Sun Yat-sen’s circle on the wire, 1916–17

01 The files

The series is 各国内政関係雑纂／支那ノ部／革命党関係（亡命者ヲ含ム）／特定人ノ発受郵便物取調, Gaimushō 1.6.1.4-2-1-2, volumes 2 to 4, on JACAR as B03050088300, B03050088400, B03050089100, B03050090000, B03050090100 and B03050090200. Each batch comes under a covering note headed 特定人發着郵便物等取調ノ件 (“inquiry into mail etc. sent and received by specified persons”), from the Director of the Communications Bureau (逓信省通信局長 田中) to the Director of the Foreign Ministry’s Political Affairs Bureau (政務局長 小池, Koike Chōzō). A typical note, 秘第八?之一號 of 25 May 1916, received the next day, reads 別紙寫電報及左記郵便物有之候: “attached are copied telegrams and the postal item noted below”.

The telegrams are copied onto the Imperial Telegraphs’ own forms. Outgoing messages from Tokyo are on “Forwarded” forms, signed 中山 (Nakayama, Sun’s Japanese alias) or, in May, with the handler’s name 林蔚陸 (“Linwailuk”). Incoming messages are on “Received” forms addressed to “Rosamonde Tokio” (Soong Ching-ling), to “Linwailuk Tokio”, or to Nakayama at 109 Harajuku, Aoyama. The correspondents are Yamada Junzaburō in Shanghai (“Junzaburo Shanghai”), “Konglipo” in Manila, Kayano Nagatomo and Ju Zheng’s men at Qingdao (Seitō), “Youngchina” at San Francisco, Honolulu, Hankow, and the Shanghai leadership: Chen Qimei, Xu Chongzhi and the rest. Some of the traffic is in clear English (“2000 through Shokinginko”), and some is in plain telegraph code. Most of it is in ten-letter condenser words.

The files hold no decodes. The Foreign Ministry did try. In B03050089100, frames 0076–0078 are a notebook copy of a long romanised kana stream (126 ten-letter groups, ending owari), and frames 0079–0081 are the analysts’ worksheets on it. They copy the stream into katakana, divide it with slashes into groups of two or three, and pencil kanji guesses beside the groups, often with a question mark: 事, 先生, 國, 電, 軍, 謀. No continuous sense comes out of them. The stream is not a condenser telegram: it is two telegrams of 20 May 1916 to Huang Xing in the three-kana code, which Satoshi Tomokiyo’s formula (September 2026) decodes. Harumi-Matsuba read both in October 2026, with many spelling corrections; they are written up with Huang Xing’s reply on their own page.

02 One family, seven keys

The system is the one described on the Swatow page. Twenty consonants (no w) and five vowels give a hundred syllables numbered 01…99, 00, and two syllables make one four-figure group of the standard Chinese telegraph code. The syllables run on across ten-letter words without regard to the word breaks, and there is no additive. Every table in the family is a rule: the alphabet rotated or reversed, the vowels in some order, the numbers filled vowel by vowel. What changes between correspondents is the rotation and the vowel order:

“Row for 01” is the consonant that takes 01 with the first vowel; the next consonant in the alphabet takes 02, and so on. The key names are those in famtest.py. Two more keys are tentative, each resting on one short telegram: Honolulu (alphabet reversed, filled consonant by consonant, vowels e a o u i; z 4.9) and Manila in May (row for 01 d, vowels e u a o i; z 4.4).

Finding and accepting a key. The whole family, 57,600 tables counting the fill orders, origins and the ±111 additive, was run on each telegram, and each table was scored with a Chinese character model. The z figures on this page are that score measured against the whole family. A short telegram can put a wrong key on top by chance, so a key was accepted only when it reached z ≥ 6 on at least one long telegram and then decoded other telegrams it had not been fitted to. For example, shanghai1916 was found on a long Shanghai → Rosamonde telegram and then decoded Sun’s own outgoing telegrams to Yamada at z 7–10. The May key yamada1916 decodes every telegram that carries the clear prefix word “Thirtyseventh”. Some short telegrams decode under their pair’s key although another table scores slightly higher on them alone. They are counted as decoded because the text makes sense and the key is fixed elsewhere; their own score does not decide it.

