Sillok Metrics

The Veritable Records of the Joseon Dynasty counted article by article, with a reliability card on every number

Scope: the 27 reigns from Taejo to Sunjong (1392–1910). The Gojong and Sunjong records are released under KOGL Type 4 (no derivatives, non-commercial), so they enter only counts that neither alter nor quote the text (articles, characters, NIKH classification, keyword hits); titles, text excerpts and the astronomical checks exclude them, and in the published coding tables their rows keep the judgment but have their title, quote and summary blanked. The 25 reigns Taejo–Cheoljong are Type 1 (attribution).

Characters per year 1392–1910 (principal editions); bands mark reigns.

Key points

1. The Sillok is more uneven than one might expect: characters per reign-year fall from over 100,000 (Seongjong, Jungjong, Sejong, Jeongjo and others) to about 13,000–14,000 (Heonjong, Cheoljong).
2. Of eclipse observation reports, 65% match an eclipse visible from Seoul and 95% one somewhere on Earth; if article types are ignored, the rate is 24%.
3. Daytime sightings of Venus on 2,937 days reproduce the planet's synodic period to within 0.04%: observation records are observations.
4. The top-10 clan share among first appointees to high office rose from 0.28 to 0.45 — the concentration shows up in the numbers.
5. The Sejong peak in death-penalty articles is sentencing paperwork; the Sillok is not a source of penal statistics.

414,501

articles · 33 editions

52.5M

Chinese characters

65%

eclipse observation reports matching an eclipse visible from Seoul

95%

matching an eclipse anywhere on Earth

583.7

Venus period (days) recovered from daytime sightings

Every number carries a reliability card: whether it was counted straight from the source, compared with an astronomical computation, or measured on a coded sample. The site carries no full text; each article links to the National Institute of Korean History.

Volume and density of the record

Articles414,50133 editionsNIKH publishes no official article count; this table fills that gap.Characters52.5Mcharacters, 33 editionsThe only complete count on which any statement about the Sillok's size can rest: 52.5 million characters across all 33 editions, 44.6 million in the principal editions of Taejo–Cheoljong alone.Characters per reign-year205,623highest: Injong, per yearTurns the commonplace that the record thins toward the end into a number.Article length (median)43median article, principal editionsThe first thing to check before comparing article-based metrics across reigns: equal article counts with twice the median length mean twice as much text.Share of articles with a historian's comment8.3%highest: MyeongjongShows which editions carry the historians' own judgments at all.Share of months without an article65most empty months: SeonjoAn empty month means no record, not no events.

Editions compared

Size of revised editions0.10Revised Seonjo / Seonjo, charactersLets you say in one line how much of the original a revision rewrites: the Revised Seonjo records are a tenth of the original; the Supplemented Hyeonjong records are longer than the original.Shared-day share60%Revised Seonjo days also in the originalHow much of the revision sits on days the original already covers.Deletions in the Daily Records of Gwanghaegun16.4%second-draft characters marked deletedSplits NIKH's 'reduced by a third' into three distinct measures (15% of articles, 18% of characters, 16% of characters marked); say which one you are citing.Historians' comments in revised editions20.2%share with comments, Sukjong Supplement and CorrectionsAccording to NIKH's tags the revisions carry almost no commentary (a single comment in the Revised Seonjo records), compared with the 52 counted by Oh Hang-nyeong.

Sky and weather

Solar-eclipse articles matched by computation65.5%of 84 observation reports match a Seoul eclipseAnswers the question 'Can the Sillok's eclipse records be trusted?' type by type: 65% of observation reports match an eclipse visible from Seoul and 95% an eclipse somewhere on Earth; most of the rest are eclipses after sunset that were recorded from calculation.Lunar-eclipse articles matched by computation46.5%of 359 articles, not classified by typeOnly 47% of all lunar-eclipse articles match a visible eclipse within ±1 day.Daytime Venus records against the Venus cycle583.7days recovered (Venus: 583.92)Four centuries of articles reproduce a planet's period to within 0.04%.Comet and guest-star articles1,417articles · precision 0.68Multiply comet article counts by the 68% precision; one in three is a memorial or debate citing a comet.Earthquake articles1,768articles · precision 0.76Of earthquake articles, 76% are actual reports, so they are usable as a source for historical seismicity catalogs.Summer frost, snow and hail days×1.3frost+snow rate 1640–1719 vs 1500–1639Tested against the Sillok, the Little Ice Age hypothesis is supported for 1640–1719 rather than across 1500–1750.

Politics and appointments

Clan concentration among first appointees to high office0.28→0.45top-10 clan share, 1392–1449 → 1800–1849Puts a number on the thesis of oligarchic concentration: the top-10 clan share of high office rose from the 17th century to the early 19th, and the rise holds even under a pessimistic lower bound.Share of royal-lecture articles8.0%of articles have a royal-lecture keyword · precision 0.82Safe to read as the share of articles in which a lecture was actually held (82% precision).Share of memorial articles6.8%of articles have a memorial keywordShows how keyword hits and NIKH classification capture different things.Share of articles classified under justice23.5%of principal-edition articles classified under justiceThe share of justice articles tracks the intensity of political trials, not routine litigation.

