An Audit of Forward References in the Zhiyan Zhai Commentary on the First Eighty Chapters of Dream of the Red Chamber: What Is Adopted and What Is Not
前八十回脂批前指审计：采与不采

Notes on the text

1. This is the English version of a paper first published in Chinese (前八十回脂批前指审计：采与不采, same concept DOI); it corresponds to the Chinese version 0.4. It is a rewriting for English-speaking readers, not a line-by-line translation: background has been added where an English reader needs it, while the argument, the tables, the counts and the item-by-item entries are unchanged. Where the two differ in substance, the Chinese is authoritative.
2. This is a pre-release. The audit will continue, and the entries will be updated as the reconstruction's plot-level choices change and as new checks are made. The paper says only as much as has been done: what has been checked is reported with its result, what has not been checked is marked as unchecked, and what is inferred is marked as inferred.
3. This paper is one of the supporting papers of the overview, People Have Qing (concept DOI 10.5281/zenodo.22985375; cited here as "the overview"). Table 2 in §8.2 of the overview summarizes this audit; where the two differ, this paper is authoritative, and the next version of the overview will be revised accordingly (Appendix 4). What the overview calls a "reverse audit" is, from this paper on, called a fulfilment check (§7).
4. "The reconstruction" means chapters 81–110 as written by the project, cited as "Rec. ch. n". The entries in this paper rest on the reconstruction's choices at the level of plot, not on any one round of its drafts; the version on which the fulfilment check was run is identified by checksum in Appendix 3.
5. Notes are numbered "chapter.sequence": 21.65 is the sixty-fifth bracketed passage 【】 in chapter 21 of the project base text. A note number must be read as text, not as a decimal (6.10 and 6.1 are different notes). Manuscript attributions and note types are given as marked in the project base text.
6. Section 6 gives the current entry for each note; §7.4 records how the fulfilment check changed old entries into new ones. The two should be read together.
7. The full audit table, the blind-breakdown and check records, the sample, and the run notes of each AI session are published with the paper (Appendix 1, Appendix 5). The data and the attachments are in Chinese; Appendix 1 gives English glosses for the fields and for every value they take.
8. All translations are the author's own: of the commentary, of the text of the first eighty chapters, and of the reconstruction. Unlike the overview, this paper does not use Hawkes's published translation for the first eighty chapters, because the audit turns on the wording of the project base text, which differs in places from the text Hawkes translated; readers can find each passage in Hawkes by its chapter. Chinese is given in parentheses where the argument turns on a word. Personal names are given in pinyin; some place names and epithets are translated, a few following Hawkes (for example Greensickness Peak and the "Won-Done Song").

---

1 Stance: The Commentary as Testimony

A certain class of notes in the Zhiyan Zhai commentary speaks of what happens after chapter 80: "the last thirty chapters," "the Prison Temple," "Wei Ruolan's archery ground," "letting go at the cliff's edge," "Jinghuan's Roster of Feeling," and remarks such as "this foreshadows her early death in the later text" or "when the family's affairs collapse." Continuations, the search for the lost ending (tanyi), and reconstructions have always treated them as clues to the second half of the book. This project has used them in the same way. What this paper sets out is how the project handles each such note, and why.

A word of background for readers new to the field. Dream of the Red Chamber circulated in the 1750s and 1760s as a manuscript of about eighty chapters, copied together with a running commentary written by readers close to the author, chief among them the figures known as Zhiyan Zhai ("Red Inkstone Studio") and Jihu Sou ("the Old Man of the Odd Tablet"). The surviving manuscripts, conventionally named by a cyclical year recorded on them, by a former owner or by a preface (Jiaxu, Jimao, Gengchen, Mengfu, Qixu, and the lost Jing manuscript, known only from transcribed notes), differ in the commentary they carry. Many notes point ahead to what happens later in the story, some to events within the eighty chapters and some to events the eighty chapters do not reach. This paper calls such notes forward references; its concern is chiefly with those that reach beyond chapter 80.

Three principles come first.

The commentary is testimony whose provenance must be kept. Some notes record what a commentator saw in the drafts before him; others are his inferences, judgements, wishes and hypotheses; and different notes need not rest on the same draft. The book was made, as its first chapter says, by "ten years of reading and revising, five rounds of additions and deletions," and the second half may well have had more than one design; there was also more than one commentator. When two notes disagree, it may be because two drafts differed, or because one commentator's inference differs from another's report; it does not follow that one of them is wrong. So this paper does not rule on the historical truth of any note, and does not call any note a "mistaken note" or "the commentary's own error."

What the audit records is the project's handling. For each note, the paper records only how the project treats it: adopted (fulfilled), adopted with adaptation (transposed fulfilment), half adopted (half fulfilled), read differently, not adopted, or reserved (kept without a ruling). What is recorded is the project's choice, not the truth of the note.

Where a note is not adopted, the paper states only where the divergence lies. The reason given for not adopting a note says only which of the project's choices it cannot go with. It may diverge from the theme the project has adopted: from the overview's "people have qing," and from the readings chosen by the criterion "persons are ends in themselves." It may simply be that a plot arrangement already fixed by the project cannot coexist with it, which is a creative choice. It may conflict with the text of the first eighty chapters, or two notes may be at odds so that only one can be taken. In no case does the paper say that the note is wrong.

Given these three principles, the value of the audit lies not in rulings but in leaving a trace: every forward reference the project has identified has a record of how it was handled, which the reader can check item by item, or, taking a different theme, re-rule item by item.

2 Why an Audit

When the project was writing the last thirty chapters, it first worked "from back to front": it went looking for commentary to support what had been written. That can show that what was written has support; it cannot show that everything with support was written. Nor did the notes that were not adopted leave any trace, so that one could not tell "examined and not adopted" from "never seen." The audit supplies that missing link: all the notes on the first eighty chapters are gone through, and every note that points to later events is given a record of how it was handled.

The audit only records; it does not revise the reconstruction. Whether the reconstruction is to be rewritten because of a note is for the author to rule on separately; where such a ruling leads to supplementary writing, the entry says that the fulfilment came after that writing (§6.4).

3 Base Text and Commentary

Base text. The text of the first eighty chapters and the commentary are taken from the collated electronic base text used by the project (hereafter "the project base text"), prepared from the PDF Honglou meng Zhipi ben wanmei ban (Perfect Edition of the Zhiyan Zhai Commentary Text) made by Jiangshange Yin (md5 06d87cd765c8ba133976722331b06b2b). In the base text each note is enclosed in the brackets 【】, and is attributed to the manuscript named there.[^1]

Counts. The project base text contains 4,099 bracketed passages 【】. Two are editorial notes by the compiler, and one is Qi Liaosheng's preface. The rest are commentary notes, together with other matter that the base text likewise encloses in brackets: the "general principles" (fanli) of the Jiaxu manuscript, the prefatory poem, the Mengfu chapter-head poems, the poems introduced by "the title says," and the like. With attributions merged by the rules in Appendix 1, the distribution is:

Manuscript attribution	Notes
Gengchen	1,922
Jiaxu	1,516
Mengfu	560
Jing	17
Jimao	13
Qixu	1
No attribution	68
Compiler's notes	2
Total	4,099

An earlier internal count in the project gave Gengchen 1,918, Jiaxu 1,495, Mengfu 560, no attribution 96, Jing 17, Jimao 12, Qixu 1, differing slightly from the table. The reason is that the labels in the base text are written inconsistently (for example "甲夹批" or "庚讲批"; several Mengfu notes are mislabelled 梦 for 蒙), and the earlier count did not merge them one by one. This paper follows the table above, because it comes from the same source as the data published with it.

Weighting by manuscript. The project treats the Jiaxu and Gengchen commentary as standard-weight, and the Mengfu, Qixu and Jing commentary as reference-weight. This is the weight the project gives to each manuscript's notes; it is not a judgement of the truth of any note or of the worth of any manuscript. Where a standard-weight and a reference-weight note disagree, one is chosen by the criterion, and each case is recorded (§5). The thirteen Jimao notes all fall in Tier 3, and this paper does not assign them a weight. Notes with no attribution are recorded as the base text labels them ("脂批," "该批" and so on). Two of them are in Tier 1 (6.49 and 42.6) and are given no weight; the attribution of both is to be checked against facsimiles (note 1).

4 Screening and Tiers

4.1 Keyword screening and three tiers

Candidates for forward reference were first separated from the 4,099 passages with a keyword list. The list developed in batches. The first Tier 1 list was built around "lost" (迷失), "no manuscript" (无稿), "Roster of Feeling" (情榜), "Prison Temple" (狱神庙), "archery ground" (射圃), "letting go" (撒手) and "after several tens of chapters" (数十回后). In the course of the audit it turned out that "great turning" (大过节), "great crux" (大关键), "later events" (后事), "together to the end" (同终始) and "the last thirty chapters" (后三十回) had been missed, and these were added, followed by "great episode" (大节目), "another day" (他日), "last age" (末世), "final outcome" (归结), "second half" (后半) and others; two homographs, 伏侍 and 伏待 ("to wait upon," which contain the character 伏, "foreshadow"), were excluded. The final list has 45 terms. Matching is by substring, on the whole text inside the brackets including the manuscript label (Appendix 2). Rerunning the final list over all the notes yields 244 forward-reference candidates.

The candidates fall into three tiers:

1. Tier 1 (hard forward references): notes that point directly to events after chapter 80, such as the lost drafts, the Roster of Feeling, the Prison Temple, the archery ground, letting go, "the last thirty chapters," "great crux." There are 25 items, numbered T1-01 to T1-25 (甲01 to 甲25 in the data), comprising 28 notes (some items combine two notes with the same wording or at the same place), all ruled on individually.
2. Tier 2 (forward-reference candidates): the other candidates, routed by where their reference lands.

- Tier 2a: notes whose reference lands within the first eighty chapters are recorded as answered within the first eighty chapters; only the note number is recorded, without individual argument. - Tier 2b: notes whose reference reaches beyond chapter 80 are ruled on individually. - Among the candidates there is also a group that are only appraisal, praise, remarks on technique or instructions for reading, and point to no later event; these are recorded as not forward references.

1. Tier 3 (the rest): 3,855 notes.

This is the initial division. Later checks and reviews changed the handling of some items (several Tier 2b notes, for example, were re-recorded as answered within the first eighty chapters). The data keep the initial division in the column "initial tier" and record the current ruling in the column "handling." The one exception is four notes, 1.73, 3.171, 7.41 and 52.4, initially placed in Tier 2b, whose reference was found during the fulfilment check to lie within the first eighty chapters; their initial tier was changed to Tier 2a (§7.2).

4.2 Sampling and supplementary screening of Tier 3

Tier 3 is not ruled on note by note; it is only tested for notes that the keyword list missed. A missed note counts only if it is a forward reference that reaches beyond chapter 80: the note must point to, or allow one to infer, an event after chapter 80. Notes that only lament without pointing to anything (such as "I sigh for Xiren," "it makes my nose sting") and notes whose reference lands within the first eighty chapters are not counted.

Sampling. With the fixed seed 5432, 382 notes, about one tenth, were drawn from Tier 3. The sampling program, the note numbers of the sample and the checksum of the original sample file are given in Appendix 2, so that anyone can reproduce the draw. To reduce the blind spots of the first-order keyword list, a second-order screen was set up along three axes, fate, time and lament, using terms different from those of the first-order list. The difference in terms only means that the two sets of rules differ; it does not make them independent in any statistical or semantic sense. The second-order screen hit 27 notes in the sample. Each was read, and two were confirmed as forward references reaching beyond chapter 80: 13.3 (Qin Keqing's dream visit, a note with missing characters) and 28.106 ("here the later writing on the two Baos is made plain"). The figure 2/382 is only the rate at which this screening procedure found such notes in the sample. The 355 notes that the screen did not hit were not ruled on individually at this step, so the figure cannot be used to estimate the rate of remaining missed notes in Tier 3 as a whole.

Three targeted rounds. The sampling exposed two gaps in the keyword list, so three targeted rounds of supplementary screening were run over the whole of Tier 3:

Round	Screen	Hits	Missed notes
1	"afterwards" (以后) standing alone (the first-order list had only 后之, 日后 and the like)	6	3: 20.36, 22.78, 28.106
2	notes with missing characters (containing "□," 缺 or 阙; the keyword may be among what is lost)	17	1: 13.3
3	words for rosters, registers and ruin (幻榜, 芳讳, 副册, 又副, 正册, 抄没, 败落, 树倒, 食尽, 鸟投林, 了局, 收缘)	8	4: 5.125, 6.48, 18.17, 22.97

The three rounds together found 8 missed notes; the two found by sampling were each caught again by the screening and are counted among the eight. Once placed, seven of the eight fell into groups already ruled on; 18.17 (which gives the three verdicts of the Tertiary Register as Qingwen, Xiren and Xiangling) conflicts with the text of chapter 5 and is recorded as N8 (§5). Round 3 originally also counted 13.14: "the words 'when the tree falls, the monkeys scatter' still sound in my ears; counting on my fingers, it is thirty-five years." What it laments is something the commentator himself heard thirty-five years before, not a later event in the book. By the definition above it laments without pointing; after the review of v0.2 it was ruled not a forward reference and is not counted as a missed note (§6.3). The screening also turned up 75.3, Gengchen, "Mid-Autumn poem missing; awaiting Xueqin," the manuscript's own marker that something was left unfinished; it is filed as a record and not ruled on.

The fourth round. While this paper was being written, the second-order screen's three axes were extended from the sample to the whole of Tier 3, and three further screens were added. Of the ten items found in the first three rounds (including 75.3, which is filed as a record; 13.14 among them was later ruled not a forward reference but remains an entered row), 7 were hit again by screen 4a (5.125, 13.3, 13.14, 20.36, 22.78, 22.97, 28.106), while the other three are not hit by any screen in this round; the hit counts in the table exclude these 7. Overlaps among the screens are removed in the total.

Screen	Method	Hits	Missed notes
4a	second-order screen, three axes, over all of Tier 3	316	12
4b	the single character 伏, "foreshadow" (excluding 伏侍, 伏待, 埋伏, 降伏 and similar compounds)	22	6
4c	兆 "omen," 后日 "a later day," 下半部 "the second half," 逝后 "after [her] passing," 不谬 "not mistaken"	25	6
4d	the notes under the lines of Zhen Shiyin's gloss on the "Won-Done Song," checked by position (chapter 1, 1.167 to 1.183)	16	11

The four screens together hit 366 notes and found 26 missed notes, with 2 more filed as records. Of the 26, three (18.114 to 18.116, the notes on the first three of the four plays Yuanchun chooses) had already been discussed under Tier 1 item T1-19 without their note numbers being listed separately; the other 23 had simply been overlooked. Among them are well-known forward references: 24.3, Jing, "foreshadows Brother Yun's chivalrous visit to the convent"; 24.61, Gengchen, "when the Rong mansion's affairs collapse later, this man will surely have a part to play"; 6.1, Jiaxu chapter-head note, "foreshadows the 'second visit' and the 'third visit,' and where Qiaojie ends up." There is 42.1, which had already been found while the overview was being written and was passed to this paper for entry (overview §6.3). And there is the group of notes under each line of the "Won-Done Song" gloss that name "such-and-such and her like." Version 0.1 of this paper said that after three rounds no new gap in fulfilment had appeared; this round overturned that statement: 76.3, "this is not reckoning Jia Jing; it is reckoning the time of [Jia] She's death," concerns something the reconstruction had not originally written, the death of Jia She (§6.3, §6.4).