The dates check themselves. Chinese telegrams give the day of the month by a rhyme-category character (韻目代日). Where a signature carries one, it agrees with the date on the Japanese form: 支 4, 元 13, 篠 17, 皓 19, 號 20, 箇 21, 漾 23, 有 25, 宥 26, 豔 29. The one exception is a Shanghai telegram handed in on 22 May that ends 馬, the 21st. The date characters come out of the decryption and the key search never looked at them, so the agreement is an independent check.

The copies, and our transcription of them. These are copies made by Japanese telegraph clerks of letter strings that meant nothing to them. Some groups decode only after a one-letter change: g↔q most often, then x against d or s, h↔k↔n, e↔c, a↔o. This page first blamed the clerks for these changes. Most of them were our own misreadings. In October 2026 Brice Dudout checked our transcriptions of the March–May 1916 telegrams against the JACAR scans. This copyist’s q has its loop on the right and his g has it on the left. Of the 47 transcribed g that Dudout checked, 30 are q on the form, and no q is a g. Of the 21 emendations in our transcriptions, 19 are already written correctly on the form. Only two are errors in the 1916 copies: 請 on 4 April and 傷 on 19 May. In all he reports 28 wrong letters. He also found that our transcription of the 3 April Shanghai → Rosamonde telegram had left out a whole word on its second sheet (fr. 0489, sixth line). That is jobusosuji, and putting it back realigns the end of the message: 已澈查局非誣騙 (“we have looked into it thoroughly; it is not a swindle”), where our version had 已[?]非誣騙[?]. Every case checked here since then agrees with him: this one, and the two 1916 forms in section 07. The March 1917 forms in section 06 show our transcription making the same kind of misreading, and two whole words left out. His full list of 28 has not yet been gone through here, letter by letter. A character in brackets on this page is therefore a change to our transcription. It is not necessarily a slip by the clerk. A dropped pair of letters makes a telegram decode as noise until the pair is restored. The 25 May telegram from Tokyo scores z 4.1 as transcribed and 6.8 once a ge is put back in its first word.

03 March–April 1916: the monarchy cancelled, the money, the Shanghai rising

Conventions: [?] = group decoded as written but doubtful; a character in brackets = the decoding after a one-letter change to our transcription (most of these are our misreadings, not the clerk’s; see above); □ = not decoded. Frame numbers are JACAR frames of B03050088300 unless stated.

Yuan Shikai announced the end of his monarchy on 22 March 1916. At that point Sun Yat-sen was in Tokyo and his party was planning a rising in Shanghai.

25 March, Shanghai → Rosamonde (fr. 0468; z 6.7)

帝制取消，馮、朱等當漸觀望。滬異派力弱，於我略可從容佈置。海軍[?]發策較萬全，然猶急備未懈…

“The monarchy is cancelled. Feng [Guozhang], Zhu [Rui] and the others will now wait and see. The opposing faction in Shanghai is weak, so we can make our dispositions at some leisure. A plan launched by the Navy … is safer, but we are still preparing urgently without slackening …” There is no signature.

26 March, Sun → Yamada (fr. 0465; z 7.4)

京信□□□策：袁暫退，使黎代；俟歐戰終，乃借英力[復]出。故滬事當發於袁退之前乃可。文

“The Peking letter … plan: Yuan to step down for now, with Li [Yuanhong] acting; when the European war ends, he comes back with British support. So the Shanghai action must be launched before Yuan withdraws. — Wen.” The same evening a 38-word telegram from Shanghai signed 美 (Chen Qimei; z 9.2) asked for arms to be shipped and a handover procedure, and closed: 穆日當局允暗濟械，已有接洽否？ (“The authorities have agreed to supply arms secretly: has contact been made?”). 穆 occurs twice with the same letters and is probably a cover word.

27 March, Sun → Konglipo, Manila (fr. 0462; manila1916)

滬電乃崇智所發，勿疑。文

“The Shanghai telegram was sent by [Xu] Chongzhi. Do not doubt it.” Earlier that day Manila had asked who was using the code to send telegrams from Shanghai, calling it 萬可疑, “most suspicious”. Xu Chongzhi’s own telegram of the 26th, signed in clear Shuichunchee, is in the Manila key, not the Shanghai one.

