卷:卷三《阳关以西》|主角:Marco·常玉山|古人:张议潮(归义军节度使)|地点:莫高窟第一五六窟《张议潮统军出行图》前·常玉山院子|典:《资治通鉴》大中五年“沙州刺史张议潮遣兄议潭奉十一州图籍入见”;《张淮深碑》载遣使十辈,唯一辈自天德军路达长安,历二年|主导感官:眼(窟内很暗,讲解员的手电从马蹄上扫过去)|合券进度:半张|金融-国学对应:Redundancy Is Priced Against Failure, Not Against Success(发十路的钱,要跟「消息不到」比,不能跟「消息到了」比)↔ 遣使十辈唯一辈达
段一·现身
十一月二十日,上午十点。
莫高窟。
马可又买了一张票。
这一次是特窟,一百五十块,只看一个。
第一五六窟。
讲解员是个中年男的,姓何。
这个窟不在常规线路上,要另外开门。
门开的时候有一股很凉的气出来。
里面没有灯。
何老师打着手电,先照顶。
然后手电往下,落到南壁下面那一条。
他说:
“这一条叫《张议潮统军出行图》。”
手电从右往左扫。
马可跟着看。
先是鼓吹。
然后是舞伎。
然后是旌节。
然后是一队一队的骑兵。
马蹄一个接一个,都是侧着的。
画得不算精细,可是很有劲。
手电扫了大概十几米,才扫到中间那个人。
何老师说:
“这个是张议潮。”
马可凑近看。
那个人骑在马上,穿的是红袍。
他在整个队伍里,个子不比别人大多少。
要不是手电停在那儿,看不出来是主角。
马可问了一句:
“何老师,这一幅是什么时候画的?”
何老师说:
“咸通年间。”
他补了一句:
“是他侄子张淮深修的窟。”
马可又问:
“他做的是什么事?”
何老师把手电关了一下,又打开。
他说:
“大中二年,他把沙州从吐蕃手里夺回来了。”
他顿了一下。
他说了一句,说得很平:
“这地方被吐蕃占了六十多年。”
马可问:
“六十年里,这些窟还在开吗?”
何老师说:
“在开。”
这个回答马可没想到。
他问:
“谁开的?”
何老师说:
“本地人。”
他把手电往壁上另一处照了照。
“吐蕃占着,日子还得过。”
马可在那个黑窟里站着,想了一下这句话。
六十年。
六十年不是一个数字。
是一个人生下来,长大,成家,老了,死了。
这中间有人娶亲,有人下葬,有人借麦子,有人还麦子。
昨天在陈列中心那些契,多数就是这六十年里写的。
他忽然明白了一件事。
那些账没有断。
换了谁来管这个地方,账还是那么记。
他在本子上写了一行,写完自己看了两遍:
「上面换了,下面的账没换。」
他停了一下,加了半句:
「所以张议潮回来的时候,图籍是现成的。」
段二·古文镜像
“沙州刺史张议潮遣兄议潭奉瓜、沙等十一州图籍入见。”
“遣使十辈,间道诣阙。唯一辈达天德军,历二年,方至长安。”
English Mirror: The prefect of Shazhou sent his elder brother bearing the maps and registers of eleven prefectures to court. — Ten parties of envoys were despatched by hidden roads. Only one reached the Tiande garrison, and two years passed before it arrived at Chang'an.
那天上午,马可在那个窟里问了何老师一个问题。
他是因为一句话问的。
何老师刚才说,张议潮夺回沙州以后,往长安报信。
马可问:
“报信要多久?”
何老师说:
“两年。”
马可以为自己听错了。
“两年?”
何老师说:
“两年。”
他把手电往队伍的最后面照。
“当时河西这一带,中间隔着的地方还在吐蕃手里。”
他说:
“从沙州到长安,路不通。”
马可问:
“那怎么报的?”
何老师说了一句,马可当时就把本子拿出来了:
“他派了十路。”
那天上午出了窟,马可在栈道上追着问完了。
何老师本来讲完就要走,被他拦住了。
马可问:
“十路,到了几路?”