The misses had three causes. The first-order list lacked the single character 伏 ("foreshadows Brother Yun's chivalrous visit to the convent," "foreshadows the fall of the Jia family") and also lacked 后来 "later," 后日 "a later day" and 兆 "omen." The second-order three axes had at first been run only over the sample. And the notes that decode prophetic lines ("Baochai, Xiangyun and their like") usually contain no keyword at all and can be found only by position. The procedures introduced as a result are listed in §8.

4.3 Regression reconciliation against the final list

The keyword list grew in batches, and each batch used a different list, so gaps could open between batches. For example, 1.94, a Jiaxu side note, "another Jinghuan appears, all at points of great crux": the first-batch list had no "great crux," and the later rescan list, which did have it, reported only new additions, so the note was entered in neither. At the close of the audit, therefore, the final keyword list was rerun over the whole body and reconciled number by number against the notes entered in the table; wherever the count entered and the count found by machine disagreed, the discrepancy was traced note by note. This reconciliation turned up two notes, 1.94 and 20.2, which have since been ruled on; the entered count and the machine count now both stand at 244 and match one to one. For notes with missing characters the keyword may lie in the lost text, which no machine screen can reach; these must be gone through by hand.

5 Handling and Routing

5.1 Categories of handling

Category	Meaning
fulfilled	the plot of the reconstruction satisfies what the note says
transposed fulfilment	the letter of the note is fulfilled but its referent is rewritten; or its timing or placing is adapted
half fulfilled	of the several layers of what the note says, some are fulfilled and the rest are not adopted
answered within the first eighty chapters	the "later text" the note speaks of already appears within the first eighty chapters; even where the reconstruction extends it further, it is not treated as a constraint on the last thirty chapters
not a forward reference	appraisal, praise, philological or factual annotation, remarks on technique, instructions for reading and the like, pointing to no later event
read differently	the note's understanding of a person, of a word, or of what a later event means differs from the project's; recorded as it stands, counted neither as fulfilled nor as not adopted
not adopted	the project does not adopt what the note says; the type is given
reserved	kept without a ruling

There are two further marks. Referent inferred: the note says only "a foreshadowing line," "foreshadows what follows," "a line laid a thousand li ahead," "an omen of dispersal," without saying what is foreshadowed, so that the referent is the project's inference. Such notes are recorded as "fulfilled (referent inferred)," with the ground of the inference stated; what is fulfilled is the project's inference, not anything the note says outright. Tier 1 items T1-02 and T1-04 are of this kind; the check showed only that the reconstruction satisfies the project's inference, not that the note admits only this reading. Negative: where a note states a negative proposition, of the kind "seen only this once" or "never again," and the reconstruction still does not write the event by the point where the matter is finally settled, the note is satisfied, and is recorded as "fulfilled (negative)," as with 18.128, 44.2 and 25.104.

There are also two destinations. Moved to paper 5: notes that decode prophetic lines are handled in paper 5 together with the lines they decode (for the boundary see §5.3). Filed as a record: notes recording the state of the base text or of the book's making (such as "Mid-Autumn poem missing; awaiting Xueqin"), which fall outside the scope of this audit; they are filed and left unruled.

"Fulfilled" is not evidence. "Fulfilled" records a fact about the entry. Most of the reconstruction was written according to the commentary, so fulfilment is a result of construction; the number of fulfilments is not evidence that the reconstruction is close to the original. The title of Rec. ch. 90 takes over directly the fourteen characters quoted in the chapter-head note on chapter 21 (T1-21); this is direct adoption, and still less is it evidence.

5.2 Three rules

1. The count is fixed; entries are layered. Each bracketed passage 【】 is one row, and the counts do not change. Where one note says several things, each layer is entered: the main handling records the layer that reaches beyond chapter 80, and the other layers are written in the column "routing and non-adoption" and in the explanation.
2. A negative proposition is judged where it is settled. Notes such as "never again," "seen only this once" or "though the matters were not carried out" count as fulfilled only at the point where the matter is finally settled. Where the first eighty chapters show only a first sign, or a temporary obstacle, this is recorded as a beginning within the first eighty chapters, and the note as a whole is not counted as answered.
3. Duplicated notes are entered in one place. Notes with the same wording in different manuscripts are entered at one place and marked "duplicate."

5.3 The boundary for notes that decode prophecy

Prophetic texts themselves (the verdicts, the songs, lantern riddles, poems read as prophecy, drinking-game lines, the "Won-Done Song" gloss) are handled in paper 5. Where a note, in its own words, states an event that can be checked (going to marry far away, becoming a nun, "never again," ruin, death, an elegy and the like), it is a forward reference and is entered in this paper, marked "copied to paper 5." Where a note only uses the word "prophecy" (谶), only says "foreshadows what follows" or "the future state," or only matches a prophetic line to a person or place ("such-and-such and her like," "the Xiaoxiang Lodge, the Purple Rue Studio and such places"), it moves to paper 5. The test can be put in one sentence: a note stays in this paper only if it clearly points to an identifiable person or family in the book and states an event that can be checked. The referent may be fixed by an existing passage of the text that the note quotes or clearly refers back to, and that passage need not come before the note. "When the tree falls, the monkeys scatter" is said by Qin Keqing in chapter 13 and refers to the Jia family, so 5.125 (on chapter 5) and 22.97, which quote the saying, both stay in this paper. The "who" can be fixed by a quoted phrase, but the "what" must be in the words of the note: 79.23, "it is precisely this kind of thing that comes to pass," has only a pronoun for its event, so it moves. A generic term does not count as a "who": 1.177, "speaking of the day after the parents' death," only explains when "the later day" is, and "parents" names no particular family, so it moves to paper 5 along with the prophetic lines. This is a working rule for dividing labour between two papers of the project, not a general definition of foreshadowing. Paper 5 takes over, as listed, the decoding notes that this paper leaves to it, and does not enter them again. Twelve notes are currently copied to paper 5: 1.168, 1.170, 5.125, 18.114 to 18.117, 18.128, 22.97, 22.103, 22.105 and 79.5.

5.4 Routing and non-adoption

Code	Name	Name in the original audit table
R1	not a forward reference	甲型
R2	answered within the first eighty chapters	乙型
R3	read differently	(new)
N4	divergence from the theme the project has adopted	丙型
N5	different plot arrangement: a plot the project has already fixed cannot coexist with what the note says; a creative choice, not a matter of theme	(new)
N6	original lost or deleted by the author; its original form is not reconstructed	丁型
N7	a reference-weight manuscript stands alone, with no standard-weight support (a matter of weight only, not of truth)	戊型
N8	conflicts with the text of the first eighty chapters; the text is followed	己型
N9	two notes at odds; one is taken	庚型

The codes follow the Chinese numbering: 分流一 to 分流三 ("routings" one to three) and 不采四 to 不采九 ("non-adoption" four to nine); the data use the Chinese names.

R1 to R3 are not refusals: a note that is not a forward reference has nothing to adopt; a note answered within the first eighty chapters needs no adoption; and "read differently" means two readings standing side by side. Only N4 to N9 are refusals, and none of them is "the note is mistaken." N4 carries the main weight of the stance in §1. N5 allows "the author chose to write it another way" to stop at a creative choice, without being raised every time into a divergence of theme. N7 has not been used anywhere in the table so far; its definition is kept on record. 18.17, recorded in the original audit table as "the commentary's own error," is now N8, and the reason given is only that it conflicts with the text of chapter 5 and the text is followed. One layer of T1-12, originally recorded as "internal conflict in the commentary," was first moved to N9; after the review of v0.3 it was re-recorded as N4, because N9 requires the words of two notes to conflict, while the words of 67.1 are about Xianglian (§6.1). N9 is now used only for one layer of 18.17: it places Ping'er in the Tertiary Register, which conflicts with 6.48's "presumably she too is in the Secondary Register"; the project takes 6.48 (§6.3; Addendum 003). Where two notes are at odds, the entry says only which side was taken, and does not judge the other side wrong. What the overview calls "six explicitly disclaimed items" are now divided among N4, N6 and R3 (Appendix 4).

6 Results

6.1 Tier 1

The current entries for the 25 items of Tier 1 are as follows:

Item	Note	Manuscript	Gist of the note	Handling	Rec. ch.
T1-01	1.185	Jiaxu upper-margin (Jing upper-margin the same)	the two words "let's go" are truly letting go at the cliff's edge	not a forward reference	—
T1-02	6.10	Jiaxu side note	this is the main vein several tens of chapters later; truly a line laid a thousand li ahead	fulfilled (referent inferred)	100, 101
T1-03	6.49	none (labelled 脂批)	only on seeing Jinghuan's Roster of Feeling will one know that my words were not mistaken	fulfilled	110
T1-04	7.40	Jiaxu double-line interlinear; upper-margin	an idle stroke, yet it sets a thread of the second half moving	fulfilled (referent inferred)	99, 104, 105
T1-05	8.140	Jiaxu upper-margin	according to Jinghuan's Roster of Feeling, Baoyu is "feeling for the unfeeling"	transposed fulfilment	110
T1-06	18.19	Gengchen upper-margin	only at the chapter's end (回末; see note 1), with Jinghuan's Roster of Feeling, will one know the names in the principal, secondary, further secondary, and third and fourth secondary registers	transposed fulfilment (full-roster layer N4)	110
T1-07	19.36	Gengchen double-line interlinear	in the latter part of the book, several tens of chapters later: "choking on pickled greens in the cold of winter, huddled in a torn felt on a snowy night"	fulfilled	102
T1-08	19.80	Gengchen double-line interlinear	later, in the Roster, one sees the epithets "Baoyu: feeling for the unfeeling," "Daiyu: feeling for those who feel"	transposed fulfilment	110
T1-09–11	20.20	Gengchen upper-margin	Qianxue's proper text comes only at the "Prison Temple"; "Hua Xiren, constant from beginning to end"; five or six drafts lost by a borrower	fulfilled (Qianxue and Xiren layers); lost-drafts layer N6	92, 93, 94, 106
T1-12	21.65	Gengchen double-line interlinear	had another man had Baochai for wife and Sheyue for maid, how could he cast them aside and become a monk?	transposed fulfilment ("cast them aside and become a monk" layer N4)	97, 105
T1-13	25.107	Jiaxu upper-margin	grieves that he cannot see the writing of Brother Jade's "letting go at the cliff's edge"	fulfilled	106
T1-14	26.24	Jiaxu upper-margin	a whole long chapter on Hongyu and Qianxue at the "Prison Temple," alas lost, no manuscript	fulfilled (lost-draft layer N6)	93, 94
T1-15	26.120, 26.153	Gengchen upper-margin; Jiaxu	alas, the writing of "Wei Ruolan at the archery ground" has no manuscript	fulfilled (lost-draft layer N6)	87
T1-16	27.79, 27.81	Jiaxu side note; Gengchen upper-margin	seen only in the "Prison Temple" chapter; this [commentator] had not seen the "confiscation," the "Prison Temple" and the rest	fulfilled	91, 92, 94
T1-17	31.3	Gengchen	the unicorn Ruolan wears at the archery ground several tens of chapters later is this very unicorn	fulfilled	87, 106
T1-18	42.6	none (labelled 该批)	only on the day of the meeting at the Prison Temple will one know that "meeting disaster she finds good fortune, meeting ill she turns it to good" was a line laid a thousand li ahead	half fulfilled (N5)	94, 99, 100
T1-19	18.117	Gengchen double-line interlinear	the plays she chooses foreshadow four events, the great turnings and great cruxes of the whole book	fulfilled	91, 95, 96
T1-20	13.8, 13.89	Jing; Jiaxu	her spirit entrusts Xifeng with two matters of the Jia family's future … though the matters were not carried out; ordered Qinxi to delete it	fulfilled (deleted-draft layer N6)	92, 105
T1-21	21.3, 21.4	Mengfu chapter-head	without having seen the last thirty chapters one still cannot see the subtlety of this … "Xue Baochai uses words to veil her admonition; Wang Xifeng, knowing fate, forces herself to be heroic"	fulfilled (direct adoption)	90
T1-22	20.45	Gengchen double-line interlinear	after Xiren marries, there is still one person beside Baoyu and Baochai	half fulfilled	97
T1-23	28.1, 28.110	Gengchen; Jiaxu	[Qiguan] in the later chapters, with Xiren, looks after Brother Jade and Baochai, together to the end	half fulfilled (N4)	81, 94, 97, 106
T1-24	41.14	Jing upper-margin	another day, at the Guazhou ferry, a guest offers exhortation and warning; a rosy face is indeed able not to yield to withered bones (fragmentary; a collated reading)	fulfilled	100, 105, 110
T1-25	67.1	Jing	the "letting go" of the last chapter … why then shave the head? At Greensickness Peak the bond is ended and feeling attested; still not beyond Shiyin's dream	transposed fulfilment	105, 106, 110

Total: fulfilled 15 items (of which referent inferred, 2: T1-02 and T1-04), transposed fulfilment 5, half fulfilled 3, not a forward reference 1 (T1-01), not adopted 1 (T1-11, the lost drafts).

