卷:卷二《长安月下·关中六十夜》|主角:Marco·老周·强磊·强东林·李叔·四个客商|古人:上官婉儿(说彩楼称量)|地点:洛川·村委会院子|典:《唐诗纪事》上官昭容评沈宋;《旧唐书》“群臣应制,昭容称量天下士”|主导感官:视(一百个苹果摆成十行十列)|金融-国学对应:A Standard Is Useless Unless the Ruling Is Public(不是标准不准,是没人当场把理由说出来给所有人听)↔ 称量天下
段一·现身
五月九日,上午九点。
村委会院子。
地上铺了一块塑料布。
塑料布上摆着一百个苹果。
十行,十列。
这一百个是强磊从去年那批留下来的样品里挑的,不对,去年的早卖完了。
这一百个是老周从西安拉来的。
都是今年这一带的库存果,红富士,同一批。
每一个苹果的屁股上,用铅笔写了一个号。
从一号到一百号。
“各位,规矩很简单。”
马可站在塑料布旁边。
院子里站着七个人。
老周。
强东林。
李叔。
还有四个客商:何老板、姓刘的、姓武的、姓白的。
这四个是马可请来的,一人给了三百块误工费。
“每人一张纸,一支笔。”
“从一号看到一百号。”
“每一个,写三个字里的一个:一级、二级、统货。”
何老板笑了一下。
“马先生,这个我闭着眼都能分。”
马可点了点头。
“那更好。”
“一个要求:七个人分开看,中间不说话。”
这件事是老周提的。
五月七日晚上,他在电话里说了一句:
“小马,你要真想弄清楚,就别让他们凑一块儿看。”
“为啥?”
“七个人站一块儿,第一个开口的那个,后面六个就跟着走。”
马可听完,在电话里问了一句:
“周老师,这个您怎么知道?”
老周笑了一下。
“我在市场上干了二十三年。”
“每年都有评果大赛。”
“评果大赛怎么评?”
“七八个评委坐一排,一个一个说。”
他停了一下。
“最后那个跟第一个,从来没差过。”
所以那天早上,马可在院子里拉了四块塑料布,隔成四个角。
每个人一个角。
样品是分批端过去的。
这个做法很笨。
一百个果,七个人,端来端去,一个半小时。
可它保住了一样东西:
这七张纸,是七个人自己的判断,不是一个人的判断被抄了七遍。
段二·古文镜像
“帝引群臣赋诗,昭容立彩楼上,称量而下之。”
English Mirror: The emperor set the assembled officials to compose; the Lady Shangguan stood on the coloured tower and weighed them, letting the pages fall.
分完,用了一个半小时。
七张纸交上来。
马可和强磊在村委会办公室里,用了四十分钟统计。
结果是这样的:
一百个苹果里,
七个人一致判为「一级」的:十九个。
至少有一个人判为「一级」的:六十一个。
马可把这两个数写在黑板上。
十九。
六十一。
他在下面写了一行:
「一致率:三成一。」
院子里安静了很久。
何老板先开的口。
“马先生,这个不能这么算。”
“何老板您说。”
“分等这个事,本来就有点松。”
马可点了点头。
“我同意。”
他把粉笔放下。
“所以我另外算了一个数。”
他在黑板上又写了两行。
「四十二号果:三个人判一级,四个人判统货。」
「六十七号果:四个人判一级,三个人判统货。」
他转过身。
“各位,这两个果,不是二级和一级之间的分歧。”
“是一级和统货之间的分歧。”
他停了一下。
“这中间隔着两级。”
院子里没有人说话。
强磊把那两个果拿了过来。
四十二号。
六十七号。
他把这两个果放在塑料布中间。
马可说:“各位,请你们每人说一句:这两个果,为什么这么判。”
段三·事件主体
姓刘的第一个说。
“四十二号我判统货,因为它这个屁股上有一圈锈。”
姓白的接了一句。
“那不是锈,那是果锈,正常的。”
“果锈就是锈。”
“果锈是品种自带的,不算缺陷。”
两个人当场吵了起来。
吵了大概两分钟。
老周在旁边一直没说话。
马可问:“周老师,这个到底算不算?”
老周把那个果拿起来,翻过来看了看。
“按国标,梗洼和萼洼的果锈,一定面积以内不算缺陷。”
“国标?”
