Corruption in Classical Chinese Astrology Ebooks Follows Character Shape

Most people who download a classical Chinese astrology text as an ebook have no way to judge whether the file is any good. The usual test is how it looks. Clean typesetting, punctuation added by an editor, a few footnotes, and the file gets treated as a reliable copy.

That test points the wrong way. When I was turning the old formulas into working code, I loaded four of these texts into my own system and started checking them character by character. What came back was not a quality ranking. Corrupted characters do not scatter evenly through a book. They cluster around groups of characters that look alike, and a single book can have one such cluster completely destroyed while another cluster is untouched. So asking whether a book is good is the wrong question. The right question is whether the specific characters you need are broken in the copy you happen to have.

I tested one cluster: 己, 已, and 巳. Three characters that differ by a single stroke, and all three carry real meaning in this field. I looked for them inside three terms that cannot avoid them, 己土 (the sixth heavenly stem), 巳酉丑 (the metal trinity), and 己卯 (a sexagenary pair).

Text	Characters	Correct form	Corrupted form	Breakdown
Yuanhai Ziping (ctext Siku edition)	32640	22	0	none
Ditian Sui Chanwei (Wikisource edition)	143095	147	6	已土 ×5, 己酉丑 ×1
Ziwei Doushu Quanshu (Ming Nanyangtang block-printed edition)	64786	13	5	己酉丑 ×2, 已酉丑 ×2, 已卯 ×1
Ziwei Doushu Jielan (Jilin edited edition, RapidOCR)	93737	0	26	已酉丑 ×23, 已卯 ×2, 已土 ×1

The last row is the striking one. In that file the correct form never appears at all. Every one of the 26 occurrences is wrong.

But the number that actually carries the argument is the control. In the same four files I counted three terms that contain none of the three lookalike characters: 亥卯未, 申子辰, 丁壬. They appear 4, 1 and 0 times in Yuanhai Ziping; 9, 6 and 15 times in Ditian Sui; 16, 12 and 5 times in Quanshu; 19, 24 and 7 times in Jielan. Corruptions from this cluster in that control set: zero, in all four files.

So the damage is not a property of the book. It is a property of the character shape. One cluster can be wiped out in a file where a different set of terms comes through clean. That is why a summary judgment about a whole edition tells you nothing useful.

Here is what the damage looks like in the actual files. These are quoted as errors, not as sources.

In Ditian Sui: 「谓穴了向寺；辛日未时，未中已土，乃辛金之印」. The hidden stem inside 未 is 己土, the sixth stem. Written as 已土 it is a different character entirely. The same file has 「不知己酉丑全金局，则亥中甲木受伤」. The metal trinity is 巳酉丑. 己 is a heavenly stem and cannot appear inside a trinity of earthly branches at all.

Quanshu has the clearest single line: 「生在已酉丑、亥卯未六阴时」. Two trinities side by side on one line. 亥卯未 is correct. 巳酉丑 has been corrupted to 已酉丑. You do not need to know any astrology to see that something is inconsistent within a single line.

Jielan is blunter. The full list of ten heavenly stems reads 「甲木乙木丙火丁火戊土已土庚金辛金壬水癸水」, and the sixth position, which can only be 己土, is wrong. The sexagenary mnemonic reads 「戊寅已卯城头土，戊申已西大驿土」, where the correct forms are 戊寅己卯城头土 and 戊申己酉大驿土. The second half contains two separate errors in one short phrase, 己 written as 已 and 酉 written as 西.

Two boundaries I want to state plainly.

First, the 26 errors in Jielan come from OCR. That file is a modern edited edition that was put through character recognition software, so a machine misread the shapes. That is a different cause from a copyist's slip carried down through centuries of hand transmission. The causes have nothing in common. The consequence for you is identical: the character you search for is not the character in the file.

Second, my checker looked at exactly one cluster and three terms. Clusters I did not check are not thereby clean; I simply have no information about them. And this is the part worth borrowing regardless of what you are checking. Before a detector is allowed to report anything, run it against input you already know is clean and confirm it stays quiet. The control terms above are doing that job. Three terms without the cluster, zero reports. Without that step, a zero tells you nothing, because you cannot distinguish a clean file from a detector that never fires.

You can run the whole thing yourself in under a minute. Open the ebook, use the reader's search box three times, and search for 已土, then 已酉丑, then 己酉丑. In a sound copy all three searches return nothing. If any of them returns a hit, that cluster is broken in your file, and from then on every passage involving the stem 己 or the branch 巳 needs to be checked against a second copy. No knowledge of astrology is required.

Quoted source text functions as a guardrail in this field. A claim gets made, and a line from an old book is placed underneath it to hold it up. If the line itself has been corrupted somewhere between the woodblock and your download, the claim is resting on nothing. Checking the wording is not pedantry. It is the floor.