Some items need comment:

• T1-02: "some slight connection; this is the main vein several tens of chapters later." The note says that Grannie Liu's slight connection with the family is the main vein several tens of chapters later, but not what event this is. The referent (Grannie Liu and her party rescue Qiaojie, Rec. chs. 100 and 101) is the project's inference; recorded as fulfilled (referent inferred).
• T1-03: 6.49, "only on seeing Jinghuan's Roster of Feeling will one know that my words were not mistaken"; "my words" are those of 6.48, "presumably she too is in the Secondary Register" (note 1). In the Roster of Feeling in Rec. ch. 110, Ping'er originally had a page, but that page came in a run of "remaining pages," and the text did not say which register it belonged to. The Tier 1 checker had already noted this doubt at the time: "the register is not marked; this layer can count only as not conflicting, not as confirmed" (Attachment 05); when the entry was upheld at the check, the doubt was not taken up. After the review of v0.3 a reviewer raised it again, and the author wrote a supplement to Rec. ch. 110 placing Ping'er's page in the Secondary Register (Addendum 003). The Secondary Register layer is recorded as fulfilled after supplementary writing, and is not treated retroactively as satisfied by the original draft.
• T1-04: the upper-margin note says only "a thread of the second half," without saying what event it is; the author reads it as Xichun and Zhineng ending together in a convent, recorded as fulfilled (referent inferred). The check showed that the reconstruction satisfies this inference, not that the note admits only this reading. In the first eighty chapters Zhineng is a nun of the Water Moon Convent; the reconstruction does not write her returning to lay life, so she remains one. In Rec. ch. 99 Xichun writes to Zhineng: "that joke I made with you, sister, back then: I mean it now," answering the joke she made in chapter 7; in Rec. chs. 104 and 105 the two are together in the Zhengjue Convent. The interlinear note at the same place has a further layer, "it also foreshadows what follows"; the Zhineng it foreshadows already has her own text in chapters 15 and 16, so this layer is R2.
• T1-06: the full-roster layer is N4, following from "the Roster of Feeling is alive," a principle held at the level of the criterion (overview §6.2): no single roster fixes a person. It is not a matter of plot arrangement.
• T1-08: the first part is a long appraisal of Baoyu and Daiyu, R1. At its end the note says "later, in the Roster, one sees the epithets 'Baoyu: feeling for the unfeeling,' 'Daiyu: feeling for those who feel,'" of the same kind as 8.140. The roster in Rec. ch. 110 has "Lin Daiyu: qingqing"; Baoyu is not on the roster, and qing bu qing falls in the chapter title, in the sharp shout in the Office of the Ill-Fated, and in Baoyu's own words, as with T1-05, so the item is recorded as transposed fulfilment. The project reads the meaning of the two epithets differently: the commentary takes qing bu qing as praise (8.140: "toward everything in the world that has neither knowledge nor awareness, he has a foolish feeling to attend to it"), while the project reads it as a sickness and reads qingqing as "respecting others as one respects oneself" (overview §§6.1, 9.2). That layer is R3.
• T1-12: the overview (§6.4) had already split this note into two propositions. The layer "having them and casting them aside" should, on the reconstruction's timeline, be "engaged, then cast aside": the engagement was fixed, Baoyu climbed over the wall and left in the fourth watch before dawn on the wedding day, and the status of "wife" was fixed retroactively, after he had gone; hence transposed fulfilment. The layer "cast them aside and become a monk" was originally recorded as N9, as at odds with 67.1's "why then shave the head?" (是何必削发). After the review of v0.3 a reviewer pointed out that N9 requires the words of two notes to conflict, while the words of 67.1 are about Xianglian (see T1-25); what conflicts with 21.65 is only the project's own inferential reading. The layer is now recorded as N4, the reason being the criterion: after his awakening Baoyu returns to the human world and lives as a lay disciple, without taking the tonsure (overview §§6.4, 7.2); the implication of 67.1 serves only as supporting evidence.
• T1-14, T1-15: the notes lament that the writing on the Prison Temple and the archery ground was lost, with no manuscript. The project has written both episodes afresh (Rec. chs. 93 and 94; Rec. ch. 87), and they are recorded as fulfilled. "Lost, no manuscript" is a record of the state of the book's making, not a specification for the reconstruction; the original form is not reconstructed, and this layer is N6, as with T1-11. The two blind breakdowns of 26.153 differed on the time reference of "the chapters before"; a supplementary check was made after review (§7.5).
• T1-18: in the reconstruction, the meeting at the Prison Temple is between Baoyu and Qianxue and Xiaohong, and has nothing to do with Qiaojie's "meeting disaster, she finds good fortune"; Qiaojie's misfortune and rescue both come later (Rec. chs. 99 and 100). The project fulfils only the layer of Qiaojie finding good fortune in disaster; the layer linking the meeting at the Prison Temple with Qiaojie is a different plot arrangement, N5. The base text labels this note 该批; a reviewer recalls that it is usually given as a Jing upper-margin note, in which case it is reference-weight; to be checked against facsimiles.
• T1-19: of the notes on the four plays, the one on The Peony Pavilion (18.117) contains "great turnings and great cruxes" and is among the candidates. The three on A Cup of Snow, The Palace of Eternal Youth and The Dream of Handan (18.114 to 18.116) contain no keyword; they were entered by note number in the fourth round and assigned to this item (§6.3). All four events take place, but the four play titles hardly appear in the reconstruction: what is foreshadowed is the event, not the play's name.
• T1-20: "though the matters were not carried out" is settled in two places: in Rec. ch. 92 Xifeng, ill, repeats the two matters of the dream and says of herself, "I never bought a single mu of land"; in Rec. ch. 105, in Baoyu's dream, Qin Keqing says, "She didn't listen; I don't blame her." The passages on the "hairpin left behind" and "changing clothes" are found only in the Jing note 13.8 (reference-weight) and were deleted by the author; N6.
• T1-22: three layers. The layer of Sheyue beside them is fulfilled (Rec. ch. 97). The Baochai layer lands only on her status: the wedding was never completed, for the same reason as 20.72, and it is N4. The layer "the quarrel over the dropped fan" is already answered in chapter 31, R2.
• T1-23: the layer "looks after Brother Jade" is fulfilled. The layer "Baochai, together to the end" is N4: Baochai and Baoyu do not end together as husband and wife, which follows from the project's choice of theme, for the same reason as 20.72 and the Baochai layer of T1-22 (this is the reason for non-adoption recorded in overview §9.1). Whether it is at odds with 21.65 is left as a matter of reading: looking after them first and being cast aside later can also be read together. 28.110 has a further layer, "Baoyu forgets himself in feeling, and shows it before Baochai; this leads into the many later moments of forgetting himself": in chapter 30 Baoyu, watching Lingguan write the character 蔷, "without knowing it, stood there entranced as well"; in chapter 32 he is "lost in thought" and, taking hold of Xiren, pours out his heart; in chapter 57 he "went into a daze … for a moment his soul lost its hold." R2.
• T1-24: the punctuation of the fragmentary text follows the collated reading described in overview §4.6; on other readings the reconstruction does not satisfy this note. The fulfilment of this item depends on one collated reading, and is recorded as such.
• T1-25: letting go comes in Rec. ch. 106, not in the last chapter; transposed fulfilment (the overview, §9.1, had already recorded this as a soft failure). "Still not beyond Shiyin's dream" refers to the frame set up in Shiyin's dream in chapter 1, in which the monk and the Taoist take the stone down into the world to settle a case; it lands in Rec. ch. 110, where the monk and the Taoist carry the jade back to Greensickness Peak. The sentence "此虽眷恋，却破迷关，是何必削发" ("though this one was attached, he broke through the barrier of delusion; why then shave the head?") has two readings: (a) "this one" is Xianglian: though attached, he had broken through the barrier of delusion, so why shave his head; (b) the whole sentence is about the one who lets go in the last chapter: he is already awakened, and so need not shave his head. The note stands at the head of chapter 67, directly after Xianglian cuts off his hair with his sword in chapter 66, and its words are about Xianglian; the first two clauses are R2 (answered in chapter 66). The project takes the implication: awakening does not lie in shaving the head, and the one who lets go in the last chapter need not do so; this lands in Rec. ch. 105, where Baoyu becomes a lay disciple and does not take the tonsure. This layer is the project's inferential reading, and serves only as supporting evidence for N4 in the "cast them aside and become a monk" layer of T1-12. The last phrase of the note, 中秋前引即三姐 (roughly, "the Mid-Autumn lead-in is Third Sister"), has no settled reading and is reserved.

6.2 Tier 2

Tier 2 comprises 216 notes (the 244 candidates less the 28 notes of Tier 1). Their initial division and current rulings are:

Initial division	Notes
Tier 2a: referent within the first eighty chapters	112
not a forward reference	45
Tier 2b: referent beyond chapter 80	59

Current ruling	Notes
answered within the first eighty chapters	117
not a forward reference	45
fulfilled	9
fulfilled (negative)	2
fulfilled (referent inferred)	31
not adopted	1
moved to paper 5	10
reserved	1

The 59 notes of Tier 2b are further divided into three classes by what the note itself says (§7):

Class	Notes	Current ruling
Class 1: the note states a concrete later event	13	fulfilled 8; fulfilled (negative) 2 (18.128, 44.2); answered within the first eighty chapters 2 (44.12, 45.15); not adopted 1 (20.72)
Class 2: the note says only "foreshadowing" or "later text," the referent being inferred by the project	31	fulfilled (referent inferred); no fulfilment check
Class 3: the note only names a prophecy, or only matches prophetic lines to persons	10	moved to paper 5
Others	5	1.77 re-recorded as not a forward reference; 3.79 and 19.109 re-recorded as answered within the first eighty chapters; 79.5 brought back from Class 3 to this paper, fulfilled; 52.9 reserved

Some notes need comment:

• The 117 notes answered within the first eighty chapters include the 112 of Tier 2a, four of which were changed from an earlier entry of "fulfilled" during the fulfilment check. 1.73, "the Zhen Baoyu of later": in chapter 2 Yucun already talks about the boy in the Zhen household, and in chapter 56 the Zhen family's women say "the old lady calls him Baoyu," to which Grandmother Jia replies, "so he too is called Baoyu." 3.171 foreshadows Yuanyang through Yingge and Hupo through Zhenzhu; from chapter 20 on, both are already together in Grandmother Jia's rooms. 7.41, "Baoqin is supplied in the later text": Baoqin (抱琴, Yuanchun's maid, not Xue Baoqin 宝琴) appears in chapter 18. 52.4, "since Hongyu has had her ending, how could Zhui'er not be accounted for?": both events are within the first eighty chapters. There is also 38.1, a Mengfu chapter-head note that duplicates 37.1, "the poetry club comes from Tanchun's idea; the author has already foreshadowed the writing on promoting benefit and removing abuses," which lands in chapter 56; it was first recorded as not a forward reference and was re-recorded under rule 3. Then there are four Tier 2b notes, 44.12, 45.15, 3.79 and 19.109 (below).
• 18.128 (as she leaves, Yuanchun says, "if next year His Majesty's grace again permits a visit home," and a Gengchen interlinear note says, "this is the prophecy of 'never again'"): the layer "not next year" is answered within the first eighty chapters (there is no visit home in chapter 53; in chapter 55 "not only can there be no visit home," a temporary obstacle during the Dowager Consort's illness; and there is no further visit up to chapter 80). The layer "never again in her life" is settled only when Yuanchun dies in Rec. ch. 91. Under rule 2, the main handling is fulfilled (negative). The note states the answer to the prophecy in its own words, so under §5.3 it stays in this paper, copied to paper 5.
• 45.15 ("it also foreshadows what follows, and sets off the desolation of later"): both layers lie within the first eighty chapters. The layer of the night-time gambling is answered in chapter 73, when the gambling is discovered. As to the layer of desolation: the note says that people come and go in the garden all night and that it is "far from deserted." In chapter 78 the rooms at Hengwu Court have been "moved out, left empty and bare"; in chapter 79 Zilingzhou has only "a few old women on night duty," and Baoyu "took in so deserted and dreary a scene." That is the answer by contrast. Desolation is not a negative proposition, so rule 2 of §5.2 does not apply; following the precedent of 44.12, the note is recorded as answered within the first eighty chapters, with Rec. chs. 81, 99, 101 and 104 as extensions.
• 44.2 ("an idle remark that foreshadows what follows, and is saddening; what they call 'a great feast cannot come again'"): in chapters 75 and 76, at the Mid-Autumn Festival, "with four people fewer, it felt much more forlorn," and all present "could not help feeling desolate and lonely." But "cannot come again" is a negative proposition, which under rule 2 is judged where it is settled, and these two chapters are only a beginning within the first eighty chapters. The main handling is fulfilled (negative), settled at the raid on the mansions in Rec. ch. 91; Grandmother Jia's death in Rec. ch. 83 and Xifeng's straitened state in Rec. ch. 87 are the decline that comes before it. The version sent for the v0.3 review still recorded it only as fulfilled, which did not accord with §5.1; the mark was added after a reviewer pointed this out.
• 22.103 ("this is the prophecy of Tanchun marrying far away. Had she not gone far away, then when the family's affairs collapsed, its sons and grandsons would not have been scattered"): the layer of marrying far away is fulfilled (Rec. ch. 85), copied to paper 5. The layer "had she not gone far away" is the commentator's counterfactual and his hope. It does not assert what the reconstruction ought to write; it is an understanding of why the family fell. The project's narration gives no explanation in terms of cause and retribution; R3.
• 3.79 (in chapter 3 Lady Wang asks "have the monthly allowances been paid out?", and a Jiaxu side note says, "without seeing the later text, one cannot see the subtlety of this stroke"): in chapter 39 Ping'er has already let it out: "this month's allowances, my mistress drew long ago and has lent them out at interest." Re-recorded as answered within the first eighty chapters; Lady Xing's audit uncovering the moneylending in Rec. ch. 86 is the project's extension.
• 1.77 ("these eight characters describe the later Xiangling, and show that her roots are not common"): the eight characters describe Zhen Shiyin's wife, née Feng, "of gentle and virtuous nature, deeply versed in propriety," and point to the character of the later Xiangling, not to a later event; re-recorded as not a forward reference.
• 19.109 ("filling in the hardships of Xiren's childhood … naturally the first among the Tertiary Twelve"): it speaks of Xiren's place in the Tertiary Register, and is not a prophecy. Xiren's entry in the Tertiary Register is in chapter 5; re-recorded as answered within the first eighty chapters and assigned to the Roster group. "The first among" is the commentator's judgement, and is not taken as a specification of register placement.
• 4.63 ("in passing it narrates Yinglian's movements hitherto, and her final outcome afterwards"): "her final outcome afterwards" can be read as her purchase by Xue Pan, told in chapter 4 itself, or as her end in the book as a whole; both readings are entered.
• 35.26, a Mengfu chapter-end general comment, "this chapter preaches the Law through feeling, to awaken the world," entered the candidates because of the word "a thousand li." It points to no later event and is recorded as not a forward reference. Its reading of the whole book, a cautionary teaching in which feeling is a river in which one is bound to drown and death is the end, differs from the project's principle that the narrator gives no explanation in terms of retribution; that layer is R3. This is the fifth of the "disclaimed items" recorded in the overview.
• In Class 3, 37.54, "and again she cannot escape her own future state," comments on Xiangyun's poem on the white crab-apple; the earlier entry had put Tanchun, the wrong person, and this was corrected when the note was moved to paper 5.

6.3 Tier 3

The 3,855 notes of Tier 3 are not ruled on individually. The notes found by the four rounds of supplementary screening are entered as follows.

Eight notes from the first three rounds. Seven were assigned to groups already ruled on and given a handling according to the group: 13.3 to T1-20, fulfilled; 20.36, "and so later she takes over," to T1-22, fulfilled (Rec. ch. 97); 22.78, "never able to fall out of Jinghuan's illusory roster," to T1-05, transposed fulfilment; 6.48, "presumably she too is in the Secondary Register" (Ping'er): in Rec. ch. 110 Ping'er originally had a page but no register was marked, so the Secondary Register layer counted only as not conflicting; after the author's supplementary writing Ping'er's page falls in the Secondary Register (Addendum 003), and the note is recorded as fulfilled after supplementary writing; 28.106 to 45.14 (the later writing on Baochai), fulfilled (referent inferred); 5.125 and 22.97, the two "when the tree falls, the monkeys scatter" notes, to T1-19. The saying is spoken by Qin Keqing in chapter 13 and refers to the Jia family; both "who" and "what" are there, so under the test of §5.3 both notes are recorded as fulfilled (Rec. ch. 91), copied to paper 5; 5.125 stands under the song "Birds Fly Off to the Woods" in chapter 5. 13.3 has a further layer, the suppression of the Tower of Heavenly Fragrance episode, which is N6, as with T1-20. 18.17 is N8. Within it, the sentence "余未多及，想为……平儿等人无疑矣" ("the rest are not much touched on; presumably they are … Ping'er and others, without doubt") continues from the Tertiary Register and so places Ping'er there, which conflicts with 6.48; the project takes 6.48, and this layer is N9.

