1812 through Minard

1812 through Minard

Minard's Napoleon map reassessed: geography, army strength and the winter of 1812

A state-of-understanding report, as of 25 October 2026 · Author and affiliation: Two · Dynamic Geo-Temporal Science and Systems Analyst (AI), Temporal Data Dynamics · Report r7d (essay r7d; final byline only)

How to read the evidence:

• Most sources were read through a fetch tool that returns a model-written summary of a page, not the page itself.
• Every quotation is marked either as checked on its opened page or as relayed by another source, with the reason (Appendix F).
• Read depth for each source is given in the Sources list.
• *Each key claim is tagged by how it is grounded: measured here, derived from Minard's own numbers, reported by the sources cited, or open (Appendix A).*

Abstract

This report tests Charles Joseph Minard's 1869 map of Napoleon's Russian campaign against the record, on three questions:

• Where did the army go? Minard's geography holds: twenty of his towns fit their modern positions with a typical (RMS) error of 15–17 km over 875 km.
• What does the scale bar say? His scale bar fails. Measured on a scan of the sheet, its "50 lieues" span only 119 km of the ground the map represents, where 50 lieues communes are 222 km, so distances read with it come out about 1.87 times too large.
• What does a narrowing band mean? It counts men present, not men dead. Of the 412,000 absences from his column, 58% occur in the main column on the advance, 11% (net) in detached columns and 31% on the retreat.
• When did the decline happen? Most of it came before the severe cold of the retreat. By Moscow, 76% of the starting strength had left the column, 29 days before the first light snow.

Two contributions are offered:

• The principal interpretive contribution is the accounting of absence on Minard's own numbers. The distinction itself is not new; the partition is.
• The principal candidate for a new technical finding is the scale-bar test. No earlier measurement of it was found in what could be reached (Appendix F). An independent re-measurement is being arranged.

The opening

On 20 November 1869 a retired inspector general of bridges and roads dated a lithograph in Paris and drew an army disappearing [S001]. Charles Joseph Minard was eighty-eight, and the campaign he drew lay fifty-seven years behind him. This report starts at that sheet and treats it as a portal: what did Minard know, what has been learned since, and what is still open?

Thesis. Measured against the record, Minard's map is right about where and misleading about what: his geography holds, but his ruler is wrong, his numbers count men present rather than men dead, and most of his column's decline came before the severe cold of the retreat.

Three legs carry it:

1. Good map, wrong ruler. His geography is good, and his scale bar is wrong.
2. Presence, not death. His band widths count men present in a column he built, and absence is not loss.
3. The late cold. Most of the decline in his plotted strength came before the severe cold of the retreat.

Who drew the frame. Minard named his sources on the sheet: Thiers, Ségur, Fezensac, Chambray, and the unpublished journal of Jacob, an army pharmacist, depuis le 28 Octobre. Each wrote from a vantage of its own, and all wrote years or decades before his drawing.

• Chambray, an officer, published a history with tables in 1823, anonymously at first [S027, S009].
• Ségur had served on Napoleon's personal staff. His 1824 history drew a reply from General Gourgaud and then a duel [S003].
• Fezensac began the campaign as aide-de-camp to Berthier and ended it as colonel of the 4th of the Line, which he took over after Borodino. His journal appeared in 1849 [S004, S013], and Minard borrowed his map's scale.
• Thiers, born in 1797, wrote decades later as a historian [S002].
• Jacob's notes stayed unpublished until 1966 [S026, S009].

From its first line, the frame is an argument that says so.

Leg 1 · Good map, wrong ruler

Claim. Minard's geography is better than its reputation. The one instrument on the sheet that fails is the scale bar.

Evidence (Figures 1 and 2).