30 March – 4 April: money

Much of the traffic is fund-raising. Tokyo asks the Fujian merchants in Manila to 盡力向閩幫籌[款?] (“do their utmost to raise funds from the Fujian guild”); Qingdao asks for fifty thousand, 惟需五萬元，懇速電匯 (“only we need 50,000 yuan; wire it urgently”); Manila reports what seems to be four thousand sent through the Hongkong Bank (a four-word telegram, weakly scored), and Tokyo acknowledges [四?]千圓. From Shanghai on 31 March (z 8.2): 田在滬交五千，到漢匯八千。近[又?]索三萬，無力應之 (“田 [a person] handed over 5,000 in Shanghai, 8,000 to be remitted on reaching Hankow. Lately another 30,000 is demanded; we have no means to meet it”). On the same day Sun warns Yamada (fr. 0485; z 7.9):

北方來電，帝制取消，軍心益振。而[滬?]反[?]觀望。恐前聯絡之人皆多不實，故托此爲辭，[?]得款耳。望兄詳察，勿受其欺。孫文

“Telegrams from the north: the monarchy is cancelled and the army’s spirit rises. But [Shanghai?] is … waiting to see. I fear the people we contacted earlier are mostly not genuine and use this as an excuse to get money. Look into it carefully and do not be taken in. — Sun Wen.”

4 April, Sun → Yamada (fr. 0494; z 9.8, no emendation)

由臺灣電六百金，請交寶昌路寶康里一八三號張宗海，往甘費。聞西林赴南寧組政府，定設法破之。文

“600 gold wired by way of Taiwan: please give it to Zhang Zonghai, 183 Baokang-li, Baochang Road, for the journey to Gansu. I hear Xilin [Cen Chunxuan] is going to Nanning to form a government: I will find a way to break it. — Wen.” This is the cleanest telegram in the files: forty characters and not one letter to change. Two more went out that afternoon. One ends 山東消息甚佳，月內當動。孫文。支 (“news from Shandong is very good; it will move within the month. — Sun Wen, the 4th”). The other, to Kayano at Qingdao (fr. 0492; seito1916, z 7.4), reports fifty thousand sent through the Yokohama Specie Bank, and arms to be shipped under the name of the Qingdao military administration (以寄青島軍政官名義).

13 April, Hankow → Sun (B03050088400 fr. 0517)

This telegram is in digits, not condenser letters. It uses the +111 code that Consul Segawa described to the Foreign Ministry that May: each four-figure code group has 111 added, here without the digit reversal. A search over every additive and both digit orders puts −111 175 log-points clear of the next candidate, and the telegram then decodes with no emendation:

武漢發動事，日人謂非念五萬不辦。迭電陸參兩部，桐以軍械運輸困難且爲時久。刻人心已熟，海陸均有七分以上把握，三萬可行。機不可[失]，望速決電示。元

“On the Wuhan rising: the Japanese say it cannot be done for under 250,000. [We] have wired the Army Ministry and the General Staff again and again; 桐 says the arms and their transport are difficult and would take a long time. Opinion here is ripe; army and navy are both more than seventy per cent sure; 30,000 would do it. The moment must not be missed. Decide quickly and wire. — the 13th.”

04 May 1916: Chen Qimei

By May, Sun was in Shanghai, where he arrived at the start of the month, and the Tokyo end was run by Xie Chi (謝持), who signs 持 and dates with rhyme characters. Every Tokyo–Shanghai condenser telegram of that month that decodes at all decodes under one key, yamada1916 (frames of B03050089100).

19 May, Shanghai → Linwailuk, Tokyo (fr. 0066; z 6.0)

英士昨下午在山田家被兇[轟?]斃，捕兇[一?]人，關係者數人。

“Yingshi was shot dead by assassins yesterday afternoon in the Yamada house; one assassin and several accomplices arrested.” Chen Qimei (陳其美, courtesy name Yingshi) was murdered in Shanghai on 18 May 1916 in the house of Yamada Junzaburō, the addressee of most of the Tokyo traffic. Tokyo replied the same day in two telegrams (both signed 持皓, the 19th): 英士被刺，傷[勢?]如何？盼覆 (“Yingshi stabbed: how are the wounds? Please reply”), and then

電悉。英士捐軀，極堪慟悼。主使爲誰？盼覆。先生[?]在滬[?]，務處處戒備。持皓

“Telegram received. Yingshi has given his life; we mourn him deeply. Who was behind it? Please reply. The Master … in Shanghai … must take every precaution. — Chi, the 19th.” Others asked the same question in other ways. From Kobe, 英士君遭難確否乞覆 came in plain code. From Manila, received on 23 May, 聞陳英士死確否？請覆 came to Tōyama Mitsuru in a key of its own. The Shanghai answer of 22 May (z 8.2) gives the instigator as 遣之暗殺主謀爲[政?]府, “the government”, and says that the pistols have been entrusted to Miyazaki to bring back.