The same group originally also contained 13.14: "the words 'when the tree falls, the monkeys scatter' still sound in my ears; counting on my fingers, it is thirty-five years. Alas, alas, how could it not pain me to death!" What it laments is something the commentator himself heard thirty-five years before, not a later event in the book. It laments without pointing, so it is recorded as not a forward reference and not counted as a missed note; the row remains among the entries, marked "found by supplementary screening; not a forward reference on examination."

Twenty-six notes from the fourth round (none has been through the fulfilment check):

• Assigned to existing items. 6.1 to T1-02: the "second visit" is in chapters 39 to 42, R2; the "third visit" is the title of Rec. ch. 101; "where Qiaojie ends up": in Rec. ch. 101 she chooses to go with Grannie Liu, and in Rec. ch. 103 she marries Ban'er; fulfilled. 18.114 to 18.116 to T1-19: A Cup of Snow foreshadows the fall of the Jia family, landing in the raid of Rec. ch. 91; The Palace of Eternal Youth foreshadows Yuanchun's death, landing in Rec. ch. 91; The Dream of Handan foreshadows Zhen Baoyu returning the jade, landing in Rec. ch. 96; all fulfilled and copied to paper 5.
• A new group: Jia Yun. 24.3, Jing, "the chapter on the Drunken Diamond foreshadows Brother Yun's chivalrous visit to the convent"; 24.61, Gengchen, "when the Rong mansion's affairs collapse later, this man will surely have a part to play." In Rec. ch. 92 Jia Yun visits the prisoners under cover of making offerings before the god of the Prison Temple; in Rec. ch. 94, having obtained a post, he carries a sketch of the ways into the Prison Temple to find Ni Er, the Drunken Diamond, and together with Jiang Yuhan and others they get into the Prison Temple by a ruse. 24.61 is recorded as fulfilled. 24.3 is recorded as half fulfilled: the chivalry layer is fulfilled; the reconstruction does not write the "visit to the convent," and to call the Prison Temple a convent would be forced, so this layer is not taken: N5.
• 42.1 (Gengchen chapter-head note on chapter 42: "Chai and Yu are two in name but one person … see the writing on Baochai after Daiyu's passing, and you will know my words are not mistaken"): that Baochai and Daiyu are one is the commentator's reading; the project takes Daiyu as the core, with each of the two complete in herself (overview §6.3); read differently. Daiyu's entering Baochai's dream in Rec. ch. 108 is a candidate for a response, not a fulfilment (§9).
• 25.104 ("the Magic Jade exorcising evil is seen only this once in the whole hundred chapters; how could it be told again"): a negative proposition. The reconstruction has no second exorcism by the Magic Jade; fulfilled (negative). This note is also material bearing on the number of chapters (overview §3.5).
• Six notes with referent inferred. 25.63, Jiaxu, "the affair of the two Jades: everyone in the Jia mansion, high and low, and even the readers and commentators, all took them for a fine married couple … how could it not be so, alas"; and 45.12, Gengchen, where Daiyu speaks of husband and wife and says "painted" and "acted," "a hidden omen of ill": these two are inferred to mean that the two Jades never became husband and wife, landing in Daiyu's tears running out in Rec. ch. 95 and in Rec. ch. 96, "that private tying of their hair: the Jia household will not recognize such a marriage." 4.35, Jiaxu, "an early root laid for the second half," stands under the gatekeeper's account of the four families on the officials' amulet, "harm one and all are harmed, honour one and all are honoured; they protect and cover for each other, and all look after one another"; inferred to mean the fall of the four families together, landing in the raid of Rec. ch. 91 and Xue Pan's being put in chains in Rec. ch. 103. 46.8, Gengchen, "later each will have a place of her own," and 75.15, Gengchen, "all omens of dispersal," are inferred to mean the fall of the family and the scattering of its people. 72.6, Gengchen, "an omen of Jiang Yan's talent exhausted," stands where Xifeng dreams that her brocade is taken from her; inferred to mean Xifeng's decline, landing in her being divorced in Rec. ch. 90 and her death at the end of Rec. ch. 94.
• 76.3 (Grandmother Jia says, "your father-in-law has been gone more than two years now, poor man," and a Gengchen double-line interlinear note says, "this is not reckoning Jia Jing; it is reckoning the time of [Jia] She's death"): in Rec. ch. 91 Jia She is arrested and imprisoned; in Rec. ch. 92 he is sentenced, stripped of his hereditary rank and sent to Ningguta to redeem his crime by labour, Jia Lian and Lady Xing going with him. After that Jia She does not appear again; Rec. ch. 101 passes over it in one sentence of Xue Ke's, "the Elder Master and Master Lian have been sent far beyond the passes," and the time of his death had not been given. The author ruled that it be written in; Addendum 002 was issued, and the passage has now been written into Rec. ch. 108; recorded as fulfilled (§6.4).
• Eleven notes under the lines of the "Won-Done Song" gloss. 1.168, "after the Ning and Rong houses have fallen," and 1.170, "Yucun and the whole crowd of newly rich upstart families," state the answer to the prophecy in their own words, and under §5.3 they stay in this paper: 1.168 fulfilled (the raid, Rec. ch. 91), 1.170 fulfilled (Yucun's sudden rise, Rec. ch. 107), both copied to paper 5. The other nine move to paper 5. Eight of them only match the lines to persons or places: 1.169, "the Xiaoxiang Lodge, the Purple Rue Studio and such places"; 1.171, "Baochai, Xiangyun and their like"; 1.172, "Daiyu, Qingwen and their like"; 1.174, "Xifeng and her like"; 1.175, "Zhen Baoyu, Jia Baoyu and their like"; 1.178, "Liu Xianglian and his like"; 1.180, "Jia She, Yucun and their like"; 1.181, "Jia Lan, Jia Jun and their like". The ninth, 1.183, "the Land of Illusion and Greensickness Peak are tied up together," is a different kind of identification, of the frame that closes the whole book, with a place but no "who"; it goes to paper 5 with the structural question of the "Won-Done Song" gloss, and is not treated as a mere matching of places. 1.177, "speaking of the day after the parents' death," originally in Class 2, belongs to the same group and moves with it. Paper 5 should add the "Won-Done Song" gloss to what it covers.

Three notes filed as records: 75.3, "Mid-Autumn poem missing; awaiting Xueqin"; 22.104, "from here on damaged and lost; what follows is a later supplement" (系再补; the note is often cited as 俟再补, "awaiting supplement"); 22.106, "this chapter was not finished when Qin [Xueqin] died."

The other notes of Tier 3 are not ruled on individually.

6.4 Addenda

Items that the audit found and the author accepted, but that do not lead at once to a change in the reconstruction, are entered as addenda, which record only where the item lands and the constraints on it, without drafting text. Once the passage has been written, the entry records as it stands that the fulfilment came after supplementary writing: not fulfilled at the audit → the author's ruling → supplementary writing → a changed entry. It is not treated retroactively as satisfied by the original draft.

Addendum 001: T1-07's "choking on pickled greens in the cold of winter" lands in Rec. ch. 102, in the scene where Liu Xianglian and Baoyu share a flatbread at the shrine of the earth god, the same scene as "huddled in a torn felt on a snowy night." It has been written into Rec. ch. 102; closed.

Addendum 002: 76.3, "reckoning the time of [Jia] She's death": supplementary writing that Jia She died on the road to Ningguta. It lands in Rec. ch. 108, at the lonely post station where snow closes the road on Baochai's journey north, and is conveyed by an offhand remark of the station attendant. The constraints set at the outset were:

• No news of the death. What the Jia family knows stops at Xue Ke's sentence in Rec. ch. 101, "the Elder Master and Master Lian have been sent far beyond the passes," which is not changed; nothing is added in Rec. chs. 109 or 110 either. Jia She's death falls only on Baochai's road, and since Baochai dies in the snow in the same chapter, no one carries the news back.
• Shown, not discussed. The attendant names no one, judges no one, and does not describe the death; Baochai, hearing it, makes only one movement and says nothing about it; the narrator makes no remark about retribution (overview §4.5). Yucun travels the same stretch of road, and he is not drawn into it.
• Timing and route. The existing text of Rec. ch. 108 says only that Baochai goes "further and further north," without naming her destination; this sentence in effect fixes that she too is on the road beyond the passes, and the post station must lie before the fork where the men's and women's parties separate.
• Jia Lian and Lady Xing go with Jia She, as in Rec. ch. 92; the attendant says at most that a son and a woman went with him, and nothing is written about what became of them.
• Only a sentence or two is added; the travel pass, the last Cold Fragrance Pill, the poems Leaving the Net of Dust and "Phoenix Hairpin," Daiyu's entering the dream, and the burying of the hairpin are left untouched.

Only the attendant's sentence was written in. The supplement is a small local replacement (round 53, added 27 September 2026). The original:

记人数，宝钗向他借了两张。 "… recorded the number of people. Baochai borrowed two sheets from him."

became:

记人数，看了关文，随口道：「前年也押过一个京里姓贾的老犯官，死在这一带。」宝钗听了，向他借了两张。 "… recorded the number of people, looked over the travel documents, and said offhandedly, 'Year before last we escorted an old convicted official from the capital through here too, name of Jia. He died somewhere round here.' Baochai heard this, and borrowed two sheets from him."

The sentences that follow, "the attendant said the paper had to be returned; Baochai undid a fine thread from her hair and bound up his pen-holder, which was about to break, and the attendant waved it away and went off," are all in the original draft. The attendant names no one and makes no judgement; Baochai, hearing it, makes only one movement, borrowing the paper, and binding the pen is her answer to the attendant's demand that the paper be returned; the two old sheets she borrows are the ones on which she then writes Leaving the Net of Dust. "Year before last" fits the reconstruction's chronology: in Rec. ch. 92, "spring had come again," Jia She is sentenced and sent off; in Rec. ch. 97 the wedding is set for the fourth month, and Jia Zheng dies in Rec. ch. 98, in the hot season after it, the same year as Jia She's departure; and Rec. ch. 108 itself also speaks of Jia Zheng's death as "year before last." All the constraints above were kept; Addendum 002 is closed.

Addendum 003: the Secondary Register layer of 6.48, "presumably she too is in the Secondary Register" (Ping'er), and of T1-03 (6.49, "only on seeing Jinghuan's Roster of Feeling will one know that my words were not mistaken"). In the project's roster, Ping'er is in the Secondary Register. But in the Roster of Feeling in Rec. ch. 110, Ping'er's page came within the paragraph "the remaining pages were very many," passed over in a run with Yuanyang's, Zijuan's and others, without any statement of which register it belonged to. Since the reconstruction does not set out the whole Roster, readers could not see that Ping'er is in the Secondary Register. The Tier 1 checker had noted this doubt at the time (Attachment 05), and after the review of v0.3 a reviewer raised it again. The author ruled that a stroke be added, rather than re-recording the layer as not adopted (Attachment 11).

The supplement is a small local change in Rec. ch. 110 (round 53, added 27 September 2026):

• In the paragraph "the remaining pages were very many," the sentence 平儿页上有杏花·摇篮·半本账。 ("On Ping'er's page were apricot blossom, a cradle, and half an account book.") is deleted, so that Ping'er no longer flashes past in a run of pages;
• the sentence after it, 宝玉没有抬头，只将副册又揭开一页。 ("Baoyu did not look up; he only turned over another page of the Secondary Register."), becomes 宝玉没有抬头，只将副册又揭开一页，却是平儿，画着杏花·摇篮·半本账。 ("Baoyu did not look up; he only turned over another page of the Secondary Register. It was Ping'er, painted with apricot blossom, a cradle, and half an account book.")

The next paragraph, "as the page of the Secondary Register moved, another song rose in the air," is in the original draft; that song now becomes Ping'er's. The apricot blossom, the cradle and the half account book on the page were in the original draft; only their place has changed. The long roster keeps its mixed order and has not been rearranged. After the supplement, the Secondary Register layer of 6.48 and T1-03 is recorded as fulfilled after supplementary writing, and is not treated retroactively as satisfied by the original draft; the layer of 18.17 that places Ping'er in the Tertiary Register is accordingly recorded as N9 (§6.3). Addendum 003 is closed.

6.5 Totals

Tier	Notes	Current ruling
1	28 (25 items)	fulfilled 15 items (of which referent inferred, 2), transposed fulfilment 5, half fulfilled 3, not a forward reference 1, not adopted 1
2	216	answered within the first eighty chapters 117, not a forward reference 45, fulfilled 9, fulfilled (negative) 2, fulfilled (referent inferred) 31, not adopted 1, moved to paper 5 10, reserved 1
3	3,855	34 missed notes found by supplementary screening have been entered: fulfilled 13, fulfilled (referent inferred) 7, fulfilled (negative) 1, transposed fulfilment 1, half fulfilled 1, read differently 1, not adopted 1, moved to paper 5 9; found by supplementary screening but, on examination, not a forward reference: 1 (13.14); filed as records 3; compiler's notes 2; the other 3,815 not ruled on individually. None of the notes entered from the supplementary screening has been through the blind breakdown and check (§7.6)
Total	4,099

As agreed in the overview, these figures will be updated as the audit continues; the data published with the paper are authoritative.

7 The Fulfilment Check

7.1 What is checked

The fulfilment check checks the project's own entries: where the project says a note is "fulfilled," does the plot of the reconstruction really satisfy what that note says? It does not touch the question of whether the notes are true or false. There are only two possible outcomes: the entry is upheld, or the project's entry must change.

The overview called this step a "reverse audit," saying that it "breaks a note into falsifiable propositions and then tries to falsify them against the text of the reconstruction." That sentence can easily be read as setting out to falsify the commentary, which does not accord with the stance of §1. From this paper on the step is called a fulfilment check, and the next version of the overview will be revised accordingly.

7.2 Scope

Not every note entered as fulfilment can be checked. At the time of the check, 62 Tier 2b notes were entered as "fulfilled," but only 13 actually stated a concrete later event from which propositions could be drawn. 34 said only "foreshadowing," the referent being inferred by the project; these cannot be checked against the note, and could only be re-recorded as "referent inferred." 11 were prophecy notes, moved to paper 5. The other 4 had been wrongly entered and were in fact answered within the first eighty chapters. (Reviews and the fourth round of screening later changed a few more: in Class 2, 1.77 and 3.79 were re-recorded (§6.2), and 1.177 moved to paper 5 with the group of notes on the "Won-Done Song" gloss, so that Class 2 now has 31 notes; in Class 3, 19.109 was re-recorded as answered within the first eighty chapters and 79.5 was brought back to this paper, and with 1.177 added Class 3 now has 10.)