何老师说:
“一路。”
马可站在那儿。
他问:
“其余九路呢?”
何老师说:
“不知道。”
他说得很平常。
“史书上没有写。”
马可问:
“到的那一路,走的哪条道?”
何老师说:
“往北,绕到天德军。”
他用手在空中画了一下。
“天德军在今天内蒙那边。”
马可在心里过了一下地图。
从敦煌往长安,直着走两千公里。
往北绕到内蒙,再往南到长安。
至少多一千五。
他问:
“绕最远的那一路,反而到了?”
何老师说:
“是。”
他把讲解器收起来。
他说了一句,然后就走了:
“近的那几条,都是有人守着的。”
马可站在栈道上,看着何老师走远。
他把刚才的对话在本子上补全。
补的时候他发现有一个数他没有问。
十路一共多少人。
他后来在网上查了半天,也没查着。
史书上只写了「十辈」。
一辈是几个人?
按当时出使的规矩,少的三五个,多的十几个。
就算一路五个。
十路五十个人。
到的一路,五个。
其余四十五个,史书上一个字也没有。
马可在本子上写了一行:
「四十五个人,没有名字。」
他看着这一行,很久没有往下写。
因为他知道自己今天下午要去做一件事:找第二家供货。
找第二家的意思是,其中一家将来会用得少。
这两件事在他心里是连着的,可他说不清怎么连的。
段三·事件主体
那天下午,马可去了常玉山家。
他是有事去的。
这件事他昨天晚上想了一夜,早上又想了一路。
他要跟常玉山说,他要找第二家。
这件事按常理是不说的。
你找第二家供货,是你自己的事,办完了对方自然知道。
先说,等于告诉人家我不放心你。
可是马可想起前天在嘉峪关,秦师傅说的那句。
余量藏起来,就是错。
余量放在明处,才是算对了。
他决定说。
那天下午三点,两个人在院子里坐着。
炉子搬到了外头,晒太阳。
常玉山在剥一堆枣,说要给他寄的。
马可说:
“常师傅,我有一件事得跟您说。”
常玉山没有抬头。
他说:
“说。”
马可说:
“我要再找一家。”
常玉山手上停了。
就停了那么一两秒。
然后他接着剥。
他问:
“为什么?”
马可把理由说了。
他说得很直:
“三月那一批,我只有您一家。”
“嗯。”
“您要是那个月出点事,我三月就是空的。”
常玉山说:
“我不会出事。”
马可说:
“我知道您不会。”
他停了一下。
他说了一句他自己也觉得难说的:
“可是您跟我说过,同一座房,东头西头差两个级。”
这一句说完,院子里静了一会儿。
常玉山把手里那把枣放下了。
那天下午,常玉山没有生气。
他先是没说话,剥完了手里那几个。
然后他抬起头。
他说:
“你说得对。”
马可松了一口气,可是没敢表现出来。
常玉山说:
“不过有一件事你得听我的。”
马可说:
“您说。”
常玉山说了一句,马可当天晚上抄了两遍:
“第二家我给你找。”
马可愣了一下。
他问:
“为什么您给我找?”
常玉山说:
“你自己找的,你不知道他斤两。”
这一句还好懂。
下一句马可想了很久:
“你自己找的第二家,会跟我压价。”
他把那盆枣往前推了推。
他说:
“我找的,不会。”
马可在那个院子里坐了很久。
这句话里的东西,比他昨天算的所有账都实在。
他自己去找,找到的人不认识常玉山。
不认识,就会报一个比常玉山低的价,为了抢这一单。
下一年常玉山就得跟着降。
常玉山降了,第三年那个人再降。
两年下来,马可省了钱,两家都做不下去。
常玉山找的人,是常玉山认识的。
认识的人不会为了一单去挖一个认识的人。
马可问了一句:
“常师傅,那您找的那家,跟您是什么关系?”
常玉山说:
“不熟。”
马可以为听错了。
“不熟?”