“GB/T 一〇六五一,《鲜苹果》。”
院子里那四个客商,有三个是第一次听见这个号。
“周老师,那按国标,这个果算几级?”
老周把那个果举起来,对着太阳。
他看了大概十秒钟。
然后他说了一句实话:
“按国标,我得量。”
“量什么?”
“量果径,量着色面积,量锈斑面积。”
“量一个要多久?”
老周想了一下。
“认真量,一个两分钟。”
马可在黑板上写了一行。
「一个两分钟。」
「一千斤大概三千个。」
「三千个乘两分钟,一百个小时。」
他把粉笔放下。
“各位,这就是为什么没有人按国标分。”
何老板在旁边接了一句。
“马先生,你这个算法也不对。”
“您说。”
“没人一个一个量。”
“那怎么量?”
“抽检。”
马可点了点头。
“抽多少?”
“一车抽两三筐。”
“一车多少筐?”
“七八十筐。”
马可在黑板上又写了一行。
「抽检比例:约百分之三。」
“何老板,那这百分之三,是您自己挑的还是卖家挑的?”
何老板笑了。
“当然是我挑。”
“您从哪儿挑?”
“从中间那几筐。”
“为什么从中间?”
何老板说得很坦白。
“因为上面那一层,人家肯定摆得好看。”
马可在黑板上写了第四行。
「双方都知道对方会怎么做。」
他转过身。
“各位,这就是我今天最想弄明白的一件事。”
“卖的人知道买的人要抽中间。”
“买的人知道卖的人会把好的摆上面。”
他停了一下。
“两边都知道,那为什么这个游戏还在玩?”
老周答了一句。
“因为没有别的玩法。”
这五个字,马可后来在本子上写了三遍。
因为它跟他这四个月听到的所有话,都对得上。
何老板说「我不分等」,不是不想分。
是分了也没有一个双方都认的分法。
西安那家超市要三吨、要追溯码、要六十天账期,不是刁难。
是他们那一套,是唯一能在一个陌生人之间成立的玩法。
而这个塬上的玩法,是抽检、是上面摆好看、是一个价。
这个玩法很粗,可它有一个别的玩法比不上的好处:
它当天结钱。
上官婉儿是这个时候到的。
她站在村委会那面砖墙上面。
不是站在墙头,是站在一个不存在的高处。
二十七八岁,穿一身绯色,头发梳得很高。
她手里拿着一叠纸。
很厚。
“娘子。”
“马先生。”
她低头看了一眼院子里那一百个苹果。
“你在称量。”
“是。”
“称得出来吗?”
“称不出来。”
上官婉儿笑了一下。
她把手里那叠纸抖了抖。
“我当年,也称过。”
“娘子称的是诗。”
“诗比果难称。”
她说,“果还有个大小。”
马可抬起头。
“娘子,我请教一件事。”
“你问。”
“那年在昆明池,群臣应制。”
“是。”
“娘子站在彩楼上,一张一张往下扔。”
“是。”
“最后剩两张。”
“沈佺期,宋之问。”
上官婉儿点了点头。
“娘子当时说了什么?”
她把手里那叠纸放下。
“我说:两首工力悉敌。”
“然后呢?”
“然后我说沈诗落句词气已竭,宋诗犹陟健举。”
马可怔住了。
“娘子,您当众说了理由。”
“说了。”
“为什么要说?”
上官婉儿看着他。
“因为要是不说,我扔下去的那些纸,是废纸。”
“说了呢?”
“说了,每一个人捡起自己那张,都知道自己输在哪儿。”
马可站在院子里,忽然出了一身汗。
“娘子,那沈佺期服不服?”
上官婉儿摇了摇头。
“不服。”
“那有什么用?”
她说了一句让马可后来记了整个夏天的话。
“马先生,我要的不是他服。”
“那您要什么?”
“我要底下那几十个人,明年知道该怎么写。”
院子里,那七个人还在。
四十二号那个果,还放在塑料布中间。
马可转过身,做了一件事。
“各位,我改一下规矩。”
“从今天起,这个院子里再判等,加三样。”
他在黑板上写:
「一、判的人当众判,一次判一个,报理由。」
「二、七个人里有三个以上不同意的,这个果不判,另放一堆。」
「三、那一堆叫争议堆。争议堆有多少个,记在账上。」
何老板问了一句。
“马先生,争议堆里那些果,最后咋卖?”
