Our journey
Where we started, what we found, and what is next.
Framed the project
The idea: compare unlike records across cultures to find things nobody has lined up. Chose the method with the best track record: a cultural record (chronicle, myth, art) set against a physically anchored one (ice cores, tree rings, DNA, astronomy, geology). Picked sky events in chronicles as the first project and comparative mythology as the second.
Started the notebook
Created NOTEBOOK.md with working rules, open-problems.md grouped by tractability, a seed table of 29 sky events with known supernovae as calibration, and a mythology leads file.
Verification pass
Ten web searches on every row marked VERIFYING or OPEN, then fetched the key papers. Lists of unsolved mysteries go stale fast: the 1181 supernova and the 1831 eruption were both solved in the last five years.
Pulled the table the search engines could not see
Search summaries said no aurora record exists for 1052. The Hayakawa 2015 catalogue is a PDF; extracted its text and read Table 2 directly. Found the 13 December 1052 white vapour to the north at Kaifeng. Lesson recorded as a working rule: always pull the primary tables.
Checked the Yuan and Ming catalogue
Extracted the Hayakawa 2017 catalogue. Found an aurora candidate on 9 February 1261 with three records, matching the proposed 1261 radiocarbon rise. Nothing at all for 1277 to 1282, so the 1279 candidate has no Chinese witness.
Downgraded and updated rows
The 5 BC star was a comet. The 369, 386, 393 and 837 guest stars are now argued to be novae, an astrophysics question, so parked. The 1458 eruption is two eruptions. The 1831 eruption is solved. The Younger Dryas platinum spike is volcanic.
Built this board
Borrowed the writing tool's idea of one link table joining any item to any other. Everything on this page is rendered from graph.json in the project folder.
Korean records for 1052 to 1053
Pulled the Yau, Stephenson and Willis 1995 catalogue (94 pages, East Asian auroras 193 BC to AD 1770) and read the 1050s. Korea logged a turtle-shaped white vapour with a blue-violet vapour inside on 18 July 1052. Nothing for the winter of 1052 to 1053. The newer Wang 2021 Korean catalogue with 2,013 records is paywalled and still to check.
Japanese records for 1052 to 1053
The same catalogue lists no Japanese aurora between 1033 and 8 March 1071. Japan was in a quiet stretch of record keeping for sky glows during the Oort Minimum. Negative result.
European annals for 1052 to 1053
Read the Annals of Ulster entries for 1051 to 1054 in full: kings killed, abbots died, a lake that moved in 1054, no sky signs. A compiled British and Irish list has only a gale on 21 December 1053 and the Rossdalla tower of fire in April 1054. The Anatolian and Constantinople catalogue is empty from 988 to 1097. Still unread: Fritz 1873 (the scan is behind a viewer that serves page images), Hermann of Reichenau's chronicle, which ends in 1054 and records sky events, and the Islamic chronicle catalogue (paywalled). Result so far: Europe is quiet for the winter of 1052 to 1053.
Pulled the raw radiocarbon numbers
Downloaded the PANGAEA dataset behind Brehm 2021 and computed Delta-14C ring by ring. Only one tree covers the 1050s: English oak STK-C10, which jumps 5.8 permil between the 1052 and 1053 rings. Then found three more studies: Kujawy oak (first high ring 1053), Furness Abbey oak (rise only at 1055, fails to replicate), and a sequoia plus Finnish pine (rise 1054 to 1055, possibly the 1054 supernova). Added the rival supernova hunch to the board.
Which ring rises first? Undecidable with today's data
Ring-by-ring Delta-14C, permil, from the papers' own tables: Brehm oak STK-C10 (England, ETH lab): 1051 -8.6, 1052 -10.6, 1053 -4.8, 1054 -3.6, 1055 -2.9, 1056 -4.9. First high ring 1053. Scifo oak BIF-E19 (Furness Abbey, England, Groningen lab): 1051 -8.5, 1052 -6.2, 1053 -7.3, 1054 -6.3, 1055 -1.8, 1056 -2.8. First high ring 1055. Rakowski oak (Kujawy, Poland): 1042 to 1052 average -11.0, 1053 -6.9, 1055 -2.6. First high ring 1053. Eastoe full-ring sequoia (California, Arizona lab): rise 1052 to 1053 per Brehm 2021 and Panyushkina 2022. First high ring 1053. Terrasi early- and late-wood sequoia (California) and Finnish pine: rise 1054 to 1055, about 4 permil. First high ring 1055. Tally: three series say 1053, three say 1055. Two of the disagreeing series are sequoias from the same mountains, so the offset is more likely laboratory or ring-dating differences than two real events. Panyushkina et al. 2022 reached the same conclusion with dynamic time warping: one event, smeared by a one- to two-year dating offset, cause undecidable. What this means for our hunch: the timing test cannot confirm or kill a December 1052 solar event right now. Three independent trees put the first high ring at 1053, which is what our hunch predicts. Three others put it at 1055. Until several trees from different places are measured ring by ring in one laboratory, the published series cannot settle it. Dee et al. 2016 found no radiocarbon rise at four other historical supernovae, which weighs against the supernova rival but does not settle it either.
