What Adam Is Reading
Two Minutes to Lille
Claude Chappe built a network that carried a signal across France in minutes and a sentence across it in half an hour. Everything that happened next in telecommunications is an argument about which of those two numbers you care about.
Topic explainer · 9 sources · 19 September 2026

On 30 August 1794 the French army retook Condé-sur-l'Escaut, a fortress town on the Belgian border about 200 kilometers from Paris. The news reached the National Convention the same day, in time for the deputies to hear it read aloud and vote the town a new name before the session closed. No rider had come. The message had been waved to Paris by a chain of men standing in towers, each one squinting through a telescope at the tower behind him and copying what he saw.

That is the moment worth pausing on. Not the machine, which was a clever arrangement of pivoting beams, and not the war, which produced a great many other things. The moment is the one where a piece of information crossed a landscape faster than a human body could cross it, for the first time since the Romans lit hilltop fires, and this time with enough resolution to carry a sentence rather than a prearranged yes or no.

Claude Chappe's telegraph ran for sixty years and then vanished so thoroughly that most people who use the word "telegraph" picture wires. It deserves better, partly because its engineers were working out, with wood and rope and eyeglasses, the two quantities that still govern every network anyone builds.


The machine, briefly

Chappe and his brothers tried several designs. The first, demonstrated between Brûlon and Parcé in March 1791, used synchronized pendulum clocks and a rotating panel, which meant both ends had to agree on time before they could agree on anything else. He threw that away. The design France adopted needed no shared clock at all.

A tower carried a pivoting beam 4.6 meters long called the regulator, with a smaller two meter arm, an indicator, hinged at each end. The indicators could each take seven positions at 45 degree steps. The regulator could physically swing to four, but only two of them, horizontal and vertical, counted as part of the code. The two oblique positions were reserved as a transition state, a way of telling the next tower "I am mid-composition, do not copy me yet." Two times seven times seven gives 98 combinations. Six were held back for service signals. That left 92.

Ninety two is a small alphabet. It is not a small vocabulary, because Chappe did not spell. Each tower crew read a pair of signals and looked it up in a code book of 92 pages with 92 entries per page. The first signal gave the page, the second gave the line. Two arm positions therefore selected one of 8,464 words and stock phrases, and later supplementary volumes added geography and proper nouns on top of that.

The crews themselves could not read the traffic. Operators were issued the apparatus and the timings, not the code book, which lived only at the terminal stations. They worked 365 days a year for the wages of a construction laborer, and were fined for errors. Telescopes of 30x to 65x sat in fixed wooden mounts, aimed permanently at the neighbor. The system ran in daylight only. Lanterns on the arms were tried for night work and never became standard, and fog closed the line outright.

THE APPARATUS THE SIGNAL SPACE regulator, 4.6 m indicator, 2 m indicator, 2 m two operators, ropes and pulleys inside the tower 2 positions carry code 2 are transition only each indicator takes 7 of 8 positions the 8th folds along the beam 2 × 7 × 7 = 98. Six held for service. 92 in use.
How ninety two signals come out of three pieces of wood. The regulator could physically swing to four positions, but only horizontal and vertical were read as code. The two obliques meant "I am mid composition, do not copy me yet." Diagram drawn for this piece from the geometry described in the sources below. Free to reuse.

The part that was actually new. The arms get the attention, and the arms are the least interesting component. The code book is the invention. It is a source coder, a lookup table that turns a thirteen bit address into a whole military phrase. A dispatch reading "the enemy has abandoned the position" costs two signals instead of forty. On a channel this slow, the compression ratio is not a nicety. It is the difference between a message arriving today and arriving tomorrow.


Two numbers, not one

Here is where the Chappe line stops being a curiosity and starts being an engineering document.

The Paris to Lille line opened in 1794 over roughly 230 kilometers. Station counts in the literature run from fifteen to twenty two depending on the year and on what you are willing to call a station. Hops were 10 to 15 kilometers. A single signal, launched at Paris, rippled down the chain and arrived at Lille in something between two and nine minutes, and the sources genuinely disagree about which. A complete dispatch of 36 signals took about 32 minutes.

