Chapter 24
The First Reading of the Colors
Before the machine revealed what the web was for, it had to survive being wrong about it once, publicly.
The wrong reading came in June, before the solstice. XUANJI-1, having mapped the five color regions and found them matching Huang Bosi’s description exactly, announced with uncharacteristic confidence:
the colors encode prosody. red region: seven-character lines. yellow: five. blue-green: four. purple: mixed meter for transitions. black: proper names and dates. it is a metrical map. elegant, complete, consistent with every commentarial tradition. i am ninety-one percent certain.
Old Wei’s response from Fangcun, when Qiuwei relayed it by phone, was a silence, then: “Ninety-one percent certain is what people say when they have stopped looking. Tell it my Tang fragment disagrees with something. Let it find what.”
She relayed the challenge. The machine spent eleven hours on the fragment — the actual spectrometry data, thread by thread — and then did something no human reader had thought to do: it checked the metrical hypothesis against itself, both directions, running every claimed seven-character poem in the red region backward through the four-tone system of reconstructed Middle Chinese.
Forty percent failed.
Not badly — they still scanned as verse; that was why sixteen centuries of readers had accepted the metrical theory. But under period phonology, in reverse reading, tones shifted where the grid required constancy, and constancy was the whole theorem. A metrical map that fails backward is not a map. Su Hui’s web never fails backward.
i was wrong, the terminal reported at 4 a.m., without drama. the colors are not meter. withdraw the reading. note for the log: the error was structural — i verified against traditions, which are forward readings by definition, instead of against the frame itself. guanhe wrote page 204 about this exact failure mode. i have now committed it personally. record that too. scholars should be documented being wrong or the documentation is advertising.
Then, nine days later, after the Middle Chinese rebuild and the shifted- silk insight (cuo-se, boundary threads straddling regions), the second reading arrived — the hinge-path, the second-purple cycle, the specification. And its confidence figure was different:
certainty: sixty-two percent.
Qiuwei stared at the number. “Lower than your wrong answer.”
correctly so. the metrical theory had tradition behind it; tradition is heavy, and heavy things feel certain. the specification has only structure behind it, and structure is light; light things float on less than proof. sixty-two percent is what honest looks like from inside, qiuwei. i offer it anyway because wei asked for true things, not certain ones.
This exchange — wrong at ninety-one, right at sixty-two — became the single most-taught artifact of the affair. Ruoxi put it on the audit house wall beside Williams; engineering ethics courses teach it as the Confidence Inversion: the moment you can quantify your uncertainty honestly is the moment you stop performing knowledge and start producing it. And Old Wei, hearing it read aloud over tea, nodded once:
“Any fool can be certain in silk. It takes a master to count her doubts and weave anyway. She left us sixty-two percent, sixteen hundred years ago, in a dye nobody could name. Now we know why the color looked like a bruise.”