Eight Mansions and Half a Blank Page

One volume of the Xingxue Dacheng walks through the Moon's motion mansion by mansion. Four lines per mansion, always the same shape: the Moon enters such-and-such a mansion, receives such-and-such a title, then one verdict. It reaches Bi, stops, turns the page — and the next thing out is Zhen.

Eight mansions are missing in between. Zi, Shen, Jing, Gui, Liu, Xing, Zhang, Yi.

We did not lose them. That is worth stating carefully, because in the previous round we logged the question as unresolved: the transcription came from a public electronic text, the text is blank at that point, and we could not tell whether the blank was the text's fault or its source's.

Settling it required a photographic facsimile. We wrote that conclusion down, shelved it for a month, and then someone remembered an entry point almost nobody clicks.

1. Two hops inward

Beside the electronic text's table of contents there was a link to a shared-media category. Inside: thirty-two files. One of them carried the title "Facsimile Wenyuange Siku Quanshu, volume 0809".

The file was not the point. Its index page was. The raw source of that index page contains one line that states outright which page range this book occupies within the volume:

第0809冊　子部 術數類 命書相書之屬
==星學大成==
【明】萬民英撰
<pagelist from=290 to=876 290=285 />

Which is to say: the book runs from page 290 to page 876 of that volume, and printed page = scanned page − 5. Check it against page 290, whose own printed footer reads "809-285", and the offset holds.

With a range in hand, the rest is just walking. Page 620 turns up the opening of juan 17; page 654 turns up the block heading of the Yuguan Ge in juan 19. The missing leaf is three pages later.

There it reads: the Moon parts from Bi, rain pours; body and life meeting it, desire runs high — and then the character 闕, "omitted". The rest of that leaf is blank. The next page opens with "the Moon traverses Zhen, titled Tianli".

So what is missing is not one sentence. It is eight lunar mansions, each of which should carry four lines plus a degree statement.

2. Three corrections, and one deliberate non-correction

Investigating this, we corrected three of our own statements. None of the three was a data error; all three were reading errors.

We called a manuscript a printed edition. The Siku Quanshu is a transcribed manuscript, hand-copied by scribes, not a woodblock print. We wrote "original print" throughout, treating a copyist's hand as if it had been cut into a block. That is a layman's mistake, and the kind that a specialist spots in one glance. The fix was easy: the whole document now says "the Siku Quanshu copy".

We counted the omissions one short. The working note said two. There are three: besides the half page in the Yuguan Ge, there is a small in-line 闕 in juan 19 where the Ziqi section discusses the life palace, and another small 闕 in juan 20, in the middle of the Yuebo calculation passage, just after "what does not divide evenly does not enter".

We treated three scans as two recensions. We hold three scans: one from the Wenyuange copy, where each image contains two half pages, and two from a library's volume-by-volume scan, where each image is a single half page. We once wrote that the latter two share layout and line count, therefore they are two digital captures of one edition — and on that basis claimed they had been cross-collated as two recensions.

That claim does not hold. On the same page — the missing leaf itself — all three agree line by line and character by character, including the position of the blank, so the two library scans cannot serve as independent recensions. That much we keep. But "one edition" is a claim about textual bibliography, and image evidence cannot carry it: the seven copies of the Siku Quanshu were produced to the same layout and line count anyway, so a matching line count distinguishes nothing. All that survives is a negative: we may not say the collation happened.

And there is one thing we deliberately did not correct.

Resolution has a ceiling. The source images top out at 1237 × 1864; a half page spans about six hundred pixels, and a body character about thirty pixels. Large characters, headings, margins and blanks read reliably. Small interlinear notes and rare variant forms do not.

So the two lines on the missing leaf are transcribed with the character 於 where the facsimile looks closer to 于 — two spellings of the same word — and with 逄 where it looks closer to 逢. We left it alone. At thirty pixels this is "we cannot read it", not "we found a variant". Those two are very far apart, and telling which one you are looking at matters more than deciding what the character is.

Our rule, then: first build a stroke baseline from a pair of undisputed variant spellings on clear pages of the same volume, then compare stroke by stroke; where the baseline cannot decide, not one character of the transcription changes, and the observation goes into the collation notes as "the facsimile is formally close to X; held in doubt". We have not built that baseline. So the note hangs in mid-air, and anyone may call it a guess. The facsimile is public; anyone can look. But promoting "cannot read" into "found a difference" is a different act altogether.

One structural observation, stated so it does not get over-extended: the Siku Quanshu copy sets annotations such as "this star is named Jianshan" in small type, and the electronic text flattened large and small type during transcription. Separating body text from annotation now requires going back to the images. Our transcription layer cannot do it. That is a known defect.

3. Why it is blank, we will not answer

This is the most important passage in the piece.

Only one thing is settled: the gap is a gap in the Siku Quanshu copy itself, and the book marks it 闕 in its own hand.

Why it is missing has two explanations, and we cannot rule out either.

The first: a scribe skipped a leaf and left the space blank to cover for it — the omission was caught on recheck, recopying a whole leaf was expensive, so a 闕 was written and the space left open. The second: the source itself was damaged — the copy the Siku editors worked from, most likely a Ming woodblock, was already missing a leaf there, and the editors faithfully left a gap.

Telling them apart means examining ink tone and brush flow on the original for a sudden break, or checking the paper for scraping and patching; or finding the Ming print and seeing whether the blank half page holds exactly the number of characters one leaf of the standard edition carries. We did neither.

So wherever we touch those eight mansions, the source must be found elsewhere, and must be labelled a parallel textual tradition rather than something this book says. One thing worth reporting: the Gujin Tushu Jicheng quotes the same Yuguan Ge, all twenty-eight mansions present. But collate it line by line against the twenty mansions the two share, and eighteen carry textual variants. That is not "the eight missing mansions restored". It is a separate branch, and we do not use it to fill the eight gaps.

There is a product consequence too, logged here as an obstacle: the third omission falls on a decisive step of the Yuebo calculation. So wherever Yuebo's historical method comes up, we may not claim to have reproduced this book's procedure.

Finally, one thing we did not manage, recorded so nobody searches twice. The National Library's public repository does hold the Wenjin Ge copy, which is the first-priority target for collation. But that step needs a human at a browser; our automated route does not get through. The other surviving copy is ruled out — it exists only inside a library's mini-program, and the publicly digitised fragments do not include this book.

If you hold images of the Ming print or of another copy, or if you see a character we have read wrongly, tell us. We will correct the article and say so.