Justice and society

Articles mentioning the death penalty11.8mentions per 1,000 in 1400–1449, the peakA sample-based warning against quoting death-penalty articles as executions: the Sejong peak is sentencing paperwork.Share of famine-relief articles1.2%of articles have a relief or famine keyword · precision 0.81With 81% precision, relief and famine articles point to real crises and relief measures; the peaks in the 1670–71 and 1695–96 famines make the series a first draft of a famine chronology.Articles mentioning the vernacular script520articles mentioning the script (principal editions)How often, and where, the vernacular script enters the state record (including overlap with trial records).Share of articles on slaves1.8%of articles have a slaves keyword · precision 0.68Re-coded in 1.2 with a narrower codebook, the slaves keyword has 68% precision (blind re-coding κ 0.86).

Record and sky — three sources

Astronomical articles of the Goryeosa Treatise on Astronomy (918–1392), the Goryeosa-jeolyo and the Seungjeongwon Ilgi (1623–1910), classified by type with the same codebook as the Sillok and compared with the same DE406 computation. Observing sites: Gaeseong for Goryeo, Seoul for Joseon. The eclipse figures here come from a v2 re-coding of the 1.0 series C (Sillok only, codebook v1), so they differ from C1; the Versions section of the Method page records what changed.

Solar-eclipse observation reports matched by computation, by source82.6%of 115 Goryeosa reports match an eclipse visible from Gaeseong · Jeolyo 81% · Sillok 73% · Ilgi 91%Across the three sources, seven to nine in ten eclipse observation reports match a computed eclipse visible at the observing site.Lunar-eclipse articles: types and match rate, by source71.3%Sillok reports matched · Goryeosa treatise 78% · Sillok 71% · Ilgi 79%Lunar-eclipse observation reports match at 71–79% in all three sources, a little below solar.Planet articles against the computed sky — Venus elongation and 犯 separations43°median Venus elongation, Goryeosa (random: 30°)On 'Venus seen by day' dates the planet's elongation has a median near 43° in all three sources against 31° on random dates, and for the '月犯X / X犯Y' pairs (Goryeosa 454 pairs: 285 Moon, 169 planet) the minimum separation that night has a median of 2.8° (2.2° with a ±1-day window) against 91° at random.Identification rate of comet and guest-star articles78%of 69 Sillok reports within ±90 d of a perihelion (chance 17%)Matching comet and guest-star report dates to known perihelia within ±90 days captures 78% of Sillok observation reports and 33% of Goryeosa articles, against 17% and 3% for the same number of random dates.Chronology of naked-eye sunspot articles170articles · Goryeosa, Jeolyo, Sillok, IlgiNaked-eye sunspot articles total 170 across the four sources (Goryeosa 32, Jeolyo 18, Sillok 50, Ilgi 70), at most a few dozen per 50-year period, so they are given as a chronology only.Same-event reporting between sources32%of 37 Sillok eclipse days also in the Ilgi, 1623–1863Of Goryeosa eclipse observation days, 91% recur in the Jeolyo, but only 32% of Sillok eclipse observation days of 1623–1863 recur in the Ilgi, while 91% of comet keyword days do.

The weather cell of the Seungjeongwon Ilgi

The 102,397 day-level weather cells of the Seungjeongwon Ilgi, decomposed with a token rule. The cell was first digitized in full by Lyu et al. (2024, Geoscience Data Journal, Zenodo CC BY 4.0) for 1623–1910, and Lee, Shin & Hong (2024, Scientific Data, Zenodo CC BY 4.0) published its categories for 1778–1907 together with the rain-gauge series. This chapter adds the day-by-day agreement with those two datasets and the list of differences, the degree to which a bare '晴' was entered as routine, the overlap with the rain gauge and with modern observations, and the structure of the gaps.

Token shares of the weather cell and agreement with two datasets75.1%of weather days carry only '晴'Three quarters of the days carry nothing but '晴' (76,308 with the bare character, 181 with a spelling variant such as 〈晴〉 or 朝晴).Routine-entry index — the bare 晴 and the rain gauge73%of 8,320 gauge rain days carry 雨 · 雨 on 1 of 36,138 days without a gauge entryOnly 73% of days with measured rain carry 雨 in the cell — and the weather string transcribed by Lee et al. 2024 gives the same 72.7%, so this is a property of the cell.Overlap with modern observations, 1907–191021%of 403 observed rain days carry 雨, 1907–1910On the 1,052 days overlapping KMA observations (1907–1910) only 21% of observed rain days carry 雨, and only 24% of '晴'-only days have mean cloud cover ≤ 2.5 tenths.Gaps — missing days, empty cells, note-only cells2,275days without a node · 2,256 empty-cell nodes · 1,317 duplicate nodesOf 104,963 calendar days, 2,275 have no day node; 2,256 day nodes have an empty weather cell (2,185 distinct dated days); 1,317 duplicate nodes (two nodes on one date) and 69 undated nodes are counted separately.

All reigns

Articles per reign (principal editions); see the Reigns menu for each reign's page.