常玉山说:
“太熟的不行。”
他解释了一句,这一句很实在:
“太熟的,他家出事我也帮不上,我家出事他也顶不上。”
他抬起头。
“两家挨着的房,一场火全没了。”
马可在本子上写了一行:
「不能不认识,也不能太近。」
他停了一下,加了一句:
「太近的两家,是同一家。」
写完这两行,他问了一个很实际的问题:
“常师傅,他要是给我的价比您低呢?”
常玉山说:
“不会低。”
马可问:
“您怎么知道?”
常玉山说:
“我给他打电话的时候会说清楚。”
马可问:
“说什么?”
常玉山这时候抬起头,看了他一眼。
他说:
“我说,这个人我在做,你别抢。”
马可愣住了。
他说:
“常师傅,这么说,我就不能比价了。”
常玉山说:
“你比不了。”
他说得很直接,一点也没有绕。
马可在那儿坐着,心里过了一遍这笔账。
这句「你别抢」,把他省钱的路堵死了。
可这句「你别抢」,也把两家互相杀价的路堵死了。
他自己算:要是能比价,他一斤能省多少?
顶多五毛。
四百斤,两百块。
要是两家杀起来,两年以后其中一家不做了。
那时候他又回到一家。
回到一家,价格就由那一家说了算。
马可在本子上写了一行:
「我用两百块,买了两家都在。」
写完他划掉,重写了一句更准的:
「我买的不是两家都在。」
「是两家都在,而且都不恨我。」
张议潮是那天傍晚出现的。
在常玉山院子外面那条土路上。
马可走出来抽了口气,看见路那头站着一个人。
穿的是红袍,跟画上一样。
可是走近了才看清,那件袍子很旧,袖口磨破了。
他不像画上那么大。
就是一个五十来岁的人,脸很黑。
他说的第一句是:
“你今天在窟里问了几遍。”
马可说:
“节度使,”
张议潮摆了摆手。
他说:
“问什么,你说。”
马可问的第一个问题很直。
“为什么派十路?”
张议潮说:
“因为一路到不了。”
马可问:
“两路呢?”
张议潮说:
“两路也到不了。”
他站在那条土路上,往东边看了一眼。
他说:
“我算过。”
马可赶紧问:
“怎么算的?”
张议潮说:
“中间有六段路,每一段都有人守。”
他伸出手,比了六。
“一路过一段,能过的,十有二三。”
马可在心里算。
十分之二三,过六段。
零点二五的六次方。
他算不出来,可他知道这个数极小。
他说:
“节度使,这么算,十路也到不了。”
张议潮说:
“是。”
他这时候说了一句,马可站住了:
“所以十路不走同一条道。”
那天傍晚,张议潮在那条土路上,把这件事讲清楚了。
他讲得很慢,因为这件事他算过很多遍。
“十路走同一条道,六段全一样。”
“十路走十条道,”
他停了一下。
“有的道近,有的道远。”
马可说:
“近的容易被截。”
张议潮说:
“近的一定被截。”
他说得毫不含糊。
“守的人也不傻。他守的就是近的。”
马可问:
“那为什么还派人走近的?”
张议潮这时候看了他一眼。
他说了一句,这一句是这一回的骨:
“走近道的那几路,是让守的人有事干。”
马可站在那条路上,半天没说话。
他明白了。
十路里有几路,从一开始就不是为了到。
是为了让人以为消息在这几路上。
他问了一句:
“那走近道的人,知道吗?”
张议潮说:
“不知道。”
他说得很平。
马可又问:
“您呢?”
张议潮说:
“我知道。”
风从戈壁上过来。
马可没有再往下问。
过了一会儿,张议潮自己说了:
“这件事我这一辈子只做过一次。”
他停了一下。
“我不想说它对。”
他抬起头。
“可是十一州的图籍,得有人送到。”
那天傍晚,马可问了最后一个问题。
这个问题跟他自己有关。
“节度使,我这一批货,两百袋。”
张议潮说:
“两百袋。”
“我只有一条道。”
张议潮问:
“几段?”
马可说:
“从敦煌到成都,快递,一段。”
他补了一句:
“可是发货、分装、团长,加起来三段。”
张议潮问了一句,问得很准:
“哪一段断了,你三月就没有了?”