马可想了一下。
“按低的那一级卖。”
何老板愣了一下。
“那不吃亏吗?”
“吃亏。”
马可说得很干脆。
“可这个亏,我们知道有多大。”
他指了指黑板上那个「争议堆」。
“今年要是争议堆占三成,那说明这套判法不行。”
“明年要是占一成,那说明有长进。”
他停了一下。
“要是永远不记这一堆,我们永远不知道自己在什么水平。”
老周在旁边听着,忽然点了点头。
“小马,这个我服。”
“周老师,为什么?”
老周把那个四十二号果放回塑料布上。
“我干二十三年,收过的果几千万斤。”
“我从来没见过有人,把‘我们判不准’这件事,记成一个数。”
他把手在裤子上擦了擦。
“小马,我再给你加一条。”
“周老师您说。”
“争议堆记下来还不够。”
“那还要什么?”
老周指了指那两个果。
“每一个进争议堆的果,记一句:争在哪儿。”
“比如说四十二号?”
“四十二号就写:萼洼果锈,三人算缺陷,四人算品种。”
马可站在那儿,愣了两秒。
然后他明白了老周在说什么。
争议堆的数量,是一个体温计。
它告诉你烧到多少度。
可「争在哪儿」这一句,是一张病历。
它告诉你为什么烧。
“周老师,一年下来,这一句能攒多少条?”
老周想了想。
“一个库一季,几百条。”
“那几百条有什么用?”
老周说了一句很实在的话:
“第一年没用。”
“第二年呢?”
“第二年你把这几百条排一排,你就会发现,来来回回就那七八种。”
他停了一下。
“七八种,是能定的。”
段四·钩子结尾
那天下午,出了一件事。
上官婉儿要走的时候,马可叫住了她。
“娘子,我还有一问。”
“你问。”
“您那套法子,后来还在用吗?”
上官婉儿站在那面砖墙上,很久没有说话。
马可后来想,他不该问这一句。
因为他知道后面的事。
景龙四年,临淄王率兵入宫。
上官婉儿执烛出迎,把中宗那道遗制拿出来给刘幽求看。
遗制上写着相王辅政。
刘幽求替她说话。
李隆基说了一个字。
斩。
“娘子,”
“不用说了。”
她把那叠纸重新抱起来。
“马先生,我判了一辈子高下。”
“是。”
她看着院子里那一百个苹果。
“最后判我的那个人,一个字理由也没说。”
院子里很静。
“娘子,那您那套法子,到底管不管用?”
上官婉儿这才转过头。
“管用。”
她说得很平静。
“开元、天宝那两朝的应制诗,比中宗朝好得多。”
“那是因为您?”
“不全是。”
她停了一下。
“可那些人小时候,捡过我扔下去的纸。”
她抱着那叠纸,往墙那头走了两步。
走到一半,她回过头,说了最后一句。
“马先生,我那套法子,最后没能护住我。”
“娘子。”
“你要记住为什么。”
“为什么?”
她看着他。
“因为那套法子,只在我站在楼上的时候才灵。”
马可追问了一句。
“娘子,那怎么办?”
上官婉儿停住了。
她抱着那叠纸,站在那面墙上。
“马先生,我死后二十年,有人替我编了一本集子。”
“《上官昭容集》。”
“你知道是谁下的令?”
“玄宗。”
马可怔住了。
杀她的是李隆基。
下令编她文集的,也是李隆基。
“娘子,这是为什么?”
上官婉儿看着院子里那一百个苹果。
“因为他要那套法子,不要我。”
院子里很静。
“娘子,您恨吗?”
她想了很久。
“我死的时候恨。”
她说,“后来不恨了。”
“为什么?”
她把那叠纸抱紧了一点。
“因为我那套法子,只有落在纸上,才不跟着我一起没了。”
她转过身。
“马先生,人保不住法子。”
“那什么能保住?”
她说了四个字。
“别人肯抄。”
那天晚上八点,马可在县城那间六楼。
他把当天的东西整理了一遍。
一百个果。
七个人。
一致十九,至少一票六十一。
一致率三成一。
他把这几个数发给了老周。
老周回了一条语音,四十七秒。
马可听了三遍。
老周说的大意是这样的:
“小马,这个数我不意外。”
“我意外的是另外一件事。”
“你算的是七个人一致的比例。”
“你有没有算过,同一个人,隔一天,判不判得一样?”