23–25 May: arms and the students

Tokyo on 23 May (z 8.2): Wu Feiji, back in Shanghai at 26 Rue Joffre, has ordered 300 Type-30 rifles and 100 Browning pistols through the Ōkura company at Dalian, with 2,000 taels already paid as deposit. Tokyo and Shanghai also wire back and forth about 席、趙兩處學生事, the business of Xi Zhengming’s and Zhao’s students at Tianjin, which had been demanding money for a rising. On 22 May Shanghai orders it stopped; on 24 May it says 學生事暫擱 (“shelve the students affair”); on 25 May Tokyo asks which of the two it means.

26 May, Tokyo → Yamada (fr. 0092–93; z 9.6)

克強言外務省意先生宜緩赴青島。又軍械事能得青木一電參謀本部較更有力云云…　宥

“Keqiang [Huang Xing] says the Foreign Ministry’s view is that the Master should put off going to Qingdao. Also, on the arms, a telegram from Aoki [probably Lt-Gen. Aoki Nobuzumi] to the General Staff would carry more weight … — the 26th.” Here Sun’s Tokyo office passes the Japanese Foreign Ministry’s own advice to Sun, and the Foreign Ministry was being sent a copy of the telegram. It had no means of decoding it. Earlier that morning Tokyo had asked 報載先生昨[潛?]離滬，確否？ (“the papers report that the Master secretly left Shanghai yesterday; is it true?”).

29 May, Tokyo → Yamada (fr. 0113–14; z 6.6)

據譚根云，岑有軍[火]船四艘由神戸回粤，今遣彼來東運機[?]二臺，交該船載[?]…謝持　豔

“According to Tan Gen, Cen [Chunxuan] has four military steamers going from Kobe back to Guangdong, and has sent him to Japan to ship two machine[-guns?] on them … — Xie Chi, the 29th.” Cen Chunxuan is the “Xilin” whose government at Nanning Sun had promised on 4 April to break.

05 After the condensers: Cai E and Huang Xing

From July the ciphered traffic changes to two systems that are not in the family (below), but some telegrams in the later volumes are in plain telegraph code. Two of them belong to the same story.

10 July, Luzhou → “Ilchang Tokio” (B03050090000 fr. 0417)

宥電敬悉。承念甚感。久罹喉病，軍中失調，醫治漸痊，幸釋綺注。鍔叩。蒸

“Your telegram of the 26th received with respect. I am most grateful for your concern. I have long suffered from a throat illness, made worse by life with the army; treatment is gradually curing it; please set your mind at rest. — E, the 10th.” This is Cai E (蔡鍔), who led the Yunnan rising against the monarchy. The throat illness killed him in Fukuoka on 8 November 1916.

13 September, Tokyo (中田) → Huang Xing, Shanghai (B03050090100 fr. 0471)

閱報憂悚。接[?]電，母[恙?]稍安。請靜養。現如何？

“Alarmed to read the papers. […] telegram received; your mother is a little better. Please rest. How are you now?” Huang Xing was already ill, and he died on 31 October 1916, eight days before Cai E. The Japanese clerk’s own kanji glosses on this form differ from the codebook at two groups (4443 and 3225).

06 Sun and Dai Jitao, March 1917

B03050090200 (vol. 4, item 5, frames 0503–0513) holds three condenser telegrams of late March 1917. Two are from Sun in Shanghai, signed Sunwen in clear, to “Toi c/o Kikuchi Rioichi” in Tokyo. The third is the reply to “Sun yatsen Shanghai”, signed Tenkiutai (天仇, Dai Jitao). “Toi” is therefore Dai Jitao (戴季陶), staying with Kikuchi Ryōichi (菊池良一).

The key is a seventh member of the family: row for 01 k, vowels a e u o i, no additive. It was found by running the family on the 23 March telegram alone, where it scores −321.6 against a population mean of −556.4 and s.d. 25.8, about nine standard deviations up; the next unrelated key is 124 points behind. The Swatow key gives nonsense. The 25 and 26 March telegrams played no part in finding the key, and both decoded under it at once.