Fritz 1873 read: Europe has no aurora on record from 1031 to 1069
Fetched the Slovenian digital library's own OCR text of Fritz's catalogue (956 kB) by opening a session first. The chronological list runs: 1030 Kassel; 1031 doubtful; then nothing until 1069 doubtful, 1074 doubtful, 1084 armies in the sky. So the largest European catalogue has no aurora anywhere in Europe for the winter of 1052 to 1053, or for twenty years either side. That is the Oort Minimum, when the Sun was quiet. Hermann of Reichenau's chronicle and the Islamic chronicle catalogue are still unread, but both are print-only or paywalled.
Checked Fritz for the other candidates too
With the whole catalogue in hand, searched 1255 to 1265 and 1275 to 1282. Found a Salzburg chronicle entry for 1262, sky as if burning after sunset, and a doubtful 1280 entry of fiery beams around Corona Borealis. The 1261 hunch now has a second culture; the 1279 candidate moves from negative back to open.
Gave 1261 (and 1268, 1279) the same scrutiny as 1052
Read every East Asian entry from 1245 to 1285 in the Yau catalogue, checked the Chinese day names against the calendar, computed Brehm's tree-ring values for 1250 to 1286, and read Scifo's 1279 table. Found a Japanese two-night flickering aurora in 1263, three Korean red glows in 1260, and a Korean violet glow in October 1279. Corrected an earlier mistake: China was not silent in 1277 to 1282. Then counted how often glows were logged at all. The early 1260s were busy, the 1050s nearly empty. That flipped the ranking: the 1052 record is the more surprising coincidence, and the 1261 hunch was moved back to 'not checked yet'. Found a new puzzle: 1268, the biggest radiocarbon candidate, has no witnesses.
Get the two paywalled papers
Miyahara et al. 2022 (GRL) decides whether our pairing of aurora records with the 1261, 1268 and 1279 candidates is new or already published. Wang et al. 2021 (JGR) is the full Korean catalogue of 2,013 records. Both need library access, an author's copy, or a request through ResearchGate.
Started the 1808 mystery eruption
Checked that it is still unsolved. Read the two main studies (Guevara-Murua 2014, Timmreck 2021) and noticed neither uses lunar eclipses. Pulled NASA's eclipse list for 1808 to 1811 and found four usable eclipses. Found the Bogota weekly that Caldas edited on the Internet Archive and read his original haze note and his report on the eclipse of 9 May 1808: a normal orange-red Moon, our 'before' picture. Ferrer's timing of the same eclipse from Havana agrees. Searched American and German journals for the later eclipses without success yet.
Find someone who described the eclipse of 3 November 1808
One description of the Moon's colour on that night decides whether the eruption came before or after November 1808. Best places: US newspapers and observers, Mexico City, Lima, Havana, ships in the Pacific, and the Japanese shogunate's astronomers at moonrise.
Checked the 2023 eclipse study for our medieval eruptions, and corrected two mistakes
Read Guillet et al. 2023 in full through Europe PMC. The 1230 eruption is now dated to spring or summer 1229 by a pitch-dark eclipse in a Japanese poet's diary. The 1286 eruption is a genuine puzzle: a big two-pole sulfur signal, but a bright colourful eclipse in October 1287. Added a new hunch for it. Corrections: (1) the 1808 eclipse idea was tried by Stothers in 2005 and stalled for lack of reports; our Caldas find still fills a gap in his series. (2) The 1262 Salzburg entry means 'in the northwest', not 'between sunset and midnight'. Also flagged that Samalas (1257) and a 1262 northern eruption could explain some red glows we had counted as aurorae.
Tested the 1286 puzzle with the monthly ice-core data
Downloaded the 2023 study's supplementary spreadsheets. No eclipse descriptions survive for 1284 to 1286. The monthly sulfate shows Antarctica rising from August 1285 and Greenland a year later, but a calibration against 1229, 1257 and 1276 showed Greenland lags like that after other eruptions too, so the lag is a dating drift and says nothing about where the volcano was. Result: the eruption probably happened in the first half of 1285, and the idea that its veil stayed in the south is weakened. A good example of why every pattern needs a calibration check.