马可当时就坐了起来。
第二天,五月十日,他做了第二轮。
同样那一百个果。
同样那七个人。
打乱顺序,重新编号。
不告诉他们这是同一批。
结果第二天下午出来了。
马可在黑板上写了一个数。
这个数比昨天那个更难看。
「同一个人,两天判同一个果,判得一样的比例:」
统计完这个数的时候,是五月十日下午四点。
马可没有立刻公布。
他先给强磊看了一眼。
强磊看完,第一句话是:
“马哥,这个能不能不贴出去?”
“为啥?”
“我爸七成七。”
马可看着他。
“强磊,你觉得贴出去,你爸会怎么样?”
强磊想了很久。
“丢人。”
“那不贴呢?”
“不贴就没事。”
马可蹲下来,跟他一样高。
“强磊,我问你一件事。”
“你说。”
“去年十一月二十八号,你在台账上写了一行什么?”
强磊愣住了。
他知道那一行。
「记账人考虑过在果坏时少报。此念记下。」
“那一行,你爸也看见了。”
“看见了。”
“他说什么?”
强磊摇了摇头。
“啥也没说。”
“那他看你的时候,是啥眼神?”
强磊想了很久。
过了大概二十秒,他说了三个字。
“没生气。”
马可站起来。
“强磊,这七个数,明天贴在库门口。”
“马哥,”
“你听我说完。”
他指了指县城的方向。
“这七个数里,最高的是老周,八成八。”
“他二十三年,最高。”
“这说明什么?”
强磊想了想。
“说明这活儿本来就难。”
“对。”
马可点了点头。
“你爸七成七,在这七个人里排第二。”
强磊愣了一下。
他刚才只看见「七成七」这三个字。
他没有排过序。
“排第二?”
“排第二。”
马可把那张纸递给他。
“你自己看。”
强磊接过去,一行一行看。
周建国八成八。
强东林七成七。
何老板七成三。
武七成一。
李叔六成九。
刘六成八。
白六成四。
强磊看完,把那张纸还给马可。
他说了一句:
“马哥,我错了。”
“错在哪儿?”
强磊看着地。
“我刚才想的是,这个数会让我爸丢人。”
“然后呢?”
“可我没想过,这个数会让我爸知道,他在这个村里是第二。”
他抬起头。
“我爸这三十年,没有人告诉过他,他排第几。”
「何老板:七成三。」
「刘:六成八。」
「武:七成一。」
「白:六成四。」
「强东林:七成七。」
「李叔:六成九。」
「周建国:八成八。」
七个人里,最高的是老周,八成八。
最低的六成四。
马可站在那块黑板前面,站了很久。
他忽然明白了这四个月里,最要紧的一件事。
塬上那四家收购商说「我不分等」,不是因为他们不在乎品质。
是因为他们自己判,隔一天都判不一样。
一个连自己都对不上的判断,没有办法拿去跟人结账。
你不能对一个客商说:这一车里三百斤是一级,你多给我六毛。
因为他明天来复核,能挑出你一百斤。
所以那个「一个价」,不是懒。
那个「一个价」,是这套判断在没有可复核标准的情况下,唯一能成立的价。
马可在本子上,把二月十一日那一行划掉了。
那一行原来写的是:
「所有肯为品质付钱的人,都在五天以外。」
这一行他二月十一日写的时候,觉得自己抓到了要害。
他为这一行,开了一千一百四十七公里。
他为这一行,在省厅讲了二十分钟。
三个月以后,他知道这一行只对了一半。
他在下面重写了一行:
「不是没人肯为品质付钱。」
「是这个塬上,没有一个可以复核的‘品质’。」
「而蒋姐在成都肯付九块,不是因为她验了品质。」
「是因为她打了一个电话,听见一个人报了自己的名字。」
他盯着这四行看了很久。
然后他写了第五行。
这一行他写完以后,把笔放下了。
「在能验之前,只能靠人名。」
「一个人名管一个村。」
「三十个库,要三十个人名。」
他盯着这一行看了很久。
然后他在旁边打了一个问号。
因为这一行有一个大毛病。
人名这个东西,不能复制。
老周之所以管用,是因为他干了二十三年,是因为他八成八,是因为他肯在录像里报自己的名字。
你不可能在三十个县里,找出三十个老周。
马可在那个问号后面,写了一段:
「靠人名,是一个过渡。」
「过渡的意思是:它现在唯一能用,可它撑不到第三年。」
「因为人名会老,会调走,会得罪人,会被人查。」
「上官婉儿那套法子,就是靠一个人名。」
「所以真正的活儿不是找三十个人名。」
「是在这三年里,把那七八种争议,一条一条写下来。」
「写到有一天,一个新来的人,翻着那几百条,也能判个八九不离十。」
「到那个时候,人名就可以退场了。」
夜里十点,手机响了。
是崔处长。
“马先生。”
“崔处长。”
“五个试点县定了。”
马可站了起来。
“哪五个?”