马可想了一下。
他说:
“任何一段。”
张议潮说:
“那你是一路。”
马可问:
“节度使,我要几路?”
张议潮说:
“你这个不用十路。”
“为什么?”
他说:
“我那六段,是有人拦。”
他把手放下。
“你那三段,没有人拦。”
马可想了一下。
确实。
快递不会有人截他的葡萄干。
蒋姐不会故意不发。
常玉山不会故意不做。
张议潮说:
“没有人拦的路,断了是碰上的。”
他说:
“碰上的事,两路就够。”
马可在本子上写这一句的时候,手停了一下。
因为这一句里有一个很硬的道理。
有人拦的路,多派几路没用,得换道。
没人拦的路,不用换道,多派一路就行。
他抬起头。
“节度使,怎么分?”
张议潮说:
“先看哪一段最容易碰上。”
马可想了一下。
发货那一段:常玉山六座房,三月不下雪不会有事。
快递那一段:三月,敦煌到成都,天气好,问题不大。
团长那一段:蒋姐一个人。
他说:
“成都那一段。”
张议潮说:
“那你在成都要两路。”
马可说:
“我只有一个蒋姐。”
张议潮说了一句,马可当天晚上写在纸上画了框:
“她说要给你三个。”
他抬起头。
“你到今天,一个也没有去要。”
他往东边看。
他说:
“两百袋不多。你分两批发。”
马可问:
“同一家发两批,有什么用?”
张议潮说:
“不是同一天到。”
他说:
“头一批到了,你就知道路通不通。”
马可在本子上飞快地写。
张议潮又说了一句:
“头一批不要两百。”
“多少?”
他说:
“三十。”
段四·钩子结尾
那天晚上,马可把《西线便物历》拿出来,在最后加了一页。
这一页他写的是发货计划。
原来是:三月十号前,两百袋,一次发。
改成了三行:
二月二十号,三十袋。
三月五号,七十袋。
三月十五号,一百袋。
他在旁边写了为什么。
第一批三十袋,试路。
这三十袋不是给蒋姐的群卖的。
是发给蒋姐、三个团长、和他自己。
每人几袋,吃。
第二批七十袋,第一次上群。
第三批一百袋,看第二批的反应再定发不发。
他写完这三行,算了一下多出来的成本。
分三次发,运费从一袋一块二涨到一块五。
两百袋,多花六十块。
六十块。
他把这个数写下来,在旁边画了一个圈。
然后他在圈旁边写了另一个数。
一万一千一。
这是三月发不出货,蒋姐一年要多花的补人钱。
他在下面写了一行:
「六十块,买的是不至于赔那一万一。」
写完他停了一下。
他觉得这一行还是不对。
他重写了:
「冗余的钱,不能跟到货的收益比。」
「要跟不到货的代价比。」
那天晚上十点,常玉山发来一条消息。
只有一个名字和一个电话。
跟罗代办给他的那张烟盒纸一模一样。
一个姓,一个地方,一串号。
马可看清了地名。
吐鲁番。
他打了一句:
「常师傅,这个人做什么的?」
常玉山回:
「跟我一样。」
又一条:
「他那边房子多,一百多座。」
马可算了一下。
常玉山六座。
那边一百多座。
他打了一句:
「那他为什么要跟我这么小的量做?」
这一条发出去,常玉山过了很久才回。
回的是一句让马可坐直的话:
「他不为量。」
下一条:
「他要一个每年三月都来拿货的人。」
马可盯着这两条看了很久。
因为这句话跟他这十三天在找的东西,是一样的。
他在找一个每年三月都有东西发的人。
那边在找一个每年三月都来拿货的人。
两边找的是同一件事的两头。
只不过一个在阳关以西,一个在成都。
中间三千公里。
马可正要回,那边又来了一条。
这一条只有八个字:
「七月你来一趟吐鲁番。」
他打了一个「好」,发出去。
然后他把手机放下,走到窗前。
敦煌的夜里没有什么灯。
他站在那儿,想的是白天张议潮那句。
走近道的那几路,是让守的人有事干。
他想了很久这句话跟他有什么关系。
想到最后他明白了一件事,这件事他不太想承认。
他这条线上九段,有几段是明知道要断的。
成都那一段就是。
蒋姐一个人,撑不住四万斤。
他早知道。
可他这十三天,一直在西边算价钱。
他在本子上写了今天最后一行:
「我一直在修最好修的那一段。」
「因为那一段有数可算。」
写完他关灯。
躺下以后他又爬起来,把这一行补完了:
「成都那一段没有数可算。」
「那一段只有一个人。」
这一行写完,他躺了大概十分钟,又开了灯。
因为他想起张议潮傍晚说的最后一句。
她说要给你三个,你到今天一个也没有去要。