Read the Lima newspaper that survives in Madrid: no eruption news, but big gaps
Ken read the Minerva Peruana through Memoria de Madrid and sent screenshots; Claude read them. Checked: issue 71 (17 December 1808) in full, and the front-page contents of 28 January 1809 and 1 March 1809. No report of an eruption, ash, darkness, booming or anything from the southern provinces. But the paper is almost all war news from Spain, and Madrid's copy is missing most of January and February 1809, the weeks when news from the Bolivian border would have arrived. Peru's national library, which may hold them, would not load for either of us. Verdict: mild evidence against an Andean volcano, not decisive. Worth retrying Peru's library another day.
Narrowed the 1458 mystery volcano with glass chemistry
Downloaded the shard-by-shard chemistry from the 2025 Vostok study and the reference glass from a 2023 study of Taupo ash in Antarctica, and scored every reference against the 1458 rhyolite. No match is convincing, Reclus is not special, and potassium rules out ocean-arc and Antarctic volcanoes, leaving the southern Andes or New Zealand. New Zealand's ash record argues against it, so the Patagonian Andes are favoured. Also checked and set aside: the Loisels drift pumice found in Maori sites (wrong date, wrong rock).
Checked the whole southern Andes ash database against the 1458 glass
Found the open BOOM database of southern Andean ash on GitHub and scored all 39 eruptions with enough rhyolite analyses. Nothing matches well, and none of the eruptions dated near the 1400s fits at all. Combined with the New Zealand and Antarctic checks, the 1458 rhyolite matches no known eruption anywhere we can test. That is as far as public data goes; the next step needs trace-element analyses of the ice shards or new lake cores from Patagonia.
Started the myths track: tested sky-fall myths against young craters
Found a public-domain copy of Berezkin's world motif catalogue with map coordinates, built a list of 15 impact sites young enough to have been seen, and tested whether stories of falling rocks, a fallen sky or a world fire cluster near them. They do not. The rare 'rocks fall from the sky' motif turned out to be the Chinese Nuwa story, with no crater anywhere near. The country-by-country shuffle caught a false signal that a simple count would have reported.
Read the Miyahara 2022 paper Ken found
Ken found the paper free to read and pasted it. It changed three hunches. 1261: already considered by the authors and the timing rules it out. 1268: our 'no witnesses' puzzle was wrong, since a Korean record of 15 February 1269 exists and is already paired. 1279: the authors say no aurora is known, but the Korean violet glow of 25 October 1279 falls inside their own timing window, so this becomes our second 'looking good' hunch. Also downloaded their ring-by-ring data and learned that oak records can lag a year.
Read the Goryeosa ourselves for every target year
Ken pasted the Wang 2021 paper, which showed its catalogue only lists red glows, so it would never contain our white 1052 or violet 1279 records. So we pulled the original Goryeosa omen and astronomy volumes from Chinese Wikisource and built a date converter, checked against five published dates. Found: a red glow on 19 December 1268, a second witness for the 1268 to 1269 event; a comet in February 1053, a possible innocent explanation for our 1052 record; strong red glows in March 1278 and a large sunspot in August 1278; a fireball in March 1261; and confirmation of the 1279 violet glow in the original wording.
New mystery: tested the sweating sickness against tree-ring wetness
Picked the English sweating sickness because it has dated outbreaks on one side and a public 2,000-year tree-ring wetness map of Europe on the other. Checked first that nobody had tested the 'it follows wet years' claim with tree rings. Result: a weak lean toward wet years one and two years before, not convincing, and 1517 breaks the rule. Next: the French Picardy sweat, which has 196 outbreaks and so much more statistical power.
Doubled the sweating sickness test with the French Picardy sweat
Found Hirsch's 1883 table of French outbreaks on the Internet Archive and extracted 79 outbreak years. Tested them against northern France's tree-ring wetness with a test that keeps the clustering of outbreak years. Same direction as England: wetter summers before outbreaks. Combined with England on one measure chosen for both, about a 3 percent chance by luck. Raised the hunch to 'looking good', with its caveats written down.
Tested both versions of the hantavirus idea outbreak by outbreak
Rebuilt Hirsch's French table from word positions (171 of 194 outbreaks, with departments), downloaded a gridded summer temperature reconstruction, and tested each outbreak against its own place. The wet hint held at the same weak level but turned out to be tiny in size, so it went back to 'not checked yet'. The seed-boom version found nothing. A good example of a hint shrinking when tested more carefully.