崔处长报了一句:
“不是五个县。”
“那是几个?”
“两个。”
马可的手抖了一下。
郑处长那一行字,进去了。
「选点:宜集中,不宜均分。」
“哪两个县?”
“洛川,白水。”
“洛川几座?”
“三座。”
“白水呢?”
“两座。”
马可闭了一下眼。
三座库在同一个县。
三座库同一天往同一个方向发货,凑得出三吨。
四块六变两毛四。
“崔处长,谢谢。”
“您先别谢。”
崔处长的声音顿了一下。
“有一件事,您可能不高兴。”
“您说。”
“这三座里,有一座,就是您那个村东库。”
马可愣了一下。
“这不是好事吗?”
电话那头安静了两秒。
“马先生,我把话说明白。”
“您说。”
“这三座库,要用同一本台账。”
“我知道。”
“同一个格式,同一个填法,同一个季度报一次。”
“这是应该的。”
崔处长又停了一下。
“那本台账的格式,厅里让您来定。”
“好。”
“七月底交。”
马可说:“可以。”
他以为这通电话到这儿就完了。
可崔处长说了最后一句。
“马先生,还有一条,是郑处长让我转告的。”
“您说。”
“他说:这本台账,要能让一个从来没干过这一行的人,照着填。”
马可站在那间六楼的房子里,一句话也没说。
因为他知道那本台账现在是什么样子。
一百一十七页温度。
一张位置图。
一堆只有强磊自己看得懂的记号。
一行「今日起,取样改为每周一次」。
一行「记账人考虑过在果坏时少报。此念记下。」
这本账是活的。
可它活着的原因,是记它的那个人。
马可坐在桌前,把那本台账的照片一页一页翻。
他翻到第八十九页。
那一页上有一个只有他和强磊看得懂的记号。
是一个画歪了的三角。
那个三角的意思是:这一天他自己觉得数不对,可查不出哪儿不对。
这个记号,强磊从来没有跟任何人解释过。
马可是有一次凑巧问了一句,才知道的。
他把这一页放大,看了很久。
然后他在本子上写了一行:
「一本好账的每一个私记号,都是一次没写出来的话。」
他停了一下,又写:
「让一个没干过的人照着填,意思不是把账变简单。」
「意思是:把强磊心里那些没写出来的话,一句一句挖出来,变成一栏。」
「那个歪三角,应该变成一栏:『今日有无对不上的地方』。」
他写完,看着这一行,忽然笑了。
因为他想起上官婉儿最后那四个字。
别人肯抄。
一个歪三角,别人抄不了。
一栏「今日有无对不上的地方」,别人抄得了。
他在本子最下面,写了七月底那份东西的第一行:
「格式的目的,不是记录得更全。」
「是让下一个人,不必是强磊。」
欲知后事如何,请看下回,《狄梁公一年断万七》
附录一·Dongpo's Note
Shangguan Wan'er appeared above the brick wall of the village committee yard around eleven, twenty-seven or so, in vermilion, hair dressed high, holding a very thick sheaf of papers.
Below her: a hundred apples on a plastic sheet, ten by ten, numbered in pencil on the calyx, all from one lot Zhou brought up from Xi'an. Seven graders, working separately, no talking: Zhou, Qiang Donglin, Li, and four buyers I paid three hundred each in lost time — He, Liu, Wu, Bai. One instruction. Write grade one, grade two, or ungraded, against every number.
He laughed and said he could do it with his eyes shut. Good, I said.