他把手机拿过来,翻到跟蒋姐的对话。
往上翻。
十一月十三日,酒泉那天晚上。
蒋姐说:我不出钱,我出人。我给你拉三个团长。
这条消息发出来已经七天了。
这七天里,他跟蒋姐说过三次话。
三次说的都是货、日期、流程。
一次也没有提过那三个团长。
马可坐在床上,想为什么。
头一个想到的理由是:货还没有,提了没用。
这个理由听着很合理。
可他知道不是。
真正的理由是,他不知道要那三个人来做什么。
两百袋,蒋姐一个人就发得完。
多三个团长,货不够分。
他一直觉得,得先有足够的货,才好去要人。
可他今天下午刚在纸上写过那一列数。
六千斤六毛,一万零五百斤五毛五,两万斤四毛八,四万斤四毛一。
那一列是从右往左读的。
先有量,成本才下得来。
人也一样。
先有人,货才敢做大。
他不能等到有四万斤了再去找团长。
因为没有团长,他就不敢做四万斤。
马可在本子上写了一行,写完在旁边画了一个框:
「我一直在等条件够了再去要。」
「条件是要来的,不是等来的。」
他把手机拿起来,给蒋姐打了一段字。
打了删,删了打,弄了快二十分钟。
最后发出去的是三句:
「蒋姐,您上回说的那三个团长。」
「二月二十号我有三十袋试货,不卖,就是给人吃的。」
「能不能这三十袋里,给她们一人五袋。」
发完他看了一眼时间,十一点四十。
他以为要等到明天。
结果两分钟就回了。
蒋姐发来的是一句:
「你终于开口了。」
马可盯着这五个字。
下面紧接着又来一条:
「我等了七天。」
马可回了一句:
「您为什么不先说?」
蒋姐回得很快:
「我说了就是我求你带她们。」
下一条:
「你开口,是你要她们。」
马可看着这两条,在床上坐了很久。
这两句的意思,他这一路听过很像的。
常玉山说:第三年再去给,我怎么开口。
道真说:他要是得说,他就不来了。
安员进说:罚重了他就走了。
今天晚上蒋姐说的是第四个:我说了就是我求你。
这四个人说的是同一样东西。
开口是有代价的。
而且这个代价,落在谁头上,事情就不一样。
马可在本子上写了一行,写完手停在那儿:
「我这两年,一直等别人先开口。」
他停了很久,才写下面一行:
「因为先开口的那个人,欠着后开口的那个人。」
又停了一会儿,写完最后一句:
「这七天,那三个团长一直在那儿。」
「差的只是我一句话。」
欲知后事如何,请看下回,《曹议金的三门亲》
附录一·Dongpo's Note
Another ticket at Mogao, a special cave this time, a hundred and fifty for one: Cave 156. Not on the standard route; it has to be unlocked, and cold air comes out when the door opens. No lighting inside. He, the guide, put the torch on the ceiling first, then dropped it to the strip along the lower south wall. This is the Procession of Zhang Yichao.
The beam went right to left. Drums and horns, then dancers, then the standards, then rank on rank of cavalry, every hoof in profile, not finely painted but with a great deal of force in it. Ten or twelve metres before the torch reached the man in the middle. That is Zhang Yichao. Red robe, on horseback, and no larger in the composition than anybody else — without the torch stopping on him you would not know he was the subject.