Those two figures measure different things, and the gap between them is the whole subject.

The first is latency. It is set by the number of hops and by how long each crew took to observe, verify and reproduce a position, call it twenty to thirty seconds under good conditions. Latency on a Chappe line scales linearly with distance, because distance is hops and hops are people. Paris to Toulon, roughly 760 kilometers and well over a hundred stations, took a signal something like twelve to twenty minutes.

The second is throughput. Once the pipeline is full, a new signal can enter at Paris every twenty or thirty seconds regardless of how far away Toulon is. Sustained rates ran two to three signals per minute. At two signals per code book word, that is roughly one word per minute at the optimistic end. Wikipedia's optical telegraph entry puts the practical figure lower, at 0.4 to 0.6 words per minute, once acknowledgments and service traffic are counted. Either way the channel carried on the order of 0.3 bits per second.

A store and forward pipeline whose latency grows with hop count and whose throughput does not. Every network engineer reading this has drawn that diagram on a whiteboard. Chappe built it out of oak in 1793.

A handwritten Chappe telegraph dispatch dated 24 August 1838, headed Ligne de Bayonne, Direction de Toulouse, announcing the birth of a prince, with an archive stamp at lower left.
A real one. Paris to Toulouse down the Bayonne line, 24 August 1838. The Minister of the Interior tells every prefect that the Duchesse d'Orléans has been delivered of a prince at three in the afternoon, that the King has named him Louis Philippe Albert, Comte de Paris, and that they are to pass it to every commune at once. Filed at Paris at half past three. Copied out in Toulouse, roughly 700 kilometers away, at six that evening. Two and a half hours for a message of that length, which is the throughput number rather than the latency one. Administration télégraphique, Toulouse, 1838. Archives départementales de la Haute-Garonne, cote 4M54. Public domain. Wikimedia Commons.
SystemRouteHopsFirst symbol end to endThroughput
Chappe optical, 1794Paris to Lille, 230 km~152 to 9 min~0.3 bit/s
Chappe optical, 1800sParis to Toulon, 760 km100+12 to 20 min~0.3 bit/s
Morse electric, 1850sany wired distance1milliseconds25 to 40 wpm
Rady et al. balloons, 2022Torfou to Séris, 103 km10sub-second50 kbit/s per link
Commercial fiber, todayNew York to London, ~6,600 kmmany~30 ms one wayterabits

What the network became, and what broke it

Strasbourg came on line in 1798, Brest in 1799. Napoleon understood immediately what a general could do with it and the map filled in. By the mid 1840s France had somewhere between 534 and 556 stations covering close to 5,000 kilometers and linking 29 cities, with branches reaching Amsterdam, Mainz and Venice, and later lines across North Africa between Algiers and Oran and Algiers and Constantine. Mohamed Aly Pasha imported the design for Egypt and ran Alexandria to Cairo, 17 stations over about 180 kilometers, in 40 minutes. A law in 1837 made telegraphy a state monopoly and banned private lines, which tells you how valuable the thing had become.

Engraved portrait of Claude Chappe, shown from the chest up in a high collared coat and cravat, with shoulder length hair.
Claude Chappe. He was 41 when he died in 1805. The network he left behind ran for another fifty years. Engraving by Henri Rousseau, Album du Centenaire. Public domain. Wikimedia Commons (the file name misdates him; he lived 1763 to 1805).

Samuel Morse's single wire scheme arrived in 1837 and France began converting in the mid 1840s. The usual explanation is that electricity was faster. That undersells it.

The electric telegraph did not improve Chappe's latency. It deleted the term. Propagation down a copper wire is so fast relative to human timescales that end to end delay stops depending on distance at all, and the only clock that matters is how quickly the operator can key. Paris to Toulon went from fifteen minutes to about four milliseconds, a factor of roughly two hundred thousand, and the same wire would have done Paris to Moscow for free. Throughput improved too, by perhaps thirty or fifty times, which is respectable and is not the revolution. The revolution is that distance stopped costing time.