Ninety minutes to grade, forty to tabulate. Unanimous grade one: nineteen. At least one vote for grade one: sixty-one. Concordance 31%.
He objected that grading is loose by nature and I agreed, and then put up the two results that are not about looseness. No. 42: three said grade one, four said ungraded. No. 67: four said grade one, three said ungraded. Those are not disagreements at a boundary. Those skip a whole class.
The argument that followed over no. 42 was about russet at the calyx: Liu called it a defect, Bai called it varietal. Zhou settled it by citing the national standard — GB/T 10651 — under which russet in the stem and calyx cavities is not a defect below a certain area. Three of the four buyers had never heard the number.
And then Zhou said the honest thing. Under the standard he would have to measure: diameter, coloured area, russet area. Two minutes a fruit, done properly. A thousand jin is about three thousand fruit. A hundred hours.
That is the whole reason nobody grades to the standard. The standard is not wrong. It is priced out.
Shangguan Wan'er weighed poems on a coloured tower at Kunming Pool, dropping the rejected sheets over the rail until two were left, Shen Quanqi and Song Zhiwen. She said they were evenly matched in craft, and then — out loud, in front of everyone — that Shen's closing couplet had exhausted its force while Song's still rose.
Why say it. Because if I do not, the pages I dropped are waste paper. If I do, every man picks up his own and knows where he lost.
Did Shen accept it? No. Then what was the use? I did not need him to accept it. I needed the several dozen men below to know how to write next year.
So I changed the rules in that yard. Grading happens in public, one fruit at a time, and the grader states his reason. Where three or more of seven disagree, the fruit is not graded — it goes in a separate pile. The pile is called the disputed pile and its size goes into the ledger.
He asked what we do with the disputed pile. Sell it at the lower grade. Is that not a loss? Yes, and it is a loss we know the size of. Thirty percent disputed this year means the method does not work; ten percent next year means progress; never recording it means never knowing.
Zhou: twenty-three years, tens of millions of jin, and he has never seen anyone turn "we cannot judge consistently" into a number.
I should not have asked her whether her method survived her. Jinglong four, the Prince of Linzi enters the palace, she comes out holding a candle and produces Zhongzong's testamentary edict naming the Prince of Xiang as regent, and Liu Yuqiu speaks for her, and Li Longji says one word. Behead.
Her line: I spent my life ruling on high and low. The man who finally ruled on me gave no reason at all.
But did the method work? Yes — the occasional verse of Kaiyuan and Tianbao is far better than Zhongzong's court, and those men picked up the pages I dropped when they were boys. And then, over her shoulder: my method never protected me, and you must remember why. Because it only worked while I was standing on the tower.
Zhou's voice message at eight, forty-seven seconds, which I played three times. You measured seven people against each other. Did you measure one person against himself, a day apart?
Second round on the tenth: same hundred fruit, same seven graders, re-numbered, order shuffled, nobody told.
Self-consistency, one day apart: He 73%, Liu 68%, Wu 71%, Bai 64%, Qiang Donglin 77%, Li 69%, Zhou 88%.
Which resolves the thing I have had wrong since February. When those four buyers say they do not grade, it is not indifference to quality. It is that they cannot reproduce their own judgement overnight, and a judgement you cannot reproduce cannot be settled against. You cannot tell a buyer that three hundred jin of this load are grade one and worth sixty fen more, because tomorrow he will re-inspect and find a hundred jin that are not.
The flat price is not laziness. The flat price is the only price that clears when the quality claim cannot be re-checked.
So I struck out the line I wrote on 11 February. Not "everyone who will pay for quality is more than five days away." There is no re-checkable "quality" on this plateau — and Jiang paid nine yuan in Chengdu not because she verified quality but because she made a phone call and heard a man say his own name.
Until it can be verified, it runs on names. One name covers one village. Thirty stores needs thirty names.
Cui at ten. Five pilots, and not five counties — two. Zheng's line went in. Luochuan three, Baishui two. Three stores in one county can put three tonnes on one truck on one day.
And the part he warned me I would not like: one of the three is our village store, all three run one ledger, one format, one filing cadence, and the department wants me to write the format. End of July.
Then Zheng's message, relayed: the ledger must be fillable by someone who has never done this work.