Tested Korean dim-Sun records as a volcano detector
Parsed the Goryeosa for every dim, red or lightless Sun and compared with the newest ice-core eruption list. Mostly spring dust; caught only the two biggest eruptions (1108 and Samalas 1257, with a 'blue Sun' in 1258). Then turned the idea toward the 1808 mystery: the later Joseon Royal Secretariat Diary records the weather every day, so a dim or red Sun in a Korean winter (no dust season) after December 1808 would support a volcano west of the Americas. Collecting that diary's daily weather for three winters.
Read the Korean court weather diary for the 1808 eruption winter
Collected the daily weather line from the Royal Secretariat Diary for three winters (1807 to 1810). The weeks after the eruption, about November 1808 to January 1809, are complete and show nothing unusual in Seoul's sky. Logged as weak evidence against an Asian or west Pacific source that would have been obvious in Korea.
Checked the Constantinople omens of May 1453 against the sky and the weather
Found that the ice-core list still has a separate big northern eruption in 1453 (not Kuwae), and that summer 1453 was the coldest in Europe for decades. Read Barbaro's eyewitness diary and Kritovoulos; computed the eclipse; counted modern Istanbul fogs. The Moon omen is an ordinary partial eclipse, the lights fit the Ottoman camp fires, but a whole day of fog in late spring is truly rare. Started downloading the Korean daily annals for 1451 to 1459 to date the eruption.
Read the Korean royal annals day by day for 1447 to 1458
Downloaded about 3,600 days of the Annals of the Joseon Dynasty in classical Chinese and counted every sky entry. The Annals of the 1450s never describe the Sun's colour and record few eclipses, so they cannot date the 1452/53 eruption. Logged the winter 1452 to 1453 dark fogs and the 16 November 1453 total eclipse as clues, with their caveats.
Searched for the 1453 volcano: Pinatubo fits the date, the sulfur says no
Listed every known eruption that could date to 1450 to 1453. Pinatubo's Buag eruption was the one big northern tropical match; recalibrated its radiocarbon with IntCal20 (1438 to 1454). Then read Burke et al. 2023: the 1453 sulfur arrived in Greenland below the ozone layer, so the volcano was outside the tropics. Pinatubo is out as the main source, kept only as a possible second eruption. Also finished the Korean check with Kuwae 1459: the Annals miss that too.
Tested the dancing plagues against ergot weather
Transcribed Hirsch's table of ergot epidemics (591 to 1879) and the list of dance outbreaks. Checked that ergot epidemics really do follow cold summers (they do, strongly). Then showed the dances do not: they follow dry, warm years, and the 1518 Strasbourg dancing came after one of the worst spring droughts in centuries, confirmed by tree rings and by written weather records.
Checked Strasbourg grain prices around the 1518 dancing
Used Allen's copy of Hanauer's Strasbourg prices. Bread grain was very dear in 1516 and 1517 and back to normal in 1518. This confirms the dearth but corrects our timeline: the dancing came a year after the worst hunger. No price signal across the other dances.
Checked the 100,000-year 'lost Pleiad' claim with star catalogues
Ran Pleione's motion backwards with Hipparcos, Gaia and the 25-year Hipparcos-to-Gaia baseline: 8.6 arcmin from Atlas 100,000 years ago, matching Norris and Norris. Then tested their glare argument against Pleiads that are already free of glare and only slightly fainter: most people still cannot see them. The six-star count comes from a brightness step, not glare, so the 100,000-year date is not supported.
Searched Japanese court diaries for the 993 storm; found a 1098 aurora instead
Ken searched the Tokyo court-diary database by hand. No red or white vapour in 992 to 993. But a Chuyuki note placed at 承徳2年8月24日 describes strange clouds in the north and two white bands across the sky; converted to 21 September 1098, the same night as an Augsburg record of blood-red clouds to the north, before auroras across Europe on 25 to 27 September. Ken then sent the printed page, which corrected our misreading of 奇 'strange' as 赤 'red'.
Checked Korea for the 1098 storm: silent
Listed every Goryeosa entry for King Sukjong's 2nd to 4th years. Nothing in the 8th month of 1098. The 1098 lead is parked as a weak same-night coincidence.
Tested the 1207 BC eclipse claim for Joshua at Gibeon
Wrote an eclipse calculator from NASA's Besselian elements and a base-rate scan of 1300 to 1100 BC, allowing for uncertainty in Earth's rotation. The 1207 BC ring of fire did cross Gibeon and is rare there; but it was near sunset, while the text says 'midst of heaven'. Results match Khalisi 2021.