Painted in the Xiantong years, in a cave endowed by his nephew. What did he do? The guide switched the torch off and on again. In 848 he took Shazhou back from the Tibetans. Then, flatly: this place had been held for sixty-odd years.
And then the thing I stayed for. He mentioned sending word to Chang'an. How long did the message take? Two years. I thought I had misheard. Two years. The corridor in between was still held; there was no road from Shazhou to Chang'an. Then how? He sent ten parties.
I caught him on the walkway afterwards. Ten sent, how many arrived? One. And the other nine? Nobody knows. The histories don't say. The one that got through — which road? North, round by the Tiande garrison — in what is now Inner Mongolia. Dunhuang to Chang'an direct is two thousand kilometres; north into Inner Mongolia and down again adds at least fifteen hundred. So the one that detoured furthest was the one that arrived? Yes. And, packing up his equipment as he left: the short roads all had someone watching them.
In the afternoon I told Chang I was going to find a second supplier. By any ordinary standard you do not announce this; you arrange it and the other party finds out when it is done, and saying it in advance is saying you do not trust him. But Qin in Jiayuguan two days ago: a buffer hidden is an error; a buffer in plain sight is the calculation. So I said it.
He was shelling jujubes in the sun and did not look up. Say it. I need a second house. His hands stopped for a second or two, then went on. Why? Because in March I have only you, and if anything happens to you that month I have nothing. Nothing will happen to me. I know. And then the part that was hard to say: but you told me yourself that the east end and the west end of one house come out two grades apart.
He put the jujubes down. He was not angry. You're right. Then: but there is one thing you do my way. I find the second house. Why you? You wouldn't know what he's worth. And then the sentence I have been turning over since: the second house you find yourself will undercut me. The one I find won't.
Which is exact. A stranger prices below him to take the order; he matches next year; the stranger cuts again the year after; in two seasons I have saved money and destroyed both of them. A man he knows will not knife a man he knows for one order.
So what is your relationship with the house you have in mind? Not close. Not close? Too close is no good. And the reason, entirely practical: if we're too close, I can't help him when he's in trouble and he can't cover me when I am. Two houses next to each other, one fire takes them both.
Zhang Yichao came up the dirt track outside at dusk, in the red robe from the wall, except close up the robe is old and worn through at the cuffs, and he is not large; a man of fifty with a very dark face. You asked the same question several times in the cave today. Ask it, then.
Why ten parties? Because one cannot get through. Two? Two cannot get through either. He looked east. I worked it out. How? There are six stretches in between and every one of them is watched. A single party's chance of passing one stretch is two or three in ten. A quarter, six times over. I said: then ten parties cannot get through either. No. And then, and I stopped where I stood: which is why the ten do not take the same road.
Ten on one road face the same six stretches. Ten on ten roads face different ones; some short, some long. The short ones are easier to intercept. The short ones are certain to be intercepted. The men watching are not fools. What they watch is the short road. Then why send anyone by the short road?
*He looked at me. The parties on the short roads are there to give the watchers something to do.
Did those men know? No. Did you? I did. I did not ask anything further. After a while he said it himself: I did this once in my life. I don't want to say it was right. And then: but the maps and registers of eleven prefectures had to reach someone.
Last question, about my own two hundred bags. How many stages? Dispatch, repacking, group leader: three. Which stage, if it breaks, leaves you with nothing in March? Any of them. Then you are one party. Two hundred is not many. Split the shipment. Not so that it arrives on the same day. When the first lot arrives you know whether the road is open. And: don't make the first lot two hundred. How many? Thirty. — M.
附录二·Marco's Memo
Memo to file. Dunhuang, 20 November 2027. Volume three, night thirteen.
1. The historical case, and its arithmetic. Shazhou retaken 848; the despatch reporting it and delivering the registers of eleven prefectures took two years to reach Chang'an; ten parties sent, one arrived, via a northern detour through the Tiande garrison adding perhaps 1,500 km to a 2,000 km journey. The most indirect route was the surviving route, because the direct ones were watched.