23 March, Sun → Dai (fr. 0504)

前電云三日後商妥□□□□□□□□□接後音。此間必俟前約取消，始可致電北京照辦。各約已取消否？速覆。公司抄件瑞士尚未辦妥。文

“My previous telegram said that after three days [the matter] would be settled by agreement … received further word. Here we must wait until the former contract is cancelled before we can wire Peking to act accordingly. Have the contracts all been cancelled? Reply at once. The company copy … [瑞士?] not yet settled. — Wen.” Words 4 to 7 on the form are pixaqaqe macorarala veyurojiji eianakerele. Our first transcription left out the two middle words, and this page then blamed the gap on the clerk. With them restored, all 22 code words on the form are accounted for. Two faults are on the form itself: pixaqaqe has eight letters, and eianakerele has eleven and begins with ei, which is not a syllable. pi completes 妥, and lebe 接 decodes as written, so “…接後音” is likely. About nine codes between them are still not decoded. If one syllable is put into pixaqaqe, the restored words give ordinary characters (再, 告, 候, 至) instead of noise, which suggests that a syllable is missing there. They do not give a text that can be adopted. 瑞士 needs an x/m change and is the least certain emendation. The telegram’s last word, qixapfete, has nine letters; read qixa pi ne te it gives 辦妥 and 文, Sun’s signature character, which also closes the 25 March telegram.

25 March, Sun → Dai (fr. 0507)

函悉。約取消後，當□□□能否。取消卽覆。文

“Letter received. After the contract is cancelled, [whether] … is possible. Reply as soon as it is cancelled. — Wen.” 能 is xubo. We first read its first letter as z, but it has the shape of this clerk’s x elsewhere on the form. Three codes do not decode, and no one-letter variant makes sense of them.

26 March, Dai → Sun (fr. 0506)

本定今日赴熱海。□□來云：彼等須先□定，約三日回。□公司事如何？電示。宥

“I was to have gone to Atami today. … came and said that they must first settle …; [I shall be] back in about three days. How stands the company matter? Wire me. — the 26th.” 公司 has the same qabo for 司 that decodes cleanly on 23 March. The two characters before 來云 do not decode. 小池 (Koike Chōzō, who had left the Foreign Ministry’s Political Bureau in 1916) is two letters away and is no more than a guess. The first word begins with the clerk’s capital Q, so 本 (qeta) is as written on the form. The other one-letter changes on the three March 1917 forms have not yet been checked again on the images.

Together the three telegrams show Sun, in the last week of March 1917, waiting for “the former contract” to be cancelled before he can wire Peking, pressing Dai for news, and Dai asking after “the company matter”. They do not say what the contract or the company was. That needs the Chinese record for March 1917, which has not been consulted.

07 July–November 1916: Brice Dudout’s two grids

The ciphered telegrams of July to November 1916 in B03050089900–B03050090100 were decoded in October 2026 by Brice Dudout, who has published his readings, transcriptions and solver at bdudout.github.io/wenmi (github.com/bdudout/wenmi, CC BY 4.0). They are in two systems. This page had earlier taken them for a single “文密 code”. Both systems number the syllables consonant by consonant, with the vowel order reversed on every second consonant. Each pair of syllables then gives one standard telegraph code, with no additive.

• Tokyo → the 軍務院 (Zhaoqing, later Canton), June–September 1916. After the 軍務院 was dissolved, this channel went on to Cen Chunxuan’s headquarters (岑都司令, 13 September, fr. 0455b; 都司令部, 30 September, fr. 0462, with 李印泉, Li Genyuan, named in clear). One telegram, fr. 0406, went to the 中華新報 in Shanghai. Five vowels, and twenty consonants that include w and y and leave out j, in the order h g f d c b z y x w v t s r q p n m l k. The syllables run ha = 00 … hu = 04, gu = 05 … ga = 09, and so on to 99. Most forms on this channel carry the clear prefix 文密 (2429 1378); in Dudout’s set only fr. 0386 and 0462 lack it.
• Tokyo → Yunnan (Tang Jiyao, the provincial assembly, the finance office) and Shanghai (中華新報), July–November 1916. Six vowels, with y as a vowel, and seventeen consonants in alphabetical order, b d f g h k l m n p r s t v w x z. A syllable’s value is 6 × consonant rank + vowel rank, with the vowels in the order y u o i e a, reversed on odd ranks. The last two cells, ze and za, would be 100 and 101, and are never used. Dudout fixed the grid from a clear-code telegram sent the same day (fr. 0489b). He then tested it on telegrams that played no part in finding it, among them the report of Cai E’s death at Fukuoka on 8 November (fr. 0496). Most forms on this channel begin with Cg in clear, which is counted in the word count; none of the five-vowel forms has it. What it stands for is not known.

