The Identifications Were Written On Afterwards

Articles 2 and 3 established that after 象34 the printed commentary stops naming people and starts hedging. This copy of the book does not leave it there. On five leaves, somewhere between 1915 and 2013, ink went onto the paper that was not printed with it.

The interesting part is not the names. It is the grammar.

Three leaves doing one thing

The ? marks a character that is on the page but cannot be resolved at this scan’s resolution. More on those below — they matter.

On 象36 and 象37 the annotation is a five-character declarative ending in 也. 「⋯也」 closes a classical judgment: X is Y. The subject is not written because the printed line has already supplied it — the woman, the head of state. A sentence that withholds a name has been read, and closed. 象38 skips the construction and gives the bare name of the event.

In each case the edition describes and withholds; the annotation names. And all three sit after 象34, on the far side of the line the commentator drew through his own work. These leaves carry annotation that completes exactly the identifications the printed commentary declined to make.

That claim survives every open question below — which glyph, how many hands, what instrument, what date — because it is a claim about form. 也 is legible. The personal names are not.

The pairing, on one leaf

What the 1915 edition prints on 象36:

疑一女子能定中原建都長安

It is suspected: a woman able to settle the Central Plain and establish a capital at Chang’an. No name, no reign, no date, and a 疑 in front of the whole thing.

And what is written beside it, in ink that is not the printed ink:

慈禧太后也

[She] is the Empress Dowager Cixi. Five characters, closing the printed sentence’s open subject.

The fourth character of that reading is not read from the page. It is restored from context, and the distinction is the reason this series exists. At this scan’s resolution one annotation character occupies roughly sixty by sixty real pixels. Four readers looking at that glyph proposed four different characters. What settles it is not the strokes but the frame: 慈禧太 is not a free-standing word — it is the first three characters of 慈禧太后 and of nothing else in use — and 也 requires a predicate noun phrase in front of it, which 太后 supplies and none of the stroke-based candidates did. Three independent constraints converge on one reading, so the corpus adopts it, and records in the same breath that it was determined by context and not by the strokes. The diplomatic reading of the page stays 慈禧太?也.

Three readings, three different strengths

It would be convenient if the five annotated leaves were five identifications. They are not, and flattening them would be the same error the annotation itself commits.

象38 is the strongest. Its four characters were read off the page — 八國聯軍, the Eight-Nation Alliance of 1900 — after the blue pixels were isolated and grouped into four distinct marks. Nothing is restored. Its record also carries something worth reading: the row was withdrawn in error and restored the same day, because a measurement that averaged across the full column width was measuring both inks at once and produced a confident number pointing the wrong way. The mistake is kept on the record rather than tidied away.

象36 is adopted from context, as above — one character restored, on grammar rather than strokes.

象37 is weaker still, and is not quoted here. The crop shows five characters ending in 也 — the same construction as 象36 — of which the first, fourth and fifth are legible and the second and third are not. A reign-era reading is the natural completion of what is legible, but unlike 象36 nothing excludes the alternatives: the frame constrains without forcing. So the corpus records it as proposed rather than adopted, and this article describes the annotation instead of printing a name. The annotation on 象37 carries more ink than any other leaf — five separate marks — and is the one an eye is most drawn to and least able to settle.

The other two are not identifications at all. 象39 carries a mark rather than a legible identification, and 象48 carries a character written over a printed one — an overwrite, on the leaf whose printed commentary already has a blank square in it where a surname should be (疑一朱姓與一□姓爭奪朝綱).

So of five annotated columns, three carry a reading record and two do not, and the three are not equally established. The corpus records each separately, with what it was determined by, at received-readings.json. Their attestation level there is physical — meaning the annotation is on the page, which is not the same as every character is confirmed. 象36’s emendation is recorded at 當作 with status adopted; 象37’s two at 疑作, proposed. 當作 asserts; 疑作 preserves the doubt. The corpus keeps them apart and so should any article using it.

How we know it was not printed with the page

This is the part that does not depend on reading anything. Printed ink and this ink separate by colour, and the comparison is made within each column — annotation bands against the printed characters standing beside them on the same leaf, so that scan lighting and paper tone cancel out.

Measured that way, on every one of the five leaves, every annotation band is between 11.8 and 41.9 blue-minus-red points bluer than the printed characters in its own column. The narrowest separation is on 象38, the widest on 象48, whose single mark is the most strongly coloured on any leaf. Those figures are band means; the corpus also reports an isolated-blue-pixel measurement for 象38 which runs a little higher, and the two methods should not be quoted as if they were one. The bands, the per-column printed reference and both methods are in pen-annotation.json.

There is a second, accidental witness. Both recognisers used to transcribe this book read the handwriting as though it were type, produced different garbage from it, and disagreed with each other — which is how the annotation was noticed in the first place. A machine failing in an unusual way is sometimes the most reliable instrument you have.

Three things this does not show

It does not show what wrote it. Colour separation establishes a different ink. It says nothing about the implement, and scan colour processing is a further confounder. Naming one would import a chronology the evidence does not carry, so this article writes “blue-ink handwritten annotation” and stops. For the same reason nothing here calls the printing anything more specific than printed ink — whether this edition is 鉛印 or 石印 is an open question in the corpus.

It does not show how many people wrote it. The measurement establishes five columns carrying handwriting and marks in the same blue-ink range. A range is not an identity, and one person, two, or five are all consistent with it. The number is undetermined and this article does not guess.

It does not show that modern belief in these identifications descends from these annotations. What it shows is narrower and firmer: these identifications are not printed here, and they were added to this copy afterwards. Where identifications recorded elsewhere came from is a question about publication history and reception that this corpus cannot answer, and nothing in it should be read as answering.

What is actually on the page

A printed line that says perhaps a woman, and handwriting beside it that says it is the Empress Dowager. The first is what the edition committed to. The second is what the page was later made to say — in a different ink, in the grammar of assertion, on leaves where the printed commentary had declined to assert.

Every crop above is a link. The scan is public, the measurements are public, and if one of these crops does not show what the sentence beside it claims, that is checkable in about ten seconds. Which is the standard the annotation itself was never held to.