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. ChronologyCatalogBandwidth Fixed-anchor analysis: Gaussian kernel with anchor fixed at 11.6 ka (precomputed in Rust) Kernel score-at fixed anchorPermutation p-value-raw p; BH-adjusted qBinary 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 kaMaximum kernel score-relative to the fixed 11.6 ka anchorMonte 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.