Both grids were checked here on our own copies of the forms. The transcriptions used were made here, not taken from Dudout’s files. Where they differed from his, the forms agree with him.

5 November 1916, Tokyo → 唐督軍, Yunnanfu (B03050090100 fr. 0489a), six-vowel grid

擬十號南歸。謹聞。伸 歌

“I plan to return south on the 10th. Respectfully reported. — Shen, the 5th.” All nine codes come out as written on the form. The first word is gyrobatapo, as Dudout reads it; our transcription had gyrotatapo and so needed a change for 十. The clear-code telegram on the same sheet, to the provincial assembly (省議會), reads 十號起程月末抵滇 伸 歌 (“leaving on the 10th, reaching Yunnan at the end of the month”). The sender’s address on both forms is written 青山南町四ノ二一 関方 趙伸: Zhao Shen, care of Seki, Aoyama Minami-machi. 伸 is therefore 趙伸 (Zhao Shen), the 軍務院’s commissioner in Japan. Dudout identified him from fr. 0498.

19 July 1916, Tokyo → 軍務院, Canton (B03050090000 fr. 0416b), five-vowel grid

文密 報傳取消，日人態度遽變，確否？速示，俾便行止。伸 [皓]

“Reported cancelled, and the Japanese attitude has suddenly changed: is it so? Advise at once, so that I can decide whether to go ahead or stop. — Shen, the 19th.” One one-letter change, i→e, gives 皓, the rhyme-day code for the 19th, which is the date on the form. Dudout makes the same change. This page first had two more changes, and a different ending, 俾便行檄 (“so that the circular can go out”). Both came from our misreadings. The form has xodugoluro, whose first letter has the same shape as both x in the next word, xoxotuhica; we had read it as r. It has qiheyuxifa, which we had read with a g. The q gives 止 (2972) where g gave 檄 (2907). Dudout pointed out that the companion telegram sent the same day on the other channel ends with the same phrase. Transcribed here and decoded in his six-vowel grid with no change, it reads:

19 July 1916, Tokyo → 中華新報 Shanghai, for Tang Jiyao (B03050090000 fr. 0416a), six-vowel grid

轉文羣轉滇督 Cg 滇索槍萬枝可辦到。因報傳軍院取消，方針遽變，確否？速示，俾便行止。伸 皓

The first six groups are in clear code: “forward to 文羣, to be forwarded to the Yunnan governor”. The rest reads: “The ten thousand rifles Yunnan asks for can be arranged. But on the report that the 軍務院 is to be dissolved, the line has suddenly changed. Is it so? Advise at once, so that I can decide whether to go ahead or stop. — Shen, the 19th.” The decoding is Dudout’s grid applied to our own transcription of the form.

A correction. This page earlier treated the July 1916 traffic as a single code and gave the six-vowel system the name 文密. On fr. 0416 the prefix 文密 stands on the five-vowel telegram to the 軍務院. The six-vowel telegram on the same frame, to 中華新報, opens instead with 轉文羣轉滇督 in clear code. The six-vowel search run here (about 200,000 tables) could not have found either grid. It was run only on the June 1917 telegrams, with rotated consonant orders and no reversal of the vowels.

08 The wider series, 1913–16

Brice Dudout has since carried the work back into the 1913–1915 volumes of the same series (JACAR B03050087300–B03050087700) and forward into the numeric telegrams of June–July 1916. By 19 October 2026 his site gave readings of 194 telegrams in eight systems, with the scans, his reviewed transcriptions, the raw decodes and working translations (telegrams, method). His count, in telegraph codes decoded:

The condenser row includes the March–May 1916 telegrams on this page, which Dudout re-read against the scans, and the tables first recovered here. Not counted: the 22 Rosamonde telegrams in enciphered English (page).