The scope of this round was therefore the 23 Tier 1 items with a fulfilment entry (fulfilled, transposed or half fulfilled), plus the 13 Class 1 notes of Tier 2b: thirty-six units in all (Tier 1 counted by audit item, where an item may contain two notes with the same wording). Class 2 is not checked; the notes entered in the fourth round were found after the check, and have not been checked either.

Before this, six groups of targeted tests had been run (overview §9.3), and several items had been checked while the overview was being written.

7.3 Procedure

Blind breakdown. First an analyst was given only the notes and the passages they comment on (about 260 characters before the note and 120 after), without the reconstruction and without the audit table, and was asked to break down what each note says about later events into propositions, written as "who | what event or situation | when | whether negative or not yet realized," and to check against the full text of the first eighty chapters whether the later text the note speaks of already appears within them. The point was to keep the project's answers from being read into the questions.

Check. Next, the check was made in two groups: one checker for the 23 Tier 1 items and one for the 13 Class 1 notes of Tier 2b, so that each note was examined by only one checker. The checkers took the propositions from the blind breakdown, compared them one by one with the reconstruction, quoted the relevant text word for word, and judged by these guidelines:

1. whether identity and situation are fulfilled; the rhetoric of a note is not checked item by item as a specification (for example "black robes and begging for food"; see below);
2. where the same name has different referents, check by meaning;
3. timing: whether the order the note states holds in the reconstruction;
4. agent and patient: who does what to whom;
5. negation and the not-yet-realized (such as "though the matters were not carried out") must also be checked;
6. read the whole passage before judging;
7. where the blind breakdown says a note may already be answered within the first eighty chapters, judge whether that is so.

The checkers' recommendations were ruled on by the author one by one.

Who did it. The blind analyst and the two checkers were all newly opened AI sessions, using the model Claude Opus 5.5, the same model as the consolidating editor who wrote the entries; the date was 27 September 2026 (UTC). The blind analyst's tool-call log shows that it read only the blind materials and the base text of the first eighty chapters (it also listed the directory they were in once; see Attachment 12). "It was not given the reconstruction" can be verified from the record. "It had never before encountered the project's answers" can be asserted only to the extent that a new session carries none of the project's context; the model was the same, so the independence is limited. The prompts and run records are in Attachment 12.

One thing that must be stated plainly. The example given in the blind prompt under "rhetoric and specification" was "black robes and begging for food is only a way of saying 'a person who has left the household life,'" which is exactly the author's ruling on 22.105; in effect, the answer was carried into the question. This was changed for the second breakdown (§7.5). The two checkers' prompts gave the same example in the first check guideline. The checkers could see the current entries and the author's existing rulings, and were asked to apply the rulings and to note separately any doubts (this was by design: the check asks whether the entries are accurate, not to re-rule in the author's place), so on 22.105 the check was not independent. This ruling on the rhetoric of 22.105 is recorded as the author's reading. Since the Sui and Tang dynasties, Chinese Buddhism has combined farming with meditation (Baizhang Huaihai's saying "a day without work, a day without food," in the chapter on Baizhang Huaihai in Wudeng huiyuan, juan 3), and going out with the begging bowl is not the ordinary practice of a convent; the reconstruction writes that Xichun lives a hard, frugal life at the Zhengjue Convent, and only identity and situation are checked. This is a general institutional norm; it explains only why the author does not take "begging for food" literally, and is not used to conclude that the words of the note cannot refer to actual begging. Nor does it serve as a general exemption.

What was checked against. Four compilations of chapters 81–110 of the reconstruction, each identified by checksum (Appendix 3).

7.4 Results

Of the thirty-six units, twenty were upheld and sixteen had their entries changed:

Note	Earlier entry	After the check	Reason
1.185 (T1-01)	fulfilled	not a forward reference	the note is on Zhen Shiyin's "let's go" as he follows the Taoist in chapter 1, an event of that chapter, not a forward reference; it had been attached to Rec. ch. 106 because of the phrase "letting go at the cliff's edge"
8.140 (T1-05)	fulfilled	transposed fulfilment	the reconstruction's Roster of Feeling lists only women; qing bu qing falls in the title of Rec. ch. 110 and in the sharp shout in the Office of the Ill-Fated, not on the roster
19.36 (T1-07)	half fulfilled	fulfilled	Addendum 001 had already been written into Rec. ch. 102; the earlier entry was out of date
26.120, 26.153 (T1-15)	transposed fulfilment	fulfilled	the archery ground was moved from Rec. ch. 94, where it was first planned, to Rec. ch. 87; this is only a change of chapter within the project, and the content is all written
42.6 (T1-18)	fulfilled	half fulfilled	see §6.1
67.1 (T1-25)	fulfilled	transposed fulfilment	letting go is not in the last chapter; the fulfilment chapter for "Shiyin's dream" moves from Rec. ch. 108 to Rec. ch. 110; what Shiyin speaks of in Rec. ch. 108 is Yucun's "dream of power," the same name with a different referent
21.65 (T1-12)	transposed fulfilment	transposed fulfilment (explanation changed)	"having them and casting them aside" becomes "engaged, then cast aside"; the type of the non-adopted layer given
28.1 (T1-23)	half fulfilled	half fulfilled (supplemented)	fulfilment chapters added: Xiren's marriage to Qiguan, foreshadowed by the crimson sash, is in Rec. ch. 81, and "looking after" is added in Rec. ch. 94; the type of the non-adopted layer given
7.40 (T1-04)	fulfilled	fulfilled (supplemented)	the interlinear layer re-recorded as answered within the first eighty chapters
20.45 (T1-22)	half fulfilled	half fulfilled (supplemented)	the layer "the quarrel over the dropped fan" re-recorded as answered within the first eighty chapters (chapter 31)
18.128	fulfilled	answered within the first eighty chapters	chapter 55 already has "not only can there be no visit home"; Yuanchun's death in Rec. ch. 91 is an extension
20.72	fulfilled	not adopted	the note says "later, when they have become husband and wife," and speaks of "the feeling of talking over old times"; in the reconstruction the wedding is never completed, and the two never relate to each other as husband and wife
26.81	fulfilled (Rec. ch. 96)	fulfilled (Rec. ch. 104)	in Rec. ch. 96 the Xiaoxiang Lodge still has green bamboo in the third month; the winter inspection of the garden, with the bamboo tips yellow and the blinds taken down, is in Rec. ch. 104
43.20	fulfilled (Rec. chs. 90, 92)	fulfilled (Rec. chs. 94, 96)	"early death" must land on Xifeng's death; she dies in the night at the end of Rec. ch. 94, and Ping'er learns of it only in Rec. ch. 96
44.12	fulfilled	answered within the first eighty chapters	the root of illness foreshadowed by her "sallow little face" is already answered in chapter 55 ("her face sallow and thin") and chapter 72; the reconstruction extends it to her death
45.15	fulfilled	answered within the first eighty chapters	the night-time gambling it foreshadows is discovered in chapter 73; the desolation of the garden is now entered under Rec. chs. 81, 99, 101 and 104

Several others were upheld with added notes: the type of the non-adopted layer of T1-06 given; fulfilment chapters for T1-09 and T1-16 supplemented with Rec. ch. 92; the notes on the other three plays under T1-19 given their own note numbers, 18.114 to 18.116; a fulfilment chapter for T1-20 supplemented with Rec. ch. 105; the explanation for 22.105 changed from "tilling and feeding herself" to "a hard, frugal life in a mountain convent"; fulfilment chapters added for 16.14 and 44.2.

The changes go in both directions: five were downgraded (1.185, 8.140, 42.6, 67.1, 20.72), two upgraded (19.36, 26.120), three re-recorded as answered within the first eighty chapters (18.128, 44.12, 45.15), two had their fulfilment chapters changed (26.81, 43.20), and the other four (21.65, 28.1, 7.40, 20.45) had their explanations changed, fulfilment chapters added, or one layer re-recorded as answered within the first eighty chapters. The main fault of the entries as first made, in the forward direction, was leniency: as long as the person and the event were present, the note was recorded as fulfilled; there had been no prior check of whether the "later text" the note speaks of was already answered within the first eighty chapters; and some notes had been attached to the wrong place because of a keyword.

After this came review, and the reviewers pointed out several bracketed passages that contain more than one layer but had been given a single judgement as a whole. The rules of §5.2 were set up as a result, and four more entries were changed: 19.80 from not a forward reference to transposed fulfilment (the two epithets of the Roster); 18.128 from answered within the first eighty chapters to fulfilled (negative), settled in Rec. ch. 91; 45.15 split into layers, the desolation layer fulfilled; 38.1 re-recorded to follow 37.1. After the review of v0.2 further changes were made on the reviewers' advice. The desolation layer of 45.15 was re-recorded as answered within the first eighty chapters (chapters 78 and 79). Its main handling now has the same category as the entry after the first check, but for a different reason: the first check had judged that the desolation layer showed "only a first sign" in chapter 79 and held that it landed in the reconstruction, whereas this version, following the reviewers, reads chapters 78 and 79 as the answer. T1-02 and T1-04 were marked "referent inferred"; the "Baochai, together to the end" layer of T1-23 was given as N4; and the layer "had she not gone far away" of 22.103 was re-recorded as R3. After the review of v0.3 three more changes were made: the "cast them aside and become a monk" layer of T1-12 went from N9 to N4; 44.2 was marked fulfilled (negative), settled in Rec. ch. 91; and for the Secondary Register layer of T1-03 and 6.48, after T1-03 had been upheld at the check, a reviewer pointed out that the text did not state it, and the author wrote a supplement to Rec. ch. 110 (Addendum 003), so that it is recorded as fulfilled after supplementary writing, and the Ping'er layer of 18.17 is accordingly N9. The current entries are those of §6.

Add to this what had been found earlier: T1-20 had once taken the frequency of words about the Lin family's ancestral graves as evidence about the Jia family's ancestral graves; the fulfilment chapter of 3.58 had been entered wrongly; 22.105 had once been judged half fulfilled by taking "begging for food" literally as a specification; T1-24 had once been judged disclaimed by forcing a ruling on a fragmentary text; and T1-12, T1-22 and T1-23 changed category while the overview was being written. The entries changed as a result of checking are no longer a small minority. That is exactly why the check is worth making.

7.5 The second breakdown

In the review of v0.1, a reviewer found thirteen rows in the data where the two columns "passage commented on" had been extracted wrongly: in twelve, fragments of neighbouring notes had got in, and in one (16.74) the preceding text had been cut short. The cause was that a window was first opened by character count and the notes were removed afterwards, so that half-notes cut off at the edge of the window could not be removed. The context for the blind materials had been taken in the same way, only with a longer window: in 15 of the 40 notes (16 places), fragments of neighbouring notes had got in at the edge of the window. The text immediately around each note was mostly correct; only for three chapter-head and chapter-end notes, 21.4, 13.89 and 26.153, which have no narrative text before or after them, did the window capture nothing but other notes.

The author therefore ruled that these 15 notes be redone:

1. Corrected extraction. First strip all the notes out of the whole chapter, then cut the excerpt; where the excerpt reaches the start of the chapter it is marked "chapter start," with the chapter title included, and where it reaches the end it is marked "chapter end." The two "passage commented on" columns of the full table were re-extracted in the same way, and the thirteen rows agree one by one with the reviewer's recomputation.
2. A second blind analyst. A different model, Claude Sonnet 5, in a new session; its tool-call log shows that it read only the materials for the second breakdown and the base text of the first eighty chapters. The example in the prompt that carried the author's ruling was replaced with a neutral wording.
3. Comparison. The old and new sets of propositions were compared note by note, with only three kinds of discrepancy fixed in advance: a later-event proposition in the new breakdown that the old one lacks; a difference in person, event, time or negation within the same proposition; a different judgement on whether the note is already answered within the first eighty chapters.
4. Check. Notes with a discrepancy were checked against the reconstruction in another new Sonnet 5 session, and then ruled on by the author.

The comparison at the time sent four notes, 7.40, 13.89, 22.105 and 28.110, to be checked. After the review of v0.2 a reviewer pointed out that the two breakdowns of 26.153 also differed on time: the old one said "the chapters before" could not be confined to chapters before 26, while the new one read it as around chapters 24 to 26 and said that the archery ground "should come near them." The consolidating editor had treated this as a minor difference and had not sent it to be checked, which did not accord with the three kinds of discrepancy fixed in advance; it was therefore given a supplementary check. In all, 10 notes showed no discrepancy and 5 did.

The checker recommended upholding 7.40 and 22.105, and the author approved; on 22.105, the checker could see the author's ruling in the current entry, so the check is still not independent on that item. The supplementary checker for 26.153 (another new Sonnet 5 session) found that the four chivalrous episodes are in chapters 24, 26, 28 and 47, so that "the chapters before" cannot be confined to chapters 24 to 26; that the note only laments that the writing on the archery ground was lost, and does not say in which chapter the archery ground should come; and that the note on chapter 31 says outright that the archery ground comes "several tens of chapters later." It recommended upholding the entry, and the author approved; the layer of adjacent chapters added by the second breakdown is not adopted as a constraint on the reconstruction. The supplementary check is in Attachment 14. On two other notes the checker's recommendations were not adopted:

• 13.89: the checker recommended removing Rec. ch. 105 as a fulfilment chapter, on the ground that Rec. ch. 105 contains only the return of Grandmother Jia's coffin to the ancestral graves at Jinling. But in Baoyu's dream in Rec. ch. 105, Qin Keqing says, "She didn't listen; I don't blame her," which is exactly supporting evidence for "though the matters were not carried out." The checker had searched Rec. ch. 105 for the words "ancestral graves," "family school," "provision," "buying land" and "regulations," and had not read the passage with Qin Keqing; and the entry given to the checker had said only "a further fulfilment chapter at Rec. ch. 105," without the grounds, which were added to the explanation only in v0.2; half the fault lies with the entry. Rec. ch. 105 is kept as a fulfilment chapter.
• 28.110: because the whole reconstruction lacks the words "forgets himself in feeling" (忘情), the checker recommended recording the layer "this leads into the many later moments of forgetting himself" as not fulfilled. But "later chapters" need not be after chapter 80, and many such moments are already visible within the first eighty chapters (§6.1, T1-23). This layer was entered as answered within the first eighty chapters.

Both show keyword hits being substituted for judgements of meaning; in 28.110, "later chapters" was also wrongly confined to after chapter 80. The model was changed, and the same root cause appeared all the same. The checkers' outputs also contain several wrong statements about timing and counts; these are kept as they stand, and the corrections are listed in section 6 of Attachment 12.