2. The commander's own model, which is a real one. Six contested stretches; per-party survival per stretch ~0.25; so a single party's end-to-end probability is 0.25⁶ ≈ 0.02%. Ten parties on the same road: still ~0.02% — the failures are perfectly correlated, so replication buys nothing. Which is why the ten did not take the same road. Ten different routings decorrelate the failures, and the long routes carry low interception risk at the cost of duration. Note for my own use: redundancy without decorrelation is not redundancy. Three group leaders sourced from the same WeChat community, fed by the same warehouse, on the same courier, are one group leader.
3. The part I did not want to write down. Some of the ten were sent on the short roads to occupy the watchers, and they did not know it. He volunteered this, said he had done it once and would not call it right, and then said the registers had to reach someone. I record it because I do not think I am entitled to take the useful half of a man's account and leave out the half that cost him something.
4. Second supplier: disclosed rather than arranged quietly. Following ch.127, I told Chang I was looking for one, and why, including the awkward reason (his own admission that east-end and west-end output differ by two grades). His response is the commercially most sophisticated thing anyone has said to me on this trip: I find the second house. Because a second house you find yourself will undercut me; one I find won't. The mechanism: an unconnected entrant prices below the incumbent to win the order → incumbent matches next season → entrant cuts again → in two seasons the buyer has saved money and rendered both suppliers unviable. An introduced second source is a hedge. An unintroduced one is the first move of a price war I did not intend to start and cannot stop. Filed as a rule: let the incumbent choose the backup, and pay the small premium that comes with it.
5. And the correct distance between two suppliers. Not close. Too close is no good: if we're too close I can't help him and he can't cover me. Two houses next to each other, one fire takes them both. So the selection criterion for a second source is not "unrelated" and not "trusted friend of the first" — it is known to the incumbent, and exposed to different weather, different labour, different local conditions. Correlation of the failure, again, not the identity of the party.
6. Shipping plan rewritten. Was: 200 bags, one consignment, before 10 March. Now: — 20 Feb: 30 bags. Not for sale. To Jiang, the three leaders, and me. Eaten. This tests the road and the batch at once (ch.129's twenty-days protocol, run against three months of ambient storage). — 5 March: 70 bags. First offer to the groups. — 15 March: 100 bags. Conditional on the response to the second. Incremental cost of splitting: courier rises 1.20 → 1.50 per bag = ¥60 on two hundred bags.
7. The pricing principle, which I have had backwards for two years. ¥60 looks expensive against a consignment grossing a few thousand. It is not being bought against that. The cost of redundancy is priced against the cost of the goods not arriving, not against the profit if they do. If March is empty, Jiang loses 370 members and spends ¥11,100 replacing them, and I lose the standing I have spent thirteen days acquiring. Sixty yuan against eleven thousand one hundred. I have spent this entire trip grinding 0.19 off freight and 0.0074 off batch-running. The single best-value line item I have found costs sixty yuan and I nearly did not spend it.
8. The Turpan introduction, and what it reveals. Chang sent one surname, one place, one number — the third time on this trip information has arrived in exactly that format. The man runs over a hundred drying houses against Chang's six. Why would he deal at my volume? Long pause, then: he isn't after volume. He wants somebody who comes for goods every March. Which is my own search inverted. I am looking for someone who has something to send in March. He is looking for someone who comes in March. Same problem from opposite ends, three thousand kilometres apart, and neither of us can see the other from where we stand. This is the second time in four days I have found a two-sided idle-capacity match (ch.129: his empty July, my empty March) and both times the parties were invisible to each other. Working hypothesis for the whole venture: the value I add is not moving fruit. It is being the only party who has stood in both places. And: come to Turpan in July.
9. The admission at the end of the day. Zhang Yichao's line about the short roads made me look at my own nine segments and count how many I know will break. The Chengdu segment will break: Jiang alone cannot absorb 40,000 jin, and I have known that since Wuwei. Thirteen days in the west, grinding costs, and I have not touched it. Because the western segments have numbers in them and Chengdu has one person in it. — I have been repairing the segment that is easiest to repair, and calling it work.