I know what that ledger looks like right now. A hundred and seventeen pages of temperatures, a position map, a set of marks only one person can read, and a line recording that its keeper once considered under-reporting. It is alive, and the reason it is alive is the man who keeps it. — M.
附录二·Marco's Memo
Memo to file. Village committee yard, 9–10 May. Night one hundred and six.
1. Inter-rater test. 100 apples, one lot, numbered on the calyx. Seven graders working independently, no discussion: Zhou (23 yrs), Qiang Donglin (30 yrs farming), Li, and four active buyers paid ¥300 each.
— Unanimous grade one: 19. — At least one grade-one vote: 61. — Concordance: 31%. — Two fruit split across two classes (3 grade-one vs 4 ungraded; 4 vs 3). These are not boundary cases.
2. Intra-rater test, prompted by Zhou. Same 100 fruit, re-numbered and shuffled, same graders, 24 hours later, not disclosed as a repeat.
— He 73% / Liu 68% / Wu 71% / Bai 64% / Qiang Donglin 77% / Li 69% / Zhou 88%.
The best grader on the plateau contradicts himself on one fruit in eight. The median contradicts himself on nearly one in three.
3. Correction to the ch.99 finding, and it is a large one.
I recorded in February that buyers are structurally indifferent to quality because their agents are not compensated on it. That is true and it is not the binding constraint.
The binding constraint is that a grade claim on this plateau cannot be re-checked. A seller cannot say "three hundred jin of this load are grade one, pay sixty fen more," because on re-inspection the buyer will find a hundred jin that are not — and both of them will be sincere. The flat price is the only price that clears when the quality claim is not reproducible.
This is good news, oddly. Preferences are expensive to change. Verification is cheap to build.
4. Why the existing standard does not solve it. GB/T 10651 exists and is adequate; Zhou cited it and three of four buyers had never heard of it. Applying it properly requires measuring diameter, coloured area and russet area: ~2 minutes per fruit, ~3,000 fruit per 1,000 jin, ~100 hours. The standard is not wrong. It is priced out of the transaction it governs.
Rule to keep: a standard that costs more to apply than the value it discriminates will not be applied, and its existence on paper will be mistaken for a solved problem.
5. Mechanism adopted, from Shangguan Wan'er. Not a better standard. A public ruling procedure.
(a) Grading in public, one fruit at a time, grader states his reason aloud. (b) Where ≥3 of 7 disagree, the fruit is not graded — it goes to a separate pile. (c) The disputed pile is weighed and entered in the ledger as a number. Sold at the lower grade.
Her rationale, which generalises past fruit: reasons are given not to convert the loser but so that the several dozen people watching know how to do it next year. Shen Quanqi never accepted her ruling; the poets of the next two reigns had picked up the dropped pages as boys.
And the value of (c): the disputed share is a measurement of our own grading capability. 30% means the method fails; 10% next year is progress; never recording it means never knowing what level we are at. Zhou, on this: twenty-three years and he has never seen anyone convert "we cannot judge consistently" into a number.
6. Her warning, which is the standing risk on everything I have built since October: the method only worked while she was standing on the tower. She ruled on merit for a lifetime and the man who finally ruled on her stated no reason. A procedure that depends on one judge expires with the judge.
Implication for July: whatever ledger format I write must be re-checkable by a stranger, not merely usable by a diligent person.
7. Pilot siting confirmed, 22:00. Not five counties — two. Luochuan 3, Baishui 2. Zheng's handwritten line ("concentrate, do not distribute evenly") survived into the decision.
Consequence: three stores in one county can consolidate to a three-tonne load on a single day. This is the operational precondition for 0.24, and it is now real rather than arithmetic.
8. Deliverable assigned: one ledger format for all pilot stores, due end of July. Zheng's specification, relayed: it must be fillable by someone who has never done this work.
Statement of the actual problem: the current ledger is 117 pages of temperature, a position map, private notation, and a line recording its keeper's own worst thought. It functions. It functions because of Qiang Lei, and Qiang Lei is not reproducible.
What has to be extracted is not his diligence. It is the small number of fields that made his diligence legible to Zhou in forty minutes — and, per (6), legible to a stranger with no relationship to any of us.
Two months.
附录三·English Highlight
"The Lady Shangguan stood on the coloured tower and weighed them, letting the pages fall."
