid	36,000 years
6	Ensipadzidana	28,800 years
7	Enmenduranna	21,000 years
8	Ubara-Tutu	18,600 years
	Total	241,200 years

The transliterations follow the composite text cited below. Like the post-flood sections of the King List, the antediluvian list is a textual tradition with variants, so the table is an analysis input rather than a claim about literal historical reigns.

One speculative interpretation treats these numbers as a distorted memory of prehistory. Under this hypothesis, the reign boundaries encode real climate shifts, eruptions, impacts, or sea-level changes. After rescaling and anchoring the list to a proposed flood date, the boundaries should coincide with dated events in the geological record.

Here, I test that idea with an exploratory analysis. The explorer rescales each chronology to a fixed 241.2 ka span, anchors one boundary, and compares the resulting dates with a catalog of Quaternary events. I use 11.6 ka BP, or about 11,600 years ago, as an analyst-chosen anchor near the Younger Dryas termination. The King List does not provide that date or suggest this paleoclimate interpretation.

Finding matches is easy. With nine boundaries and freedom to change the anchor or bandwidth, chance alignments are common. The relevant question is whether the observed Sumerian reign order scores unusually high under a clearly defined null model.

Results

• In the primary paleoclimate catalog, the Sumerian sequence does not show a statistically significant alignment. At the fixed 11.6 ka anchor and a kernel bandwidth of σ = 1.60 ka, the permutation p-value is 0.350. After adjustment for multiple comparisons, q = 0.622.
• Expanding the analysis to all 103 usable catalog entries increases the number of apparent matches but does not change the conclusion. At the same bandwidth, the wide-catalog p-value is 0.148 and the adjusted q-value is 0.430.
• The smallest raw p-value for the Sumerian sequence occurs in the catastrophic exploratory catalog at σ = 1.60 ka: p = 0.021. This was the best result found in a larger exploratory search. After adjustment for all comparisons, q = 0.222, so the result does not support the hypothesis.
• As a secondary check, I also count how many events fall within a fixed distance of a boundary. The kernel score is the main measure because it gives less weight to events farther away instead of using an abrupt cutoff.

What Would Count as Evidence?

A credible alignment would need to meet three conditions.

First, the text must determine which boundary to anchor. The Sumerian King List places the flood after the eighth reign, so I anchor the end of the antediluvian sequence. The text does not assign that boundary an absolute date. I use 11.6 ka BP as an analyst-chosen date near the Younger Dryas termination.

Second, the Sumerian sequence should look unusual next to other ancient chronologies. I compare it with a sequence derived from Biblical patriarchal ages, seven god and demigod reigns from the Manethonian fragment preserved in the Excerpta Latina Barbari, and the first eight listed Kish I rulers. These are examples for comparison, not independent statistical controls, and I do not test whether one chronology outperforms another. They show how readily unrelated ancient sequences can produce apparent matches under the same procedure.

Third, a credible result should remain significant after accounting for the tested chronologies, catalog tiers, and bandwidths. None does.

Interactive Explorer

Use the selectors below to switch chronology, catalog tier, and bandwidth. The primary paleoclimate catalog defines the main analysis. The wide and catastrophic catalogs provide exploratory sensitivity analyses. The fixed-anchor section uses p-values precomputed in Rust. The sliding-anchor section recomputes the anchor sweep in the browser.

Sumerian King List alignment explorer with fixed-anchor p-values and exploratory anchor sweeps.

Chronology

Catalog

Bandwidth

Fixed-anchor analysis: Gaussian kernel with anchor fixed at 11.6 ka (precomputed in Rust)

Kernel score

-

at fixed anchor

Permutation p-value

-

raw p; BH-adjusted q

Binary hits

-

sensitivity measure

Adjustment for multiple comparisons across all fixed-anchor permutation tests: no result has q < 0.05.

Exploratory anchor search: best anchor from 10 to 13 ka (live)

Best anchor in window

-

maximum kernel score from 10 to 13 ka

Maximum kernel score

-

relative to the fixed 11.6 ka anchor

Monte Carlo exceedance estimate

-

3,000 windows with centers uniform from 0 to 100 ka

Kernel score as the anchor slides from 0 to 30 ka BP

Anchor (ka BP)11.6

Timeline at current anchor

-

How to Read the Numbers

The main statistic is a Gaussian kernel proximity score:

S = Σᵢ Σⱼ exp(-(tᵢ - bⱼ)² / 2σ²)

Here tᵢ is an event date and bⱼ is a reign boundary. A nearby event contributes almost one point, while the contribution of a distant event approaches zero. This avoids the abrupt discontinuity of a hard cutoff, where an event just inside the window counts and one just outside does not.