Everything else was gravy, and there was a lot of gravy. Wire works at night. Wire works in fog. Wire does not need two men in a tower every twelve kilometers, every day of the year, and the payroll of 534 stations was the dominant operating cost of the French system. The last Chappe dispatch is usually dated 1855, carrying news of the fall of Sevastopol, though most of the network had already gone dark a year or two before that.

Claude Chappe did not see any of it. He took his own life in January 1805, at 41, exhausted by years of rivals claiming his design as theirs.


The first attack on a data network

Between 1834 and 1836 two bankers in Bordeaux, the brothers François and Joseph Blanc, bribed the operators at a station near Tours to insert a particular deliberate error into passing government traffic, followed by the backspace signal that told downstream towers to discard it. The error was stripped before the message reached Paris or Bordeaux, so the official dispatch arrived clean and nobody at either endpoint saw anything. But an accomplice watching the Bordeaux tower with a telescope could read the discarded symbol, and that symbol encoded which way the bond market had moved in Paris.

The mail took days. The Blancs had the direction of the market in minutes, and they ran the scheme for two years before anyone noticed.

Read that again as a security engineer. They did not break the code, which they could not read anyway. They hid a payload inside the error correction layer, where the protocol guaranteed it would be invisible to the endpoints and visible to anyone with line of sight to a tower. It is a covert channel in a system that had not yet been given the word "channel." They were prosecuted and they were not convicted, because no French law contemplated the misuse of a telegraph. The statute came later, as statutes do.


Eleven balloons over the same hills

In 2022 a group from Inria, Orange Labs and INSA Lyon did something I find completely delightful. They took the historical map of Chappe tower sites running south from Paris toward Orléans, drove to eleven of the villages on it, and at each one inflated a helium balloon, hung a battery powered radio under it, and let out up to 80 meters of rope.

The villages were Torfou, Etampes, Angerville, Arbouville, Toury, Artenay, Chevilly, Bucy, Baccon, Cravant and Séris. Ten hops, the longest 12.67 kilometers, total path 103.13 kilometers. A man sat at Torfou with a laptop and typed messages that came out of a laptop at Séris. Deployment took six hours, in clear weather at 14 degrees with gusts to 18 km/h. Nothing failed, no balloon and no node.

The engineering is more interesting than the stunt. The constraint that set the balloon height was Fresnel zone clearance, the ellipsoid around a radio path that has to stay unobstructed. At 868 MHz over a 12 kilometer hop the required radius works out to about 33 meters, which is why the balloons had to float at 50 meters or better and why the team cared so much about terrain elevation between villages. Chappe's surveyors were solving the same problem with the same hills and a simpler rule, which was whether you could see the next tower.

The relay protocol is Chappe's protocol with the humans removed. The root transmitted once a second. Each node listened for the packet addressed to it, incremented the source and destination fields, and retransmitted three times at 20 millisecond spacing. Incrementing the address is what stops a packet from travelling backward, which is the same job done in 1794 by the rule that you only ever copy the tower on one particular side of you.

Two corrections to the paper, since we are being careful.

First, the radio. The abstract frames the work in terms of LoRa and long range IoT, and a casual reader comes away thinking this was a LoRa network. It was not. The experiment ran IEEE 802.15.4g FSK Option 1 at 868 MHz, 50 kbps with forward error correction, on OpenMote B hardware using the AT86RF215 radio, plus some proprietary Falco motes. The authors are explicit about this in the body and they note that a comparable LoRa SF-7 test in prior work delivered nothing past 10 kilometers. The framing is looser than the work.