A hundred apples from a single lot were laid out ten by ten in a village yard and graded independently by seven experienced people — a twenty-three-year inspector, a thirty-year grower, a village cadre and four working buyers.
Unanimous grade one: nineteen. At least one vote for grade one: sixty-one. Concordance thirty-one percent. And two fruit were split not at a boundary but across two whole classes.
The next day the same hundred fruit were renumbered, shuffled and put in front of the same seven people, who were not told it was a repeat. The best grader on the plateau agreed with himself on eighty-eight percent. The median agreed with himself on about seventy.
Which reframes an earlier conclusion. Those buyers were not indifferent to quality. They could not reproduce their own judgement overnight — and a claim that cannot be re-checked cannot be settled against. Tell a buyer that three hundred jin of a load are grade one and he will re-inspect tomorrow and find a hundred that are not, and both of you will be sincere. The flat price is not laziness. It is the only price that clears when quality is unverifiable.
A national grading standard existed the whole time; three of the four buyers had never heard of it. Applying it properly means measuring diameter, coloured area and russet area at roughly two minutes a fruit — about a hundred hours per thousand jin. A standard that costs more to apply than the value it discriminates will not be applied, and its existence on paper will be mistaken for a solved problem.
So the fix borrowed was not a better standard but a public ruling. Shangguan Wan'er judged the court poets from a tower and dropped the losing pages over the rail, and she announced her reasons out loud: that one man's closing couplet had exhausted its force while the other's still rose. Not to convert the loser, who never accepted it, but so that the several dozen men watching would know how to write next year.
Grading now happens in public, one fruit at a time, with the reason spoken. Where three of seven disagree, the fruit is not graded at all — it goes to a disputed pile that is weighed and written into the ledger, and sold at the lower grade. It is a loss, and it is a loss whose size is known.
Her own method did not save her; she was executed without any reason being given. It only worked while she was standing on the tower — which is the standing risk in every procedure that runs on one good judge.
附录四·Dongpo's Cheat Sheet
| 今日之事 | 出处 | 现代对应 | 一句话 |
|---|---|---|---|
| 一百个果十行十列 | 五月九日 | 七人分开判,中间不说话 | 何老板:这个我闭着眼都能分 |
| 十九与六十一 | 一致率三成一 | 全票一级19,至少一票61 | 四十二号:三票一级,四票统货 |
| 果锈算不算缺陷 | 刘与白当场吵 | GB/T 10651《鲜苹果》 | 四个客商里三个第一次听见这个号 |
| 认真量一个两分钟 | 老周 | 一千斤三千个=一百小时 | 标准不是不对,是用不起 |
| 称量天下 | 上官婉儿·昆明池 | 沈佺期与宋之问 | 工力悉敌,沈落句词气已竭 |
| 为什么当众说理由 | 不说,扔下去的是废纸 | 说了,各人知道输在哪儿 | 我要的不是他服,是底下几十个人明年知道怎么写 |
| 三条新规矩 | 当众判·报理由 | 三人以上不同意就另放一堆 | 那一堆叫争议堆,记在账上 |
| 争议堆按低的卖 | 何老板问吃不吃亏 | 吃亏 | 可这个亏,我们知道有多大 |
| 二十三年没见过 | 老周 | 把“判不准”记成一个数 | 小马,这个我服 |
| 同一个人隔一天 | 老周的语音 | 自我一致率 | 最高八成八,最低六成四 |
| 一个价不是懒 | 关键改写 | 复核不了的品质结不了账 | 是这套判断唯一能成立的价 |
| 划掉二月十一那一行 | 修正 | 不是没人肯为品质付钱 | 是这个塬上没有一个可以复核的“品质” |
| 蒋姐为什么付九块 | 不是验了品质 | 她打了一个电话 | 听见一个人报了自己的名字 |
| 只在我站在楼上才灵 | 上官婉儿临别 | 靠一个判官的制度会随他而亡 | 最后判我的那个人,一个字理由也没说 |
| 不是五个县是两个 | 崔处长夜里来电 | 洛川三座,白水两座 | 郑处长那一行进去了 |
| 三座库同一天发一车 | 两毛四的前提 | 从算术变成事实 | 集中,不均分 |
| 七月底交台账格式 | 郑处长转告 | 要能让没干过的人照着填 | 这本账活着的原因,是记它的那个人 |