The two statistical questions are different. The simplest way to see the difference is to ask what is allowed to move:

Method	Held fixed	Allowed to move	What is reported	Role
Exhaustive permutation	11.6 ka anchor, catalog tier, bandwidth, and the same set of reign lengths	The order of the reign lengths	Fraction of all labeled orders that score at least as high as the observed order	Primary p-value
Random-anchor Monte Carlo	Reign order, catalog tier, and bandwidth	The anchor, or the local anchor window	Fraction of sampled anchors or windows that score at least as high	Secondary sensitivity check

The two bandwidths are calibrated to have the same total weight as hard windows extending 1 ka and 2 ka on either side of a boundary. Using σ = τ·√(2/π) gives σ ≈ 0.80 ka and σ ≈ 1.60 ka. The browser truncates the kernel at 4σ for speed, matching the Rust precomputation.

The primary analysis fixes the anchor at 11.6 ka BP and reports the observed kernel score, a raw permutation p-value, and a q-value adjusted for multiple comparisons. The p-value comes from an exhaustive permutation test, not Monte Carlo sampling. The program checks every possible reign order. With the anchor, catalog tier, bandwidth, and set of reign lengths held fixed, the p-value is the fraction of those orders whose kernel score is at least as large as the observed score. Under this null model, every reign order is treated as equally plausible. The test does not account for choosing the anchor, catalog, bandwidth, or score after inspecting the data.

The primary catalog includes the Younger Dryas termination used to motivate the 11.6 ka anchor, so the terminal match is built into the setup. Its contribution is constant across reign-order permutations and is not evidence that the internal sequence is unusual.

The secondary anchor search asks how the result changes when the anchor is optimized after inspecting the data. It compares the maximum in the fixed 10 to 13 ka window with maxima from 3,000 windows whose centers are sampled uniformly from 0 to 100 ka. Both the fixed and random windows use a 3 ka width and 0.1 ka anchor spacing. The displayed Monte Carlo estimate adds one to both the exceedance count and the trial count so that a finite simulation never reports a probability of exactly zero. It is not the article’s primary p-value.

The fixed-anchor result card also reports a separate random-anchor sensitivity value. For each chronology, catalog tier, and bandwidth, the Rust program compares the score at 11.6 ka with scores from six million anchors sampled uniformly from 0 to 100 ka. This calculation is distinct from the browser’s 3,000-window maximum-score estimate. Neither value is the primary p-value.

Data and Comparators

The source catalog contains 104 dated events from the Quaternary period, grouped into 11 categories. The primary analysis uses a narrower paleoclimate catalog containing Heinrich-event dates, Greenland Interstadial onset dates, and selected Holocene climate-event dates. The resulting primary catalog contains 39 events within the 0 to 260 ka BP analysis window.

The wide exploratory catalog contains all 103 usable entries, including the 39 primary entries. It also includes meltwater pulses, Marine Isotope Stage boundaries, large volcanic eruptions, impact structures and contested impact hypotheses, geomagnetic excursions, extreme solar proton events, megafauna extinction nodes, and major cultural transitions. The catastrophic and deep-time tiers provide overlapping sensitivity analyses. Lonar is kept in the source catalog but excluded from the dashboard because it falls outside the analysis window.

Each entry records a selected date, an approximate uncertainty, source information, and whether the event is contested. The uncertainty values were estimated in different ways, so they are not directly comparable and should not all be read as standard errors. The current analysis uses only the selected dates. The wide catalog is still exploratory, even though every entry has a documented source.

For the Biblical comparator, the first eight values are patriarchal lifespans and the final 600 is Noah’s age at the flood. For the Excerpta Latina Barbari comparator, I use the seven named god and demigod reigns listed in that fragment: Hephaestus 680, Helios 77, Sosinosiris 320, Orus 28, Typhon 45, Anubes 83, and Amusis 67 years. For Kish I, I use the first eight rulers in the ETCSL composite text; that tradition has manuscript variants, so the comparator should be read descriptively rather than as a fixed historical chronology.