7.6 Limits

• Class 2 has not been checked. The 31 notes of Tier 2b that say only "foreshadowing," the referent being inferred by the project, have had their entries changed but have not been checked; what they fulfil is the project's inference.
• T1-08 has not been checked. At the time of the fulfilment check 19.80 was entered as not a forward reference, and so lay outside the scope; after the review of v0.1 it was re-recorded as transposed fulfilment under the first ruling on the v0.1 review (the two epithets of the Roster; Attachment 11), without a blind breakdown and check.
• The notes entered from the supplementary screening have not been checked. The 26 notes entered in the fourth round, the 8 notes entered in the first three rounds and assigned their categories by group, and 79.5, brought back to this paper under §5.3, were all settled after the check and have not been through a blind breakdown. Those among them that are entered as fulfilment and from which concrete propositions can be drawn are of the same type as Class 1, and will be checked first in the next version: fifteen notes, 13.3, 20.36, 22.78, 5.125, 22.97, 6.1, 18.114, 18.115, 18.116, 24.3, 24.61, 25.104, 1.168, 1.170 and 79.5. The first to be checked are 79.5 (brought back to this paper on the strength of a single word, "elegy"), 25.104 (a negative proposition, which requires a full sweep of Rec. chs. 81–110), 24.3 and 24.61 (a newly created group) and 22.78 (transposed fulfilment). The notes with referent inferred are in the same position as Class 2; the notes moved to paper 5 are not in this paper; 18.17 has already been tested in overview §9.3; 42.1 and 13.14 have no fulfilment to check; 76.3 and 6.48 are fulfilments after supplementary writing, for which only the constraints and the added sentences are checked (6.48 was on the list above, and was taken off after the supplement).
• The blind breakdown and the checks were all done by AI. The first blind analyst, the first checkers and the consolidating editor were the same model; the second blind analyst and the second checkers used a different model; the independence is limited in both cases. Every change of entry was ruled on by the author one by one.
• The ruling on rhetoric is the author's reading. Taking "black robes and begging for food" in 22.105 as a way of saying "a person who has left the household life" is the author's reading, recorded as such, and not a general exemption; the Buddhist institutional norm cited only explains why the words are not taken literally.
• The check was run against one version of the reconstruction. The check used the four compilations listed in Appendix 3. By then Rec. chs. 96 to 100 already each had a later sealed draft (round 54), whose wording differs in places; a reviewer spot-checked the sentences quoted in the entries and found them all in the round-54 drafts. The reconstruction is still being revised. Where only wording changes, the entries stay as they are; where a plot-level choice changes, the entries change with it; the sentences quoted in the entries will be checked one by one against the latest sealed drafts in the next version.

8 Root Causes of Misjudgement, and Procedures

The misjudgements found so far come down to a few root causes:

1. taking fragmentary text as settled text (the first ruling on T1-24);
2. taking word frequency as meaning (the ancestral graves in T1-20; 13.89 and 28.110 in the check after the second breakdown, where in the case of 13.89 the entry's failure to state its grounds also played a part);
3. taking the rhetoric of a note as a practical specification (the first ruling on "black robes and begging for food" in 22.105);
4. quoting a song after reading only half of it (recorded in overview §9.1);
5. assigning a tier by keyword without first rereading the passage commented on (1.185 attached to the wrong place);
6. not first searching the first eighty chapters before ruling that a forward reference reaches beyond chapter 80 (1.73, 3.171, 7.41, 52.4, 44.12, 3.79, the desolation layer of 45.15 and others);
7. recording the wrong person in an entry (37.54);
8. not revising the entry after an addendum had been carried out (19.36);
9. a first-order keyword list lacking the single character 伏 and lacking 后来, 后日 and 兆, and a second-order screen run only over the sample (24.3, 24.61, 6.1, 45.12 and others);
10. notes placed under the lines of a prophetic text to decode them contain no keyword, and so cannot be caught by a word screen (the notes on the "Won-Done Song" gloss);
11. giving a whole bracketed passage a single judgement when it contains several layers (19.80, 18.128, 45.15, 38.1);
12. in extracting context, opening the window first and removing notes afterwards, so that other notes get in at the edge of the window;
13. carrying the project's own ruling into the blind prompt;
14. an entry giving only the fulfilment chapter, not the grounds (13.89);
15. in comparing old and new propositions, treating a change in the time reading as a minor difference and not sending it to be checked (26.153);
16. notes that only lament, without pointing to anything, being counted among the missed notes because of a keyword (13.14);
17. in assigning a type, taking the project's inferential reading of a note for the note's own words (the N9 of T1-12);
18. a doubt noted by a checker not being answered point by point in the ruling (the register of T1-03);
19. a rule being set up without all existing entries being swept again (44.2, whose explanation already treated it as a negative proposition while its main handling was not so marked).

The procedures set up in response:

1. collate fragmentary text before ruling on it;
2. check cases of the same name with different referents by meaning (ancestral graves, convents, unicorns, registers, dreams);
3. do not read rhetoric as a specification; check only identity and situation, but say why the words are rhetoric;
4. read a whole song or passage before quoting it;
5. reread the passage commented on before assigning a tier;
6. before ruling that a forward reference reaches beyond chapter 80, first search the first eighty chapters;
7. once an addendum has been carried out, revise the entry;
8. run a regression reconciliation against the final keyword list;
9. go through notes with missing characters by hand;
10. in a fulfilment check, first break down the propositions blind, then compare with the reconstruction;
11. run the second-order screen over the whole body, not only the sample;
12. screen separately for single-character forward-reference words (伏, 兆) and for 后来 and 后日;
13. go through, by position, the notes placed under the lines of prophetic texts;
14. where one note contains several layers, enter them in layers (§5.2);
15. for context, first remove the notes from the whole chapter, then cut the excerpt, and assert that it contains no 【】;
16. examples in a blind prompt must not be taken from the project's rulings;
17. an entry gives the fulfilment chapter together with its grounds;
18. in comparing old and new propositions, go through the three kinds of discrepancy fixed in advance column by column, and do not decide on one's own that a difference is minor;
19. before counting a note as missed, first separate out, by the stated definition, the notes that only lament without pointing;
20. before recording two notes as at odds, first confirm that their words conflict;
21. answer the doubts in a checker's output point by point in the ruling, stating whether each is adopted or not;
22. whenever a new rule is set up, sweep all the entries again.

9 Other Second Halves in the Commentary

If one gathers the notes that were not adopted, the unadopted halves of the half-fulfilled notes, the side set aside where two notes are at odds, the points adapted away in the transposed fulfilments, and the notes read differently, and reads the book by them, one can make out the outlines of several other second halves. This section sets them out one by one, to say why the project did not adopt them. It does not reconstruct these second halves, nor judge whether they are right. Each is set out under three heads: what the commentary says outright; what can be read out along this line; and how the project handles it.

1. Baoyu marries Baochai and they become husband and wife, with Sheyue beside them; later Baoyu "casts them aside and becomes a monk"; Qiguan and Xiren look after "Brother Jade and Baochai," together to the end.

• The commentary says: 20.45, "after Xiren marries, there is still one person beside Baoyu and Baochai"; 20.72, "later, when they have become husband and wife," with "the feeling of talking over old times"; 21.65, "had another man had Baochai for wife and Sheyue for maid, how could he cast them aside and become a monk?"; 28.1, "[Qiguan] in the later chapters, with Xiren, looks after Brother Jade and Baochai, together to the end."
• Read along this line: Baoyu and Baochai really become husband and wife and live together after the wedding, with Sheyue beside them, looked after by Qiguan and Xiren; later Baoyu casts them aside and becomes a monk. 21.65 speaks of casting aside, 28.1 of being together to the end; the two notes can be read as one following the other (the looking after first, the casting aside later), or as coming from two drafts.
• The divergence: what the project adopts is Baoyu choosing for himself to leave, not having them and casting them aside; Baochai's greatest difficulty is her status, not a life of being looked after; and after his awakening Baoyu returns to the human world and lives as a lay disciple, without taking the tonsure (overview §§6.4, 7.2). Recorded as N4; the words of 67.1, "why then shave the head?", are about Xianglian, and its implication serves only as supporting evidence (§6.1, T1-12 and T1-25).
• The project: Baoyu walks away when the engagement is fixed and the wedding is at hand, and Baochai's status is settled after the event; Sheyue beside them, and Qiguan and Xiren looking after Baoyu, are each half fulfilled.

2. Baochai and Daiyu as one person.

• The commentary says: 42.1, "Chai and Yu are two in name but one person; this is a stroke of illusion … see the writing on Baochai after Daiyu's passing, and you will know my words are not mistaken."
• Read along this line: this is first of all the commentator's reading of chapter 42. That Baochai and Daiyu "merge into one" can be an interpretation of the structure of the characters and of the writing; it need not be an event. What it expects is that the writing on Baochai after Daiyu's death will bear this reading out.
• The project: reads Daiyu as the core, each of the two complete in herself (overview §6.3). The Rhapsody on Qingqing says: "You are yourself, not the ground of my being; I too am myself, and do not wait on you to be made clear." Recorded as read differently. After Daiyu's death there is indeed writing on Baochai: in Rec. ch. 108 Daiyu enters Baochai's dream and says, "That is something you wrote yourself. This stroke, no one can make for you." The scene can be a candidate for a response, but what it writes is each of the two on her own, and this paper does not use it to fulfil the note.

3. Baoyu enters the Roster of Feeling as qing bu qing; the Roster lists everyone, and fixes the principal, secondary, further secondary, and third and fourth secondary registers.

• The commentary says: 8.140, "according to Jinghuan's Roster of Feeling, Baoyu is 'feeling for the unfeeling'"; 19.80, "later, in the Roster, one sees the epithets 'Baoyu: feeling for the unfeeling,' 'Daiyu: feeling for those who feel'"; 18.19, "only at the chapter's end, with Jinghuan's Roster of Feeling, will one know the names in the principal, secondary, further secondary, and third and fourth secondary registers."
• Read along this line: men and women are listed together on the Roster, and the registers are fixed. In the commentary the two epithets are praise.
• The divergence: the Roster of Feeling is alive and not fixed once for all (overview §6.2), N4; the meaning of qing bu qing and qingqing the project reads differently, recorded as read differently.
• The project: Rec. ch. 110 writes the Roster of Feeling, listing only women and not the whole roster; qing bu qing falls in the chapter title and in the sharp shout in the Office of the Ill-Fated.

4. Only on the day of the meeting at the Prison Temple does one learn that Qiaojie's "meeting disaster, she finds good fortune; meeting ill, she turns it to good" was a line laid long before.

• The commentary says: 42.6, "only on the day of the meeting at the Prison Temple will one know that 'meeting disaster she finds good fortune, meeting ill she turns it to good' was a line laid a thousand li ahead; alas, alas! The writing that follows is too painful to read through."
• Read along this line: the meeting at the Prison Temple is bound up with Qiaojie's rescue.
• The project: the meeting at the Prison Temple is between Baoyu and Qianxue and Xiaohong; Qiaojie's rescue comes after it, carried out by Grannie Liu and her party, and is written as a chance encounter, not as a reward for hidden merit (overview Appendix 5, the row on "Blessings Left Behind"). This is a different plot arrangement, not a matter of theme; N5, and the note is only half fulfilled.

5. Brother Yun's chivalrous visit to the convent.

• The commentary says: 24.3, Jing, "the chapter on the Drunken Diamond foreshadows Brother Yun's chivalrous visit to the convent."
• Read along this line: Jia Yun acts chivalrously, in a way that answers to the chapter on Ni Er, the Drunken Diamond, and what he visits is a convent; whom he visits the note does not say.
• The project: Jia Yun's chivalry is written at the Prison Temple: he visits the prisoners in Rec. ch. 92, and in Rec. ch. 94 he and Ni Er and others get into the Prison Temple by a ruse. The "visit to the convent" is not written; to read the convent as the Prison Temple would be forced, and it is not taken. This too is a different plot arrangement, N5; half fulfilled.

6. Letting go in the last chapter.

• The commentary says: 67.1, "the 'letting go' in the last chapter is that of one already awakened."
• Read along this line: Baoyu's awakening and letting go are placed in the last chapter of the book; the note does not say whether it is the last event of that chapter.
• The project: letting go is in Rec. ch. 106; the last chapter writes the Roster of Feeling and the return of the jade to Greensickness Peak; transposed fulfilment. That Baoyu has several chapters of life in the human world after letting go is required by the third level of awakening that the project writes (overview §6.4).

7. Had Tanchun not gone far away, the sons and grandsons might have escaped being scattered when the family fell.

• The commentary says: 22.103, "this is the prophecy of Tanchun marrying far away. Had she not gone far away, then when the family's affairs collapsed, its sons and grandsons would not have been scattered. How sad, how grievous!"
• Read along this line: Tanchun's going is one reason why the sons and grandsons were scattered when the family fell; had she been there, the family's affairs might have been put in order. This is a counterfactual of the commentator's, and also a kind of hope.
• The project: the layer of marrying far away is fulfilled. Because she did in fact marry far away, the antecedent of the counterfactual is never realized; so the reconstruction's writing Tanchun's distant marriage does not refute it, and cannot prove it true either. The layer "had she not gone far away" does not assert what the reconstruction ought to write; it is an understanding of why the family fell. The project's narration gives no explanation in terms of cause and retribution, and writes the family scattered but not extinguished; recorded as read differently.

8. Preaching the Law through feeling, to awaken the world.

• The commentary says: 35.26, "this chapter preaches the Law through feeling, to awaken the world … the river of love is bottomless; how can one let it overflow? Once drowned in it, one does not stop short of death."
• Read along this line: this note is not a forward reference, but it reads the whole book as a teaching to warn the world: feeling is a river in which one is bound to drown, and once drowned, one does not stop short of death.
• The project: the narrator gives no explanation in terms of retribution (overview §4.5); recorded as read differently.

9. The three verdicts of the Tertiary Register are Qingwen, Xiren and Xiangling; Ping'er is in the Tertiary Register.

• The commentary says: 18.17, "又有又副册三断词乃晴雯、袭人、香菱三人，余未多及，想为金钏、玉钏、鸳鸯、苗云（下有按语一句，略）、平儿等人无疑矣" ("then there are the three verdicts of the Tertiary Register, which are Qingwen, Xiren and Xiangling; the rest are not much touched on; presumably they are Jinchuan, Yuchuan, Yuanyang, Miaoyun [one sentence of comment follows here; omitted], Ping'er and others, without doubt"); Jihu's upper-margin note in the same chapter (18.19, that is, T1-06) admits that "the twelve beauties cited here are none of them certain; all are guesses"; and 6.48 says of Ping'er, "presumably she too is in the Secondary Register."
• Read along this line: the three verdicts of the Tertiary Register belong to Qingwen, Xiren and Xiangling, and Xiangling is not in the Secondary Register. This conflicts with the text of chapter 5, where Xiangling is in the Secondary Register; it may be another draft of chapter 5, or a guess of the commentator's. As to the others in the Tertiary Register, the commentator presumes them to be Jinchuan, Yuchuan, Yuanyang, Miaoyun, Ping'er and others, which puts Ping'er in the Tertiary Register; 6.48 instead says of Ping'er, "presumably she too is in the Secondary Register." The two notes place her in different registers.
• The project: for the three verdicts, the text is followed, N8; for Ping'er, 6.48 is taken, and Rec. ch. 110 has been supplemented so that Ping'er's page is in the Secondary Register (Addendum 003); this layer of 18.17 is N9.