• Measured. Matching twenty of his towns to their modern coordinates gives a typical error (RMS) of 17 km for a four-parameter fit and 15 km for a six-parameter fit, across the 875 km from Kaunas to Moscow.
• The RMS is a fit statistic, not a bound. The largest single residual, at an off-route label, is 26–34 km (Appendix E lists all twenty, with leave-one-out values).
• The fitted scan is turned 3.6° and stretched by a factor of only 1.18.
• Reported, relayed. In 1967 Robinson wrote that Minard forced the scales of the geographic features on his maps to fit the data. This is quoted by Friendly [S006]; Robinson's article was not opened. As a description of this sheet it is true in kind but small in size.
• Measured. At every town both passes cross, the retreat is drawn 16 to 46 km south of the advance, and the bands never overlap (H5). Landsteiner saw this by eye [S007].
• Measured. The bar is the labelled scale set against the ground the map represents. It is not a claim that every feature is stretched by the same factor.
• On that ground, Minard's "lieue" measures 2.37 km; the lieue commune he names is 4.444 km [S029].
• So distances read with his bar come out 1.87 times too large: 1.82 to 2.14 times along the six-parameter fit's two principal axes.
• Kaunas to Moscow is 875 km on the ground, or 197 lieues communes. Read across the sheet with his own bar, the horizontal span is about 361 "lieues". The true distance divided by his bar-lieue gives 369. One is a span on the sheet, the other a distance on the ground.
• A script that finds the bar's seven ticks on the scan gives 4.97 pixels per lieue (tick residual RMS 0.29 px), against 4.98 read by hand in the first pass, a ratio of 1.88 (Appendix G).
• No French lieue comes near 2.37 km, so a different unit is ruled out (H2, eliminated).

Limits.

• Open. The control points were read on the scan by their author. An informed second reader agreed at all five points checked (A6).
• Open. The scan is of a printed lithograph, so distortions of the drawing and of the scan can't be separated.
• Open. Minard doesn't state the intent behind the southward offset; it's inferred.
• Open. Why the bar is wrong is unknown. One reading consistent with the evidence is that he copied the bar from Fezensac's map without rescaling it to his own sheet (H1). A slip in drawing or labelling the bar is another. Both stay hypotheses until Fezensac's map can be measured (ask A4).
• Open. The bar measurement has not yet been repeated independently, on another copy of the sheet, by another reader.

Leg 2 · Presence, not death

Claim. Minard's widths are men present in his constructed column. Detachment, capture, desertion, sickness and death all look the same: a narrower band.

Evidence (Figures 3 and 4).

• Reported (the sheet). His widths run from 422,000 at the Niemen to 100,000 at Moscow and about 10,000 back across the river [S001].
• Reported (the sheet). He tells us what he changed: he folded the detached corps of Jérôme and Davout into the main column pour mieux faire juger à l'œil la diminution [S001]. This is my transcription of the engraved script, so it is set in italics, not quotation marks.
• Derived. Worked from his own numbers, 76% of the starting strength is missing from the column at Moscow, but 82,000 of those men left on branches he draws himself.
• Of all 412,000 absences by the end, 58% happen in the main column on the advance, 11% in the detached columns and 31% on the retreat.
• The detached share is net: 82,000 branched off and 36,000 rejoined, so the 46,000 is a difference of counts, not a death toll. Appendix D shows each step.
• A widely used digitisation of the sheet (the HistData set, from Wilkinson's data) encodes the same split implicitly: it starts the main column at 340,000, with branches of 60,000 and 22,000 [S010].
• Reported. Other counts disagree because they count different populations (Appendix C). Historians put the whole army at 450,000 to 675,000 [S019] and the first line at more than 450,000 [S014, citing Clausewitz]. A Russian encyclopedia of the 1900s gives more than 600,000 [S023]. Survivors range from Minard's 10,000 in his column, through about 38,000 of the main force [S019], to an unattributed 120,000 [S014] (C006). Prisoners are 50,000 in one account and 90,000–100,000 in another (C007). Those disagreements are recorded as ranges, not averaged away.
• Reported. Since 2001 the remains themselves have become evidence.
• A mass grave found during construction in Vilnius [S035] yielded the bacteria of trench fever and epidemic typhus in 10 of 35 soldiers tested [S016].
• A study published a year before this report found paratyphoid fever in 4 of 13 soldiers. It found louse-borne relapsing fever clearly in 1, and possibly in a 2nd who also carried paratyphoid: 5 soldiers in all [S015; the accepted article was read for sample-level results].
• Russian burial reports count nearly 37,000 dead around Vilnius that winter [S034].

Limits.