Reran the Odyssey eclipse date search over 1,400 years
Installed Astronomy Engine, checked it against the paper's own numbers for 1178 BC, then applied the paper's rules to every new Moon from 2000 to 601 BC. The clues fit a date about every 64 years, not once in 2,000; and the eclipse had only about a 1 in 3 chance of being total at Ithaca.
Checked whether Europe could see the 1054 supernova
Computed the star's height at dawn in July 1054 for China, the Near East and Europe (nearly the same), showed the Japanese '4th month' date is impossible (star beside the Sun) while the 5th month fits, and compared Europe's record of the three medieval supernovae. Visibility is not the answer; record keeping is. A full comet base rate (Hasegawa 1980) could not be downloaded.
Read US death rates by cause 1910 to 1938 and tested sleepy sickness against flu
The 1943 Census volume prints its tables sideways and the scan text is garbled, so the influenza and encephalitis rows were read from rotated page images. Encephalitis did not rise in 1918; it peaked in 1920 and 1923. No one-year lag behind flu; a modest, patchy same-year link.
Tested sleepy sickness against flu in England and Wales
Collected influenza and encephalitis lethargica figures for 1916 to 1937 from 19 yearly Registrar-General reviews. No same-year link and no one-year lag. The modest US same-year link does not replicate.
Compared China's 11th-century comets with Europe's
With the NASA library still down, used Williams 1871 for China and Pingre 1783 for Europe. Europe caught about 4 or 5 of 11. The 1054 silence is ordinary for the time.
Added Scotland as a third country for sleepy sickness vs flu
Read eight Scottish Registrar General reports (1930 to 1937) for influenza and encephalitis lethargica deaths, 1921 to 1937. No link in the same year or a year later. Three countries now show no consistent tie between flu waves and sleepy sickness.
Tested the cocoliztli epidemics against the Mexican Drought Atlas
Found the epidemic list (Acuna-Soto 2004) and the 600-year Mexican Drought Atlas. No drought in the year before; epidemic years were wet in central Mexico, and the 16th-century megadrought was a northern event. A second central record (Puebla maize) agrees about the wet 1570s but not the 1540s.
Added the Queretaro bald cypress as a third central-Mexico record
The Queretaro record shows 1514 to 1539 as the driest stretch in 600 years and 1540 to 1580 as the wettest, agreeing with the drought atlas over the Puebla maize record for the 1540s. It shows no wet signal in individual epidemic years, so that result was downgraded.
Tested the Maya city shutdowns against two drought records
Extracted the last dated monument for 89 cities (Ebert et al. 2014) and compared with the Yok Balum stalagmite and the Cariaco record. The broad drier era is confirmed, but the timing of individual shutdowns does not follow dry spells, and the two climate records disagree about the final decades.
Checked the Roanoke and Jamestown droughts with the seasonal rainfall atlas
The many-site atlas confirms an extreme drought at Roanoke but places it in 1586 to 1588 (1587 the worst), overlapping the first colony too; Jamestown's 1606 to 1612 drought is real but less extreme. The published mortality-drought correlation could not be reproduced from the archived data.
Tested the Ugarit eclipse tablet dates for the Bronze Age collapse
Computed the three proposed eclipses at Ugarit with NASA's elements, under both the 2004 and 2016 Earth-rotation models, plus a base-rate scan of 1400 to 1180 BC with Mars positions. None is securely total; 1223 BC needs a rotation value about 2 sigma off. The drought evidence (Manning 2023) was not re-tested.
Checked the Mary Celeste's drift against the 1872 weather
Fetched the 20th Century Reanalysis around the Azores (too blurry: it misses the gales) and the 1872 ship reports (ICOADS, Dutch logs). The ships logged gales and a force 10-12 storm on the drift route. The 375-mile drift needs only about 5 percent of the real wind speed, but she must have sailed across the wind on the port tack for much of the time. The 1873 inquiry's 'no heavy weather' reason for a late abandonment does not hold. Prior work by MacGregor and Richardson (2007) still to be read.
Read the 2007 Smithsonian account: the drift result was already known
Ken pasted Blumberg 2007. Richardson and MacGregor had already used ICOADS and found the ship sailed herself from Santa Maria. Our check reproduces that. Problem set to PARTIAL (their scenario: false alarm about leaking near land). The storm after abandonment, against the 1873 reasoning, is the only possibly new detail.