Interesting nevertheless.
"One theory" implies that someone somewhere has ever claimed this. But has anyone? TFA gives no source for the claim, AFAICT (although I skimmed a bit).
I noticed the deviation because there are mechanical watch movements that run at 18000 and 21600 beats per hour.
The dark ages were anything but dark.
Ancient Europe was anything but primitive.
And yet it’s so rarely discussed in the mainstream.
I used to think the same thing, but I've come to appreciate exposure to highly-technical content (outside my wheelhouse) that also "gratifies ones intellectual curiosity." I don't necessarily deeply care about the merits of GUIs and TUIs, but once it's pointed out, I can see the importance of being keyboard driven (https://news.ycombinator.com/item?id=49479837). And I find it valuable fairly often to have a passing familiarity with the fact that something like this is being discussed by intellectuals in the tech space.
For smart, curious people, exposure is valuable even if you don't have the time to understand something very well.
Would you mind being a little clearer about what it is you’re saying here?
So you're not really filling me with confidence about your idea here. Perhaps you can provide some sort of evidence?
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.
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.
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
-
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.
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.
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.
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.
The event catalog and generated result table are public and machine-readable:
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.
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.
This is a forum run by a Silicon Valley startup incubator. Unfortunately tech stories and whatever the current tech thing is will always dominate. And HN's design ethos of "no silos" means everyone has to see all of that content, and it will always push everything else down, especially anything non-technical (unless the mods force it up, which they can and sometimes do.)
The only way to get what you want (I do think it's an interesting idea and I'd like to see it) would be to start another forum entirely or else build a third party client to filter out all of the AI slop. Such stories will always be diamonds in the rough here.
Some people also interpret some symbols on mammoth tusks going back 50k years as writing but that's a stretch IMO.
I did find a pattern though. If you look at ALL the kings, rather than just those 8 it becomes obvious. The recent numbers are realistic and drift further apart, becoming less and less accurate, the further back you go.
Interestingly it also compared some other clay tablets and they were close. So it seems that they were probably written down based on a verbal history that had already drifted wildly well before it was written down .