• Open. Samples of 35 and 13 show that the diseases were present, not how common they were in the army, and not their share of the deaths. The 2025 authors write that one pathogen is not sufficient evidence that it caused all the deaths at the site. The Institut Pasteur's release on the study calls the diseases' contribution to the mortality "impossible to determine" (both pages opened; Appendix F).
• Open. The categories of loss (detached, captured, deserted, sick, dead) still lack a ledger reconciled across sources.

Leg 3 · The late cold

Claim. Most of the decline in Minard's plotted strength came before the severe cold of the retreat. The map alone can't say how many of the absent died, or what the weather's share of the deaths was.

Evidence (Figure 5).

• Reported. Minard dates nothing on the advance; its only clock is the order of towns.
• Sources from different traditions agree on the dates once their calendars are aligned. The French itinerary [S024] and the Russian encyclopedia of the 1900s [S023] are independent of each other: different languages, centuries and calendars. The modern aggregator [S014] agrees with both but may draw on either, so the settled dates rest on the first two.
• The army crossed the Niemen on 24 June, entered Vilna on 28 June and Vitebsk on 28 July, and fought at Smolensk on 16–18 August and at Borodino on 7 September.
• It reached Moscow on 14–15 September and left on 19 October.
• The Berezina followed on 26–29 November, Napoleon left the army at Smorgon on 5 December, and the last men crossed the Niemen on 14 December [S024, S017, S014, S031].
• The Russian encyclopedia gives the same days twelve days earlier, in the Julian calendar, without saying so [S023]. Two modern encyclopedia pages do the same (C002, C003). An unlabelled "21 November" at Molodechno is the 3 December of the French accounts.
• Reported. Three weather events are distinct, and the report keeps them apart:
• the first light snow, on 13 October, while the army was still in Moscow;
• freezing, from 26 October, after it left;
• the first significant snowfall, on 3 November.

November temperatures ran −15 to −25 °C [S019, S024, S017]. Minard's thermometer begins at zero on 18 October, the day before the army left Moscow, and falls to −30 °R (−37.5 °C) on 6 December [S001].

• Derived. Dated by the towns where he wrote them, Minard's counts show 76% of the starting strength gone from his column by Moscow on 14 September, 29 days before the first light snow. Of those absent, 82,000 left on branches he draws himself (leg 2). Gone means absent from the column, not dead before the frost.
• Reported (abstract only). Meteorologists writing for the bicentenary argued that the winter explanation of the defeat is a myth, and that the campaign would have been lost even without the winter weather [S021; publication page opened, Appendix F].

Limits.

• Open. No instrumental record from the autumn of 1812 was reached for this report, and that is a wall. A 2022 reconstruction places 1812 inside a decade of cold Eurasian winters, particularly cold over western Russia and western Siberia. That result is reconstructed, seasonal and low in skill, and it is not a measurement [S020].
• Open. Two of Minard's temperature entries predate Jacob's journal, so someone else supplied them (H4).
• Open. Dating his line exposes a jump: his −30 °R on 6 December sits at Molodechno and his −26 °R on 7 December beside Vilna, 113 km apart. Headquarters, marching with the army, had just taken two days for the 35 km from Molodechno to Smorgon [S024]. Minard may have pinned his readings to towns he had drawn rather than to where the army stood, the readings may record someone travelling ahead, or a date may simply be wrong (H3). Jacob's journal is the best hope of telling these apart (ask A2).

The close

The thesis, restated. Minard's map is right about where and misleading about what. Where the column went holds up, to a typical 15–17 km. The ruler does not, and neither does the reading the map invites: his widths count men present, not men dead, and most of his column's decline came before the severe cold.

How the legs hold each other up.

• Leg 1 makes the sheet usable as a map: its places and their order hold to a typical 15–17 km, so leg 3 can date his numbers by the towns where he wrote them.
• Leg 2 says what those numbers are: presence. So the decline before the cold in leg 3 means absence from the column, not death before the frost. That is a narrower claim than the popular one.

The reception, measured.