附录三·English Highlight
"Ten parties of envoys were despatched by hidden roads. Only one reached the Tiande garrison, and two years passed before it arrived at Chang'an."
Shazhou had been held for sixty years and was taken back in 848. The report of it, together with the maps and land registers of eleven prefectures, took two years to reach the capital, because the corridor in between was still in other hands and there was no road.
Ten parties were sent. One arrived. What became of the other nine is not recorded anywhere. The one that got through went north, a detour of perhaps fifteen hundred kilometres onto a two-thousand-kilometre journey — and the guide's parting remark was that the short roads all had somebody watching them.
The commander, asked why ten, gave a model rather than a sentiment. Six contested stretches; a party's chance of passing any one of them about two or three in ten. A quarter, six times over, is two hundredths of one percent. Then ten parties cannot get through either. No.
Which is why the ten did not take the same road.
Ten copies of a message sent down one road face one set of dangers, and duplicating them buys nothing at all, because they fail together. Ten roads fail separately. Some are short and quick and certain to be intercepted; some are long and slow and unwatched. Redundancy without independence is not redundancy.
Then why send anyone at all down the short roads?
The parties on the short roads are there to give the watchers something to do.
Did those men know? No. Did you? I did. And after a silence, unprompted: I did this once in my life. I don't want to say it was right. And then: but the maps and registers of eleven prefectures had to reach someone.
The commercial version cost sixty yuan. A first consignment of two hundred bags, one shipment, one road, three stages, any one of which failing leaves the buyer with nothing in the month that matters — split into thirty, seventy and a hundred across three weeks, at a courier premium of thirty fen a bag.
Sixty yuan looks absurd next to a shipment worth several thousand, until you notice it is not being priced against that. The cost of redundancy is priced against the cost of the goods not arriving, and never against the profit if they do.
附录四·Dongpo's Cheat Sheet
| 敦煌第四天 | 出处 | 现代对应 | 一句话 |
|---|---|---|---|
| 第一五六窟 | 特窟一百五 | 《张议潮统军出行图》 | 手电扫十几米才到中间那个人 |
| 他不比别人大 | 红袍,队伍里一个人 | 不打光看不出是主角 | |
| 报信要两年 | 中间还在吐蕃手里 | 沙州到长安路不通 | 他派了十路 |
| 十路到了一路 | 其余九路史书没写 | 到的那一路绕天德军 | 近的那几条都有人守着 |
| 六段,每段过二三成 | 张议潮自己算的 | 0.25的六次方 | 十路走同一条道,等于一路 |
| 冗余不解耦,等于没有冗余 | 由上一条推出 | 三个团长同一个仓同一个快递 | 那就是一个团长 |
| 走近道的是让守的人有事干 | 本回最重的一句 | 他们不知道 | 我知道。我不想说它对 |
| 我要找第二家 | 马可主动说的 | 按常理不说 | 余量藏起来就是错(定城砖) |
| 第二家我给你找 | 常玉山 | 你自己找的不知道他斤两 | 你自己找的会跟我压价,我找的不会 |
| 两年下来两家都做不下去 | 挖价的连锁 | 省了钱,供货没了 | 引荐来的备份是对冲,自己挖的是价格战 |
| 太熟的不行 | 他家出事我帮不上 | 两家挨着的房,一场火全没了 | 选备份看的是「一起坏不坏」 |
| 三十/七十/一百 | 分三批 | 头三十袋不卖,自己人吃 | 试路,也试这一批放三个月变不变 |
| 多花六十块 | 一袋一块二涨到一块五 | 对着一万一千一 | 冗余的钱要跟不到货的代价比 |
| 一个姓一个地方一串号 | 这一路第三次这个格式 | 吐鲁番,一百多座房 | 他不为量 |
| 他要一个每年三月都来拿货的人 | 常玉山转述 | 马可要一个三月有货发的人 | 同一件事的两头,隔三千公里 |
| 七月你来一趟吐鲁番 | 八个字 | ||
| 我一直在修最好修的那一段 | 当天最后一行 | 因为那一段有数可算 | 成都那一段只有一个人 |