Second, the history. The paper's own historical section says the Chappe apparatus had four regulator positions and eight per indicator for 256 symbols, and that Paris to Lille was 190 kilometers with 18 towers. Both conflict with the standard account of 98 combinations, 92 in service, over roughly 230 kilometers. The radio engineering is sound. The history was written from a secondary source that nobody checked.


The problem did not go away

It is tempting to file all of this as a charming prelude. It is not. The quantity Chappe was fighting reasserted itself the moment networks got big enough.

Light in glass moves at about two thirds of its vacuum speed, so a fiber from New York to London imposes a floor of roughly 30 milliseconds each way that no amount of money can remove. Plenty of money has tried. The Hibernia Express cable cost several hundred million dollars and justified itself on a round trip of just under 60 milliseconds, a handful of milliseconds better than the route it beat. Trading firms have since gone to microwave over land and hollow core fiber undersea for the same reason. Distance costs time again, exactly as it did in 1794, five decimal places further down.

Hop count came back too. Every mesh network, every LEO satellite constellation with inter-satellite links, every blockchain gossip protocol has the Chappe structure, which is that the pipeline fills at one rate and drains at another and the two are not the same number. The 2022 balloon experiment measures a ten hop store and forward chain over 103 kilometers because that is what a low power radio mesh actually is. So did Chappe, over the same ground, with the same hop count, at roughly a thousandth the speed and a hundred thousandth the capacity.

What changed underneath all of it is less flattering to us than it looks. Chappe's channel carried 0.3 bits per second and he spent his whole design budget on a code book that made every one of those bits carry a full military phrase. We carry terabits and spend the budget on autoplaying video. The bandwidth went up by fourteen orders of magnitude. The compression ratio went down.

So What

Chappe did not invent fast communication. He invented the measurement of it. Once you have a network whose delay you can count in hops and whose capacity you can count in words per minute, the two numbers separate, and every argument since about wires, fiber, mesh and orbit has been an argument about which of them you are willing to pay for.

Confidence: the mechanism and the 2022 experiment are solid and primary sourced. Specific transmission times for named routes vary across the historical literature by a factor of three or more, and are given here as ranges on purpose. The Blanc brothers' method is well attested; the precise station and the precise signal they used are reported slightly differently in different tellings.

Sources

Replication experiment: Rady M, Muñoz J, Abu-Aisheh R, Vučinić M, Astorga Tobar J, Cortes A, Lampin Q, Barthel D, Watteyne T. "A Historical Twist on Long-Range Wireless: Building a 103 km Multi-Hop Network Replicating Claude Chappe's Telegraph." Sensors 2022;22(19):7586. doi.org/10.3390/s22197586

Apparatus, code book, network extent, Blanc fraud: Chappe telegraph. en.wikipedia.org/wiki/Chappe_telegraph

Throughput and route timings: Optical telegraph. en.wikipedia.org/wiki/Optical_telegraph

Signal geometry, per station timings, transition to electric: Dilhac J-M. "The telegraph of Claude Chappe, an optical telecommunication network for the XVIIIth century." IEEE History Center. ethw.org

Chappe brothers, operator conditions, 1837 monopoly law: Selin S. "Napoleonic Telecommunications: The Chappe Semaphore Telegraph." shannonselin.com

Speed comparison with horse messengers, code book size: History of Information, entry 429. historyofinformation.com

Narrative background: "Claude Chappe and the Napoleon Telegraph," The History Guy. youtube.com

Images: the two photographs are public domain and are reproduced from Wikimedia Commons with the credits given in their captions. The 1838 dispatch is File:Depeche telchappe.jpg (Archives départementales de la Haute-Garonne, cote 4M54). The portrait is File:AduC 175 Chappe (engraving by Henri Rousseau, Album du Centenaire). Neither requires attribution under its licence; it is given anyway. The apparatus diagram is original to this piece.

Verification: Empirical claims were submitted with authorship stripped to GPT-5.2 and Grok 4.6 for independent error and omission sweeps. Both flagged the station count and single signal timing for Paris to Lille as disputed in the literature, which is why those appear as ranges.