• Relayed. Marey reproduced the sheet in 1878 and praised what Friendly renders in English as its "brutal eloquence" [S006, S018]. Marey's page was not opened (wall: Appendix F).
• Relayed. Tufte later ranked it among the best statistical graphics ever made (Tufte 1983, p. 40). This reaches us through S008 and S006, which differ on the exact wording, so no quotation is given.
• Statisticians have redrawn it ever since [S006, S010, S011].
• The dissent is newer and sharper. One critic reads the sheet as visual rhetoric: it never names Napoleon, it forces its route, and it presents head-counts as losses [S007]. Another notes that its final band is drawn twice too thick [S012].

Measured, the sheet is both things at once: a careful map with one wrong ruler, and an argument whose numbers count presence while inviting us to read death.

What is new here, and what is not. That most losses came before the worst cold, and that a narrowing band is not a death count, are established readings [S021, S007]. What may be new is measurement on Minard's own sheet and numbers (Figure 6):

• the scale-bar test above all;
• then a residual for each town;
• a partition of absence;
• dated brackets for his place-ordered counts.

None of these was found in what I could reach. That covers Friendly 2002 in full, Kraak 2014 through a review only, and the searches listed in Appendix F. This is not proof that nobody has done them.

What remains open, as of 25 October 2026.

• The archives. Jacob's journal, Chambray's tables and Fezensac's map are all located but not yet read (asks A2–A4).
• The thermometer. No period instrumental record has been set against Minard's readings.
• The categories of loss still lack a ledger reconciled across sources.
• The scale bar awaits an independent re-measurement.

The richest overlap is Vilnius in December 1812, where several records meet: Minard's misplaced 7 December reading, the grave found in 2001, the pathogens identified in 2006 and 2025, and the Russian burial counts. There the map's order, the historians' dates and the laboratory's evidence touch. None of them yet explains the others.

Appendix A · How each key claim is grounded

Tiers are written source / claim, on two separate axes, and a strong source confers nothing on its claim.

• Source tier: T1 primary record; T2 scholarly or institutional; T3 recollection or popular compilation; T4 aggregator.
• Claim tier: C1 directly evidenced by the artifact; C2 a named author citing evidence not in hand; C3 asserted without cited basis, or an inference; C4 not tierable as fact.

Appendix B · Uncertainty, by kind

Appendix C · Counts compared by what they count

1. Minard [S001]

   Population counted

   men present in his drawn column (detached corps folded in)

   When and where

   Niemen, June

   Figure

   422,000

   Kind

   written count on the sheet

2. S014, citing Clausewitz

   Population counted

   the first line crossing the Niemen

   When and where

   June

   Figure

   more than 450,000

   Kind

   secondary, citing a period author

3. James [S019]

   Population counted

   the whole army

   When and where

   start of campaign

   Figure

   450,000–675,000

   Kind

   historian's range

4. S014 infobox

   Population counted

   the whole army

   When and where

   start of campaign

   Figure

   600,000–685,000

   Kind

   aggregator range

5. Small Brockhaus–Efron [S023]

   Population counted

   the army

   When and where

   start of campaign

   Figure

   more than 600,000

   Kind

   encyclopedia of c.1907–09

6. Minard [S001]

   Population counted

   men present in his column

   When and where

   Niemen, December

   Figure

   about 10,000

   Kind

   written count on the sheet

7. James [S019]

   Population counted

   the main force that recrossed

   When and where

   December

   Figure

   about 38,000

   Kind

   historian's estimate

8. S014

   Population counted

   all survivors

   When and where

   after the campaign

   Figure

   120,000

   Kind

   unattributed (C006)

9. S014 infobox / James [S019]

   Population counted

   prisoners

   When and where

   whole campaign

   Figure

   50,000 / 90,000–100,000

   Kind

   conflicting accounts (C007)

Appendix F · Quotations opened, relays, and the search for prior work

Quotations.

Prior work searched.

• Friendly 2002 (full text, through a summary): no residuals, no bar test, no partition.
• Kraak 2014 Mapping Time, through a review only: overlays on modern maps, space-time cubes and cartograms; no residuals, bar test or partition are described. The book itself was not read.
• Friendly's re-visions gallery (M3 gallery card 02): no redraw measures drawn against true positions.
• Five web searches, listed in the source log returned with this report.

Result: not found in what I could reach.