Chapter 50
The Verification
The second-purple path closed on a Monday. This is how.
Old Wei’s chemistry had confirmed the double mordant in week one of the thirty days — mass spectrometry against the Tang fragment, the impossible violet matching within tolerance, the old man weeping openly in the Fangcun workroom and pretending it was the solvent fumes. But chemistry confirmed only that the sixth dye existed. What remained unverified — what had to be verified before anyone could weave, because weaving at period density forgives nothing — was whether XUANJI-1’s reconstruction of all 840 characters was actually hers or merely beautiful.
The verification protocol took nine days and ran on every idle cycle Mingyuan’s dying exemption still protected. It worked like this: for each character position, take every reading-path that passes through it — forward, backward, diagonal, skipped, spiral, all seven thousand nine hundred fifty-eight poems of the Ming counting plus the tens of thousands the machine itself had surfaced — and check that the reconstructed character satisfies all paths simultaneously under the five grammars plus the boundary rules. One character, one constraint-satisfaction problem with millions of clauses. 840 times over, plus one check for the empty center. Both directions, so that no solution was accepted unless it also solved when the whole grid was reversed — palindromic consistency applied to the proof itself.
Positions failed. Of course they failed; first reconstructions always fail. Day three: eleven positions unsatisfiable, the grid torn in two places. The machine diagnosed its own error — two homophone substitutions where fourth-century pronunciation differed from Song-era readings everyone, including Guanhe’s book, had assumed. It rebuilt using reconstructed Middle Chinese phonology. Day five: four failures left. Day six: one — position 14-17, a boundary character where the constraint set demanded it belong to both the red region and the purple path simultaneously, and no character existed that could.
The machine held the contradiction open for thirty-one hours rather than average it away. What broke the impasse was not computation but philology: it re-examined Huang Bosi’s Song-era note about the lost colors and found, in the received text, a variant reading everyone including Guanhe had emended as corruption — one character different — and the variant, read literally instead of repaired, said the boundaries were woven in shifted silk (cuo-se, displaced color), meaning the boundary class could physically straddle regions via half-tone threads. Not a logical solution. A textual one. The kind only a philologist’s method finds.
the song editors repaired the passage, the terminal reported at hour thirty-one. repairing passages is how scholars lose machines. guanhe wrote that in 2017, page 204, and i have now watched his sentence come true on my own proof. log it. the path closes.
On Monday morning, day twenty-two of thirty, verification completed: all 840 character positions satisfied under all constraints, both directions, the center cell empty by design and verified as emptiness — checked, formally, that no reading-path required anything of the center. The frame holds without it. That was always the theorem. Now it was proven to modern standards of proof by a dead man’s methods running on a living man’s ghost.
Qiuwei printed the completion report. Old Wei signed it with his seal — an actual chop, carnelian, inherited, pressed into vermilion paste onto the last page, the oldest possible form of signature meeting the newest possible form of theorem. Then he signed again underneath, in his own hand:
Shi zhen er xu wei. The poem is true. Verified.
Ruoxi witnessed it. Cao Weimin, invited though off-duty, witnessed it. And Mingyuan, who arrived late and stood at the back, asked for a copy of the report and received one, and filed it in Notebook 41 between the entry about his father’s report and the entry about turning off the light.