The condenser tables he added are all of the same family as the keys in section 02. Each is given here as consonant order / vowel order / filled by vowel (V, value = 20 × vowel rank + consonant rank) or by consonant (C, value = 5 × consonant rank + vowel rank) / first number:

• “Laisaosang”, Tokyo ↔ Shanghai, June 1914 – February 1915: c…b / o a e u i / V / 01.
• “New York”, New York → Sun, July 1914 – February 1915: b…z (p and q merged) / a e i o u / C / 00, the simplest table of the family.
• “Hsuweilin”, San Francisco → “Hsuweilin”, July 1914: b…z without k / a e i o u / C / 01.
• “Thirtieth”, Shanghai ↔ Tokyo, February–May 1915, named in clear on the forms: v…t / u i e o a / V / 01.
• “Thirteenth”, Tokyo → Shanghai, February–March 1916, also named in clear: n…m / o a e i u / V / 01.
• “Sixteenth” and “Seventeenth”, Shanghai ↔ Tokyo, April 1916: t…s / e u o i a and p…n / i u a e o, both V / 01.
• “Thirtyeight”, Tokyo → Shanghai, early May 1916, the yamada1916 construction with other parameters: k…j / e a i u o / V / 01. It decodes the telegram of 12 May about the iron trunk left with Inuzuka (箱已交犬塚，十五日附法船來) and that of 11 May asking whether Zhang Ji is back in Japan.
• Huang Xing’s table, Tokyo ↔ Shanghai (“Ylchang”, Huang Xing’s Tokyo address), late May – June 1916: consonants d c b t h j f q g k n m p s r l x w z v / u i o e a / C / 01. The Tokyo telegrams end 覆。興 (“reply. — Xing”), and the Shanghai ones are signed 文 or 孫文. On 3 June Huang reports that he has seen Tanaka and that 槍二千、砲六門不日可運 (“2,000 rifles and 6 guns can be shipped shortly”).

The “Hong Kong” telegrams of March 1914 are not condensers. Each letter stands for one digit (y k u l o c a p i t = 0–9), and four digits make one standard code: kcpk = 1571 = 已. Two keys this page had marked tentative now stand on more than one telegram. The Manila table reads a Tokyo → “Konglipo” telegram of 6 January 1915 with no change (無其事。文覆, “there is nothing of the sort. — Wen, replying”), sixteen months before the telegram it was found on. The Honolulu table reads the “Aloha” telegrams of April–May 1916 if w stands for v.

Dudout’s site also names the sender and recipient of every telegram it publishes, with the evidence for each attribution (signature, address, header or series) and a grade of certain or probable. None of these readings has been checked here on the forms, apart from the three in section 07.

09 Brice Dudout’s tables in the repository

On 27 October 2026 Brice Dudout sent the full texts behind section 08 as pull requests (#144–#163): for each telegram the transcription of the form, the decode group by group, the Chinese text with punctuation and a working English translation. One key is one folder, so each table below has its own folder in the repository. They are his readings, published on his site; they were read through as sentences when merged, but the forms were not re-read here. These tables have no page of their own: this list is their entry on the site.

Six more tables share the telegrams of this page and are filed with them, in sunyatsen/y1916/dudout/: the “Shanghai” table (36 telegrams, July 1914 – April 1916), the “Yamada” or “Thirtyseventh” table (29, May–June 1916), the “Qingdao” table (19, March–April 1916), the “Manila” table (8, January 1915 – May 1916), the “San Francisco” table (5, September 1914 – April 1916) and the “Honolulu” table with its “Aloha” alphabet (2, April–May 1916). The same folder holds his 28 corrections to our transcriptions (corrections.tsv) and his re-reading notes. The numeric codes of June–July 1916 are on their own page, with his N01 and N09 乾密 texts added the same day, and the Rosamonde telegrams on theirs.