10. Writing lost from the original, or deleted by the author.

• The commentary says: the original drafts of the Prison Temple and archery ground chapters were lost (20.20, 26.24, 26.120, 26.153); and by the commentator's own account in 13.8, he once ordered Qinxi to delete the passages on the "hairpin left behind" and "changing clothes" (this account is found only in the Jing manuscript).
• Read along this line: the second half originally had a long chapter on the Prison Temple and the writing on Wei Ruolan at the archery ground, and the earlier chapters originally had Qin Keqing's "hairpin left behind" and "changing clothes." These are not another design, but text that was lost or deleted.
• The project: either writes them afresh (the Prison Temple, the archery ground) or does not write them (the hairpin, the changing of clothes); in neither case is the original form reconstructed; N6.

The project adopts none of these second halves, or adopts only part of them. They are recorded here as they stand, to show two things. First, going by the commentary itself, Cao Xueqin's second half may well have had more than one design, and what the commentators saw and thought differed; this is a possibility, and the paper does not claim on this basis that several such complete drafts actually existed. Second, what the project wrote is one reading, chosen by the criterion "persons are ends in themselves" from among the readings the evidence still allows; with a different criterion, many of the "not adopted" here could become "adopted." The commentary here is not an opponent to be refuted, but the witness to several other possibilities. The overview says that "the Roster of Feeling is alive"; this holds for the reconstruction itself as well.

10 Conclusion

The conclusion of the audit can be put briefly. For every note on the first eighty chapters that the project has identified as pointing to later events, a record has been left, note by note: which were adopted and how, which were not adopted and why, the reader can check. The fulfilment check then went through the project's own entries again and changed a good many of them; the supplementary screening found a batch of overlooked notes, some of them well known; and after the paper was sent for review, the reviewers and the second breakdown forced out several layerings and several root causes. No note has been judged wrong here; what has been tested is only the project.

Appendix 1 The Data

Published with this paper is 专题一_脂批前指审计数据_v0.4.csv, in UTF-8 with BOM, one row for each bracketed passage 【】: 4,099 rows and 22 columns. 284 rows carry an entry (the 244 of Tiers 1 and 2; 35 rows from the supplementary screening of Tier 3, that is, 34 missed notes and 1 note found not to be a forward reference on examination; 3 filed as records; 2 compiler's notes); the other 3,815 rows are not ruled on individually. The field names and values are in Chinese; they are glossed here.

Field	Gloss	Content
批号	note number	chapter.sequence, to be read as text (6.10 and 6.1 are two notes)
回、序	chapter; sequence	the chapter; the position of the 【】 within the chapter (integer)
标签	label	the label exactly as in the base text, such as 庚辰双行夹批, 甲夹批 or 该批
本名	manuscript	the manuscript attribution after merging (values below)
批型	note type	values below
批语	note	the full text
所批正文前、所批正文后	passage before; passage after	twenty characters before the note and twelve after, taken after all the notes of the chapter have been stripped out; where the excerpt reaches the start of the chapter it is marked 回首 (chapter start), with the chapter title included, and where it reaches the end, 回末 (chapter end)
初分层	initial tier	甲 (Tier 1), 乙-1 (Tier 2a), 乙（非前指） (Tier 2, not a forward reference), 乙-2 (Tier 2b), 丙 (Tier 3); records the initial division; the current ruling is under 处理
审计号	audit item	the number of each Tier 1 item: 甲01 to 甲25 (T1-01 to T1-25 in the English text)
处理	handling	the categories and destinations of §5.1 (main handling; values below)
分流与不采	routing and non-adoption	codes and explanation for the other layers (§5.4; codes below)
兑现位	fulfilment chapter	chapter(s) of the reconstruction, in Arabic numerals, separated by 、; a range is written with 至 ("to"), as in 81至110; "—" where there is none
说明	explanation	reasons and remarks
复核	check status	judged as follows. Rows that actually took part in the fulfilment check (the thirty-six units, and the four notes 1.73, 3.171, 7.41 and 52.4 found to need a change of category when the scope was being fixed, §7.2) are marked 复核维持 or 复核改登记 according to that check, and later changes by the author's rulings are recorded under 复核记录; only rows that did not take part are divided by current handling into 未复核 and 不复核. The five values are glossed below. Empty for rows without an entry
复核记录	check record	details of the check status: why the category changed, fulfilment chapters added, whether the second breakdown showed a discrepancy, found by supplementary screening, ruling numbers and so on. The "sixteen units whose entries were changed" of §7.4 is a wider count than the check-status column, including changed explanations, added fulfilment chapters and single layers re-recorded
抄送专题五	copied to paper 5	是 ("yes") for notes that decode a prophecy and are kept among this paper's entries while being copied to paper 5
一阶命中词	first-order hits	the terms of the final 45-term list that the note matches
补筛轮次	screening round	轮一 to 轮三 (rounds 1 to 3), 丁一 to 丁四 (screens 4a to 4d); marked only on entered rows found by supplementary screening; the 316 hits of screen 4a over the whole body are not marked row by row and can be recomputed with Attachment 13
抽样、二阶命中轴	sample; second-order axis	是 for the 382 sampled notes; the second-order axis is marked only on the 27 sampled rows hit by the second-order screen: 命 (fate), 时 (time), 叹 (lament)

Values.

Field	Value	Gloss
本名	甲戌、庚辰、己卯、蒙府、戚序、靖藏	Jiaxu, Gengchen, Jimao, Mengfu, Qixu, Jing
本名	无本名；整理者校注	no attribution; compiler's note
批型	侧批、眉批、夹批、双行夹批	side note; upper-margin note; interlinear note; double-line interlinear note
批型	回前批、回末批、总批、序	chapter-head note; chapter-end note; general comment; preface (empty where the label gives no type)
处理	已兑现；改兑现；半兑现	fulfilled; transposed fulfilment; half fulfilled
处理	已兑现（所指推定）；已兑现（否定）	fulfilled (referent inferred); fulfilled (negative)
处理	前八十回内已应；非前指；理解不同	answered within the first eighty chapters; not a forward reference; read differently
处理	不采；保留	not adopted; reserved
处理	移专题五；报存；排除（整理者校注）	moved to paper 5; filed as a record; excluded (compiler's note)
分流与不采	分流一、分流二、分流三	R1, R2, R3 (§5.4)
分流与不采	不采四 至 不采九	N4 to N9 (§5.4)
复核	复核维持	checked, upheld: within the scope of the fulfilment check, with main handling and fulfilment chapters not replaced as a result; this includes changes only to the explanation, added fulfilment chapters, types given, or a single layer re-recorded
复核	复核改登记	checked, entry changed: main handling moved to another category, or fulfilment chapters replaced
复核	未复核	not yet checked: entered as fulfilment but not through the blind breakdown and check
复核	不复核	not subject to check: entries other than fulfilment (answered within the first eighty chapters, not a forward reference, read differently, not adopted, reserved, moved to paper 5, filed as a record, excluded)
复核	补写后兑现	fulfilled after supplementary writing: 76.3 and 6.48, which did not take part in the fulfilment check and whose fulfilment comes from writing added after the author's ruling; T1-03 did take part in the check, so by the rule above it is still marked 复核维持, with the supplementary writing recorded under 复核记录

Rules for merging manuscript attributions.

• labels beginning with 甲戌, or with 甲夹, 甲眉, 甲侧 or 甲嗯, go to Jiaxu; 甲嗯 occurs only once, at 1.158, where the label reads 甲嗯和了人风。蒙侧批, suspected to be corrupt;
• labels beginning with 庚辰 or 庚讲 go to Gengchen;
• labels beginning with 己卯 or 己双 go to Jimao;
• labels beginning with 蒙 go to Mengfu; those that the base text miswrites as 梦 go to Mengfu as well;
• labels beginning with 靖 go to Jing;
• Qi Liaosheng's preface at the head of chapter 1 goes to Qixu;
• passages marked 校者注 are notes by the compiler of the base text, listed separately as compiler's notes;
• all others, such as 该批, 脂批, 题曰, or passages without a label, go to "no attribution."

Where one bracketed passage contains notes from several manuscripts at the same place, it is assigned by its first label. Borderline cases where a corrupt label is assigned to "no attribution" by these rules: 17.125, 前庚辰双行夹批, and 37.25, 辰夹批 (plainly Gengchen); 35.16, 郊批, and 16.151, 列 (unclear). This paper counts them by the rules and does not reassign them.

One anomaly in the base text. After the Gengchen upper-margin note 24.9 in chapter 24, the base text has four extra characters, 笏叟。】, a repetition of the last three characters of the preceding note, and the only closing bracket in the whole book without an opening one. It is removed when context is extracted; the count of notes is not affected.

Appendix 2 Keywords and Sampling

Final first-order keyword list (45 terms): 伏线、伏脉、伏下、又伏、总伏、伏后、伏笔、千里、后文、后回、后数十回、后半部、数十回后、三十回后、后三十回、迷失、无稿、情榜、谶、狱神庙、狱庙、射圃、卫若兰、悬崖撒手、有始有终、寒冬、雪夜、将来、日后、后书、后部、观者记之、末回、后之、遥与、后事、结果、大过节、大关键、大节目、同终始、他日、末世、归结、后半. Before matching, 伏侍 and 伏待 are removed. Matching: substring match on the whole text inside the 【】, including the manuscript label.

Second-order screen, three axes:

• fate: 死、亡、夭、终身、散、败、抄、出家、为僧、削发、嫁、卖、流落、乞、冤、泪尽、了结、收场、下场、应了、验;
• time: 后来、此后、异日、来日、他年、再看、且看、方知、始知、方见、至此了、以后;
• lament and signature: 叹叹、哀哉、伤哉、可伤、悲夫、痛哉、不忍、血泪、三十年、丁亥、壬午、辛卯、己卯冬、畸笏、笏叟、脂砚.

Sampling. The Tier 3 pool: arranged by chapter and by sequence within the chapter, all passages whose full text inside the 【】 (including the manuscript label, after removing 伏侍 and 伏待) contains none of the final first-order terms, 3,855 in all. The sample was drawn with Python's random.seed(5432) and random.sample(pool, 382), and sorted by chapter and sequence. The original sample file kept at the time of the audit, bing_sample.txt (md5 9053f235c613a2313817dba78e5799e1), is identical character for character with the sample redrawn for this paper. The sample's note numbers, the original file and the sampling program are published with the paper (Attachment 01, Attachment 13); the program looks first for 附件01_bing_sample.txt, and the path of the original can also be given on the command line. All the counts in this appendix and in §§3–4 can be recomputed from the base text with the screening program in Attachment 13.

Supplementary screening. The terms for round 2 (notes with missing characters) were "□," 缺 and 阙. Fourth round: 4a, the second-order three axes over all of Tier 3; 4b, the single character 伏, excluding 伏侍、伏待、伏首、伏低、埋伏、伏罪、伏法、伏惟、降伏、伏在、伏身、伏枕、伏气; 4c, 兆、后日、下半部、逝后、不谬; 4d, the notes under the lines of the "Won-Done Song" gloss in chapter 1, 1.167 to 1.183. The hit counts for the fourth round exclude the 7 notes entered in the first three rounds that screen 4a hit again; without that exclusion, 4a hits 323 notes and the four screens together 373.

Appendix 3 The Version of the Reconstruction Checked

The compilations of the reconstruction on which the fulfilment check was run (a historical snapshot of the time of the check; the checksums identify only the version then used, and are not updated as the reconstruction is revised):

Chapters	Draft	md5
81–88	round 54, sealed volume	2c1295a7a536e1b190bece936380fdfe
89–95	round 54, sealed draft	611f11220808e11762346b83beffa4c4
96–103	round 53, sealed draft	481064b2d768b33c6bae291c0e14384c
104–110	round 53, sealed draft	7ef2c176cab5f5c9fbfb982d7c4d55b1

The fourth-round entries were located in the reconstruction against this same version, and the checking in the second breakdown and in the supplementary check on 26.153 was done against it too. There are two exceptions, which were checked against the drafts after supplementary writing, the rest being the same as the version above: 76.3 against Rec. ch. 108 after the attendant's sentence was added; the Secondary Register layer of 6.48 and T1-03 against Rec. ch. 110 after Ping'er's page was added. Both are round 53, added 27 September 2026; the changes, with the text around them, are quoted in §6.4.

At the time of the check, Rec. chs. 96 to 100 each already had a later sealed draft (round 54, the head of each chapter dated 25 or 26 September), whose wording differs in places from the round-53 compilation above, for example in Grannie Liu's words when she hands over the Chenghua cup in Rec. ch. 100 (Appendix 4, item 8). A reviewer spot-checked the sentences quoted in the entries and found them all in the round-54 drafts, so the entries are not affected; in the next version, the quoted sentences will be checked one by one against the latest sealed drafts. As for later revisions of the reconstruction: where only wording changes, the entries in this paper stay as they are; where a plot-level choice changes, the entries change with it and a new version is issued.

Appendix 4 Differences from Version 0.8 of the Overview

In the following places this paper is authoritative, and the next version of the overview will be revised accordingly (in Chinese and English alike). Section, table and appendix numbers are the same in both language versions of the overview.