Each chronology is transformed in the same way. Reign lengths are rescaled so the total span equals the Sumerian antediluvian total of 241.2 ka, then converted to absolute dates by anchoring one boundary and accumulating durations backward in time. This normalization removes total duration as a factor and compares only the relative spacing of the boundaries.

The Rust program reads the same events.json catalog published with this article. It runs six million random-anchor trials per chronology, catalog tier, and bandwidth, runs exhaustive permutation tests, applies the Benjamini-Hochberg correction to the permutation p-values, and writes the result table used by the page. The TypeScript explorer loads the public catalog and result table, then calculates and displays the live anchor sweep.

Implementation Note

The expensive work runs offline in Rust rather than in the browser. The primary test is small enough to enumerate exactly: the Sumerian and Kish sequences each have 8! = 40,320 labeled orders, the Excerpta Latina Barbari sequence has 7! = 5,040, and the Biblical comparator has 9! = 362,880. Repeated reign lengths are still treated as labeled positions, so the null model asks whether this order is unusual among all reorderings of the same values.

The core calculations are explicit in the source: Gaussian kernel scoring, Heap’s algorithm for permutations, the random-anchor sensitivity check, the Benjamini-Hochberg step-up adjustment, and the deterministic JSON writer. The random-anchor sensitivity checks use a fixed seed, so rerunning cargo run --release regenerates the published result file rather than producing a new simulation each time.

Caveats

A Benjamini-Hochberg adjustment for multiple comparisons covers the 32 fixed-anchor permutation tests. It does not correct for earlier experimentation with event inclusion, anchor choices, score definitions, or comparator construction. Because this is an exploratory analysis and the full set of tests was not specified in advance, the q-values summarize this search rather than confirm a finding. No comparison has q < 0.05.

The catalog was assembled editorially rather than through a systematic review or an inclusion protocol defined in advance. Several entries describe related parts of the same climate sequence, so the 39 primary dates should not be interpreted as 39 independent observations. The permutation test keeps this catalog fixed; it cannot make the catalog independent or complete.

Catalog density varies substantially over time. Recent events are more numerous and generally better dated, so chronologies with Holocene boundaries have more opportunities to match an event. The permutation test partly addresses this imbalance by holding the anchor and event catalog fixed while changing only the reign order.

The deep-time sensitivity tier removes the crowded recent record by using only dates older than 60 ka. The Sumerian sequence scores at or below chance at both bandwidths.

The analysis does not carry dating uncertainty through the calculations. Several entries have uncertainties wider than the fixed matching windows, and the Younger Dryas impact hypothesis remains disputed. Treating every entry as an exact date overstates the precision of the result.

The reported p-values and q-values do not fully account for analyst choices, including decisions made after inspecting the matches. The interface exposes some of these choices, but the analysis remains exploratory.

Data and Code

The event catalog and generated result table are public and machine-readable:

• Event catalog with source metadata
• Precomputed dashboard result JSON
• Fixed-anchor Rust precompute, browser script, catalog, and results release v0.2.1

The event catalog records the analysis tier, date estimate, uncertainty, source URL, contested flag, and editorial notes for each entry. The result file contains the values generated by the Rust program and displayed by the browser explorer.

Selected Sources

• The Electronic Text Corpus of Sumerian Literature, Sumerian King List
• CDLI record for the Sumerian King List tradition
• Hemming 2004, Heinrich events
• Rasmussen et al. 2014, INTIMATE event stratigraphy for Greenland ice cores
• Lisiecki and Raymo 2005, LR04 benthic stack
• Earth Impact Database
• The Fragments of Manetho, Book I, including Excerpta Latina Barbari

Next Steps

A future version should put the dating uncertainties on a more consistent basis and carry them through the calculations. Repeatedly sampling plausible dates for each event would produce a range of scores and hit counts instead of a single value for each.

A stronger analysis would use a catalog assembled independently and defined before any chronology is scored. Repeating the procedure with several such catalogs would test whether the result depends on the underlying event data or on editorial curation.

This analysis explores a specific alignment claim. It finds no statistically significant evidence that the Sumerian reign order encodes the paleoclimate catalog used here.