10 What remains open

• The June 1917 telegrams between Sun and Dai (B03050090200 fr. 0509, 0511) are in neither 1916 grid. They also contain clusters that are not consonant–vowel pairs (myzoasdoke, liriopen). The second word of fr. 0509 is written vemucäyri. In Morse, ä (·−·−) is e + k run together, so Dudout reads vemucekyri. Dudout’s search on these telegrams, every table scored with the same language model on the standard telegraph code, has found nothing:
  • alphabetical grids of 17–20 consonants × 6 vowels, in every rotation, both directions, all 720 vowel orders, serpentine or not: about 194,000 tables on a machine transcription, 241,000 on his own and 840,000 with y as a consonant;
  • free consonant and vowel orders, by simulated annealing, for 17 × 6 and 18 × 6 tables filled consonant by consonant (with and without the 1916 alternating vowel order), vowel by vowel (as in the March 1917 key), and vertically serpentine, from 0 or 1. As a positive control, Chinese text of the same length (about 90 codes) enciphered with a random table of each layout came back whole, 88 of 88 codes in every layout. On the June 1917 text the best tables score about +0.9 per code, against about +0.6 for the null, and no two runs agree; a true table should give +3 to +5;
  • the same tables with the Hankow-style transformations (±111, reversal, digit-wise −111);
  • a 20 × 6 coordinate model with anchors (文 or 孫文 at the end of 0509, 戴 at the end of 0511, dates), and 42 probable words (田中, 秋山, 復辟, 張勳, 國會, 政府, 孫文 …) in all 2,101 compatible placements;
  • the known keys: shanghai1917, the Yamada table, seven 1916 condenser keys and both 1916 grids.
  What is left is a renumbered code book, another table structure, or too little text. Dudout also notes a structural lead. If the second sheet of 0511 is shifted by one syllable, four four-letter groups repeat: cafe twice there and once in 0509, and kige, fojy and mafy twice each. A shift of that kind could come from a dropped letter at the join of the two sheets. Transcribed, not decoded.
• Late May 1916. No key here decoded the five Tokyo → Junzaburo telegrams without the “Thirtyseventh” prefix, or the Shanghai → “Ylchang Tokio” telegrams that use w as a consonant. Dudout found that Tokyo telegrams without a prefix and the Shanghai → “Ylchang” ones are Huang Xing’s traffic, and decodes nine with a table of their own (section 08). Whether all five of the telegrams listed here are among them has not been checked. Two Shanghai → “Ylchang” telegrams, of 29 May and 8 June, from another sender, still do not decode in it.
• The kana stream of B03050089100 fr. 0076–0078, which the Foreign Ministry’s analysts worked on, is a different system, the three-kana code: two telegrams of 20 May 1916 to Huang Xing, decoded with Tomokiyo’s formula by Harumi-Matsuba (October 2026) and checked here; see Huang Xing’s three-kana telegrams.
• The Swatow original. Every Latin-letter form in B03050088300 and B03050088400 was examined, and none is from Swatow. The copy in B03050738800 is still the only one found, and its three garbled tail groups stay lost.
• Undecoded groups. Most long telegrams keep a few groups marked [?]: names that are not in the model, codes one letter from several readings, and supplementary-code characters. The two keys first marked tentative (Honolulu; Manila) are now supported by further telegrams (section 08). About nine codes of the 23 March 1917 telegram, and the identity of the contract and the company, are open.
• Dudout’s 28 corrections to our March–May 1916 transcriptions are reported above. He checked each one on the scans, and every check is in his re-reading table (solver/data/relecture-1916.csv). They have not been checked here one by one, and the per-telegram transcripts in this repository still have the old letters except where this page says otherwise.
• Numeric codes 和密 and 乾密. A few short numeric telegrams of June 1916 carry these prefixes, or none. They are too short to attack on their own.

11 Sources

• JACAR (Japan Center for Asian Historical Records), 外務省記録 1.6.1.4-2-1-2, 各国内政関係雑纂／支那ノ部／革命党関係（亡命者ヲ含ム）／特定人ノ発受郵便物取調, vols 2–4: B03050088300, B03050088400, B03050089100, B03050090000, B03050090100, B03050090200.
• JACAR B03050738800, for the Swatow telegram and Consul Segawa’s report of 11 May 1916 on the +111 code; see the Swatow page.
• S. Tomokiyo, “Chinese Cryptography: 1871–1945” (cryptiana), for Yamada Junzaburō’s two condenser tables, which first suggested the family.
• Brice Dudout, bdudout.github.io/wenmi (repository github.com/bdudout/wenmi), October 2026, for the two grids of July–November 1916 and the readings made with them, and (emails of 11 and 12 October 2026) for the corrections to our 1916 transcriptions and the account of his June 1917 search; and (emails of 15, 16 and 19 October 2026) for the 1913–1916 tables, the Rosamonde telegrams and the numeric codes of June–July 1916.
• Unicode Unihan database, kTaiwanTelegraph / kMainlandTelegraph, for the standard telegraph code.

Per-telegram transcriptions, headers, keys, z-scores, emendations and every undecoded group are in the repository: targets/sunyatsen/y1916/transcripts_A.md–transcripts_D.md and NOTES.md, targets/sunyatsen/y1917/NOTES.md, the family search targets/sunyatsen/famtest.py, and the Hankow decoder targets/sunyatsen/y1916/hankow.py. The JACAR PDFs are not committed.