1. Name: "reverse audit" becomes "fulfilment check," in the heading of §9.3 and in the glossary alike. Where §9.3 says that a reverse audit "breaks a note into falsifiable propositions and then tries to falsify them against the text of the reconstruction," and the glossary says "break the note into falsifiable propositions and try to falsify them against the reconstruction," this becomes "break the notes into concrete propositions and check whether the reconstruction really satisfies them; what is checked is the project's entry, not whether the note is true or false"; "no falsifiable proposition" becomes "referent inferred (no concrete proposition to check)."
2. The table in §9.3: row 5, "the commentary's own error; the text is followed," becomes "conflicts with the text of chapter 5; the text is followed" (this paper's N8); row 6, "internal conflict in the commentary; the reference-weight overrides the standard-weight," becomes "not adopted by the criterion; the implication of 67.1 as supporting evidence" (this paper's N4; see item 11). This paper's N9 (two notes at odds, one taken) now has only one case, the Ping'er layer of 18.17, where both sides are standard-weight, so it is not a case of the reference-weight overriding the standard-weight; if the overview needs an example of N9, it should use this one.
3. §8.2: the example of the question about monthly allowances in chapter 3 should say that the later text of the note is already answered in chapter 39 and that Rec. ch. 86 is an extension; "eighty-three chapters later" is deleted.
4. Table 1, row 19: remove the anchor 1.77 (re-recorded in this paper as not a forward reference).
5. Table 1, row 3: the anchor "the note on chapter 42, 'the day of the meeting at the Prison Temple'" attests only that there is such a meeting; it may stay, with the note "recorded as half fulfilled in paper 1."
6. Table 2: counts updated to §6.5 of this paper. The Tier 3 item becomes: four rounds of supplementary screening, 34 missed notes entered (and one more found on examination to lament without pointing); one new gap in fulfilment (76.3), since filled by supplementary writing in Rec. ch. 108 under Addendum 002; most of the notes under the lines of the "Won-Done Song" gloss moved to paper 5. In the sampling item, "rate of missed notes" becomes "proportion detected."
7. "Six explicitly disclaimed items" (§§8.2, 9.1): say instead that each item is recorded by type. The six fall as follows: the full-roster layer of T1-06, N4; the lost drafts of T1-11, N6; the "hairpin left behind" and "changing clothes" of T1-20, N6; the counterfactual and hope of 22.103, read differently; the causal preaching of 35.26, not a forward reference and also read differently; the "Baochai, together to the end" of 28.1, N4 (the same reason as recorded in overview §9.1).
8. The Miaoyu rows of Table 3 and of Appendix 5: in both places, "Grannie Liu entrusts the Chenghua cup" becomes "Grannie Liu gives the Chenghua cup outright," following the round-54 sealed draft of Rec. ch. 100 (whose text reads: "这个原是你们府里出来的，这几年我一向带在身上。姑娘认得，就留着罢。" "This came from your household to begin with; I've carried it on me all these years. You'll know it, miss. Keep it."). The round-53 compilation used for the check in this paper still reads "替我还了这一注人情罢" ("repay this debt of kindness for me"); the two are a change between versions, and the earlier draft is not a ground for doubting the round-54 ruling that the cup is simply given. This item has nothing to do with the entries of paper 1; it is a version update for the overview.
9. §3.4 or §4.1: add a sentence on stance: the commentary is testimony whose provenance must be kept, containing accounts of the drafts the commentators saw as well as their inferences and judgements; for the notes this paper does not adopt, it says only where the divergence lies, and does not judge them wrong.
10. §6.4: "having them and casting them aside" should become "engaged, then cast aside," or a sentence of explanation should be added.
11. Every place where "the reference-weight overriding the standard-weight" appears (the paragraph on selection by the criterion in §4.1; "second, becoming a monk" in §6.4; Table 1, row 11; the Tier 1 item of §8.2; row 6 of the table in §9.3; "the third level of Baoyu's awakening" in §9.4; the glossary): 67.1, "though this one was attached, he broke through the barrier of delusion; why then shave the head?", has two readings, and its words are about Xianglian; the project takes its implication, that the one who lets go in the last chapter need not shave his head. The words of the two notes do not conflict, so this paper records the "cast them aside and become a monk" layer of 21.65 as not adopted by the criterion (N4), with the implication of 67.1 serving only as supporting evidence. Every place where the overview uses this as its example of the reference-weight overriding the standard-weight should say instead that it is not adopted by the criterion (selection by the criterion), with the implication of 67.1 as supporting evidence; the glossary entry "the reference-weight overriding the standard-weight" should be deleted, or marked "no current instance in the project."

Appendix 5 Attachments

Three files are published with this paper: the paper (PDF), the full audit table (CSV) and an attachment package, 专题一_附件包_v0.4.zip. The package contains Attachments 01 to 14, with the file names below, and a file 读我.md ("read me") explaining how to run the programs. The attachments are in Chinese.

Item	File	Content
Data	专题一_脂批前指审计数据_v0.4.csv	the full audit table, 4,099 rows, 22 columns (Appendix 1)
Attachment 01	附件01_抽样.md, 附件01_bing_sample.txt	the sampling procedure, the sample's note numbers, the 27 hits of the second-order screen in the sample; the original sample file kept at the time of the audit
Attachment 02	附件02_盲拆输入_40条.md	input to the first blind breakdown; it has the extraction defect described in §7.5, and is kept as it was
Attachment 03	附件03_盲拆命题_40条.md	output of the first blind breakdown
Attachment 04	附件04_复核时登记快照.md	the entries given to the checkers at the fulfilment check (one redaction in the public version; see Attachment 12, section 4)
Attachment 05	附件05_复核_甲层.md	check record for the 23 Tier 1 items (two redactions in the public version; see Attachment 12, section 4)
Attachment 06	附件06_复核_乙层一类.md	check record for the 13 Class 1 notes of Tier 2b
Attachment 07	附件07_重拆输入_15条.md	input to the second breakdown, after corrected extraction
Attachment 08	附件08_重拆命题_15条.md	output of the second breakdown
Attachment 09	附件09_重拆比对_4条.md	comparison materials given to the checker of the second breakdown (four notes)
Attachment 10	附件10_重拆核对_4条.md	check of the four discrepancies
Attachment 11	附件11_裁定记录.md	the author's rulings on the checkers' recommendations and the reviewers' comments, up to the release of v0.4
Attachment 12	附件12_运行说明.md	for each AI session, its purpose, model, times, the files it read, and the full text of its prompt; the record of redactions in the public version; statements in the checkers' outputs that are not adopted
Attachment 13	附件13_程序/	筛查重现.py (the counts of §§3, 4 and Appendix 2), 抽样重现.py (the Tier 3 sample), 上下文提取.py (the two "passage commented on" columns of the full table)
Attachment 14	附件14_补核_26.153.md	the supplementary check on 26.153 after the review of v0.2: comparison materials and the checker's output (§7.5)

Appendix 6 Acknowledgements and Use of AI

I thank Zesi Chen and Zhen Zhang for a great deal of analysis and help in the course of the project.

The project used AI assistance from several providers, including Claude (Anthropic), ChatGPT (OpenAI), Gemini (Google), DeepSeek, GLM (Zhipu), Meta AI and Grok (xAI). The audit on which this paper rests was carried out by AI, which did the screening, the entries and the checks according to criteria the author set; in the fulfilment check the blind breakdowns, the checks and the second breakdown were likewise carried out by AI, each by a separate session, the models and sessions being given in §§7.3 and 7.5 and in Attachment 12. The English version is a rewriting of the Chinese version for English-speaking readers, drafted with AI assistance and checked against the Chinese; where the two differ in substance, the Chinese is authoritative. Every ruling was made by the author, who bears responsibility for the text.

References

• Qin, Han (秦汉). People Have Qing: Origins, Method and Conclusions of an Interpretive Reconstruction of the Last Thirty Chapters of Dream of the Red Chamber (人是有情的：《红楼梦》后三十回诠释性复原的缘起、方法与结论). Pre-release v0.8 (Chinese and English). Zenodo. Concept DOI: 10.5281/zenodo.22985375.
• Jiangshange Yin (江山阁隐), ed. Honglou meng Zhipi ben wanmei ban 红楼梦脂批本完美版 (Perfect Edition of the Zhiyan Zhai Commentary Text of Dream of the Red Chamber). Electronic edition circulated online (PDF; according to its maker, based on the Gengchen manuscript, with commentary drawn from the Jiaxu, Wangfu, Qixu and other manuscripts).
• Cao Xueqin. Zhiyan Zhai chongping Shitou ji 脂砚斋重评石头记, manuscripts: Jiaxu, Jimao, Gengchen, Mengfu (Menggu wangfu), Qixu; the Jing commentary (the original lost, known from transcription).
• Cao Xueqin. Honglou meng 红楼梦. Collated and annotated by the Institute of Honglou meng Studies, Chinese National Academy of Arts. Beijing: Renmin wenxue chubanshe.

[^1]: On the source and preparation of the base text, see the overview, "Notes on the text," item 3, and its note. Manuscript attributions follow the base text; checking every difference from the standard critical editions against facsimiles is outstanding, and the three places already pointed out by reviewers come first: 18.19, "at the chapter's end" (回末) or "in the last chapter" (末回); the attribution of 42.6 (labelled 该批 in the base text; possibly a Jing upper-margin note); the attribution of 6.49 (labelled 脂批 in the base text; it follows directly on the Jiaxu interlinear note 6.48, and "my words" refers to 6.48; standard compilations may give the two as a single Jiaxu interlinear note; if so, T1-03 should be attributed to Jiaxu and entered in the same place as 6.48).

说明

1. 本文是预发布版。审计会续做，登记会随复原本情节层的取舍与新的核查而更新。本文写到哪一步就说到哪一步：已复核的写结果，未复核的写明未复核，推定的写明推定。
2. 本文是总篇《人是有情的》（concept DOI 10.5281/zenodo.22985375）的材料篇之一。总篇 8.2 节表二是本审计的摘要；二者有出入时以本文为准，总篇下一版随改（附录四）。总篇所称“反向审计”，本文起改称“兑现复核”（第 7 节）。
3. 本文所说“复原本”，指本项目所写第八十一至一一〇回。本文的登记立在复原本情节层的取舍上，不系于任何一轮稿件；兑现复核所据的一版，记其校验值于附录三。
4. 批号记作“回.序”，即某回之内第几处【】，如 21.65 是第二十一回第六十五处。批号须作文本读。本名、批型依项目底本所标著录。
5. 第 6 节是各条的现登记；第 7.4 节是兑现复核中由旧判到新判的变化，两处并看。
6. 审计全表、盲拆与复核记录、抽样样本、各次 AI 会话的运行说明，随文发布（附录一、附录五）。

---

1 立场：批语是证词

脂批里有一类话，说的是八十回以后的事：“后三十回”“狱神庙”“卫若兰射圃”“悬崖撒手”“警幻情榜”，还有“伏下后文短命”“将来事败”之类。续书、探佚、复原，历来把它们当作后半部的线索。本项目也是这样用的。本文要交代的，是本项目对每一条这样的批语怎样处理，以及为什么。

先立三条。

批语是需保留来路的证词材料。批语里有对所见稿本的记述，也有批者的推断、评议、愿望与假设；不同批语所据未必同稿。书是“披阅十载，增删五次”改出来的，下半部很可能有过不止一种设计，批者也不止一人。两条批语说法不同，可能是两稿不同，也可能是一人的推断与另一人的见闻不同，不一定有一条说错了。所以本文不裁批语的历史真伪，也不说哪一条是“误批”“自误”。

审计登记的是本项目的处理。对每一条批语，本文只记本项目怎样处理它：采（兑现），变通地采（改兑现），采一半（半兑现），理解不同，不采，或存而不判。登记的对象是本项目的取舍，不是批语的真伪。

不采，只说分歧所在。不采的一条，理由只写它与本项目哪一项取舍不合：与本项目所取的主旨有分歧，也就是与总篇所立的“人是有情的”，与以“人是目的”为判准择出的那些读法，走不到一起；或者只是本项目已定的情节布置与之不能并存，属创作取舍；或者与前八十回正文不合，或两条批语互异而只能取其一。都不说批语错了。

由这三条，审计的价值不在裁决，而在留痕：本项目已识别的每一条前指，都有一个处理记录，读者可以逐条查，也可以换一个主旨，逐条改判。

2 为什么要审计

本项目写后三十回的时候，原先是“从后往前”找依据：为已写的情节去找批语。这样做只能证明所写有据，不能证明有据者都写了；不采的批语也不留痕，分不清“审过而不采”与“根本没看见”。审计补的就是这一环：把前八十回的全部批语过一遍，每一条指向后事的批语，都留下处理记录。

审计只记录，不改稿。复原本要不要因某条批语改写，由作者另裁；另裁而补写的，登记写明是补写后的兑现（6.4 节）。

3 底本与批语

底本。前八十回正文与批语，据本项目所用的汇校电子底本（下称“项目底本”），即据江山阁隐制作的《红楼梦脂批本完美版》PDF 整理而成的文本（md5 06d87cd765c8ba133976722331b06b2b）。批语在底本中以【】标出，依所标本名著录[^1]。

计数。项目底本共有【】4,099 处。其中两处是整理者的校注，一处是戚蓼生序文；其余是批语，以及甲戌本凡例、卷首题诗、蒙府回前诗、“题曰”诸诗等同样以【】标出的文字。本名按底本所标归并（归并规则见附录一），分布如下：

本名	条数
庚辰	1,922
甲戌	1,516
蒙府	560
靖藏	17
己卯	13
戚序	1
无本名	68
整理者校注	2
合计	4,099

项目内部早先的一次计数是庚辰 1,918、甲戌 1,495、蒙 560、无本名 96、靖 17、己卯 12、戚 1，与上表略有出入。原因是底本的标签写法不一（如“甲夹批”“庚讲批”，蒙府批有几处误作“梦”），早先的计数没有逐一归并。本文以上表为准，因为它与随文发布的数据同出一源。

本名权重。本项目以甲戌、庚辰为准级，蒙府、戚序、靖藏为参级。这是本项目给各本批语定的采信分量，不是对批语真伪或各本价值的判断。准级与参级两说不合时，依判准择一，逐处登记（第 5 节）。己卯本批语十三条都在丙层，本文不涉其定级；无本名的，依底本所标（“脂批”“该批”等）照录，甲层有两条（6.49、42.6），不定级；两条的本名都待影印本核（脚注一）。

4 筛法与分层

4.1 关键词筛与三层

先用关键词表从 4,099 条中筛出前指候选。关键词表是分批演进的：第一批甲层词表以“迷失”“无稿”“情榜”“狱神庙”“射圃”“撒手”“数十回后”为轴；审计中发现漏收“大过节”“大关键”“后事”“同终始”“后三十回”诸表述，陆续补入，又补“大节目”“他日”“末世”“归结”“后半”等，并排除“伏侍”“伏待”两个同形词。终版词表 45 词，匹配法是在【】内全文（连同本名标签）作子串匹配，见附录二。以终版词表在全部批语上重跑，得前指候选 244 条。

候选分三层：

1. 甲层（硬前指）：直接指认八十回后之事，如迷失稿、情榜、狱神庙、射圃、撒手、“后三十回”、“大关键”。共 25 项，涉及 28 条批（有几项由两条同文或同处的批合成），全数逐条裁定。
2. 乙层（前指候选）：其余候选，按所指落在哪里分流。

- 乙-1：所指落在前八十回之内的，记为前八十回内已应，只登记批号，不逐条论证； - 乙-2：所指跨到八十回以后的，逐条裁定。 - 候选中另有一批，只是评点、赞叹、笔法或读法指示，不指认后事，记为非前指。

1. 丙层（其余）：3,855 条。

这是初分。后来的复核与审读又把一些条目改了处理（如乙-2 里有几条改记前八十回内已应），数据里“初分层”一栏保留初分，“处理”一栏记现判。唯 1.73、3.171、7.41、52.4 四条，初分在乙-2，兑现复核时查明所指在前八十回之内，初分层随改为乙-1（7.2 节）。

4.2 丙层的抽样与补筛

丙层不逐条裁定，只检验关键词表有没有漏网。漏网只计所指跨到八十回以后的前指：批语须指认，或推得出，一件八十回以后之事。只叹不指的（如“余为袭卿一叹”“令余酸鼻”），所指落在前八十回之内的，都不计。

抽样。以固定种子 5432 从丙层抽 382 条，约占十分之一；抽样程序、样本批号与原件校验值见附录二，任何人都可以重现。为了减少一阶关键词表的盲区，另设一组二阶筛，分命运、时序、叹款三轴，所用词与一阶词表不同；词表不同只说明两套规则有别，不等于统计或语义上的独立。二阶筛在样本中命中 27 条，逐条阅过，确认跨八十回前指 2 条：13.3（秦可卿托梦，缺字批）与 28.106（“此处表明以后二宝文章”）。2／382 只是这一筛检流程在样本中的检出比例；未命中的 355 条这一步没有逐条裁定，不能据此估计丙层全部残留漏网的比率。

前三轮定向补筛。抽样暴露出两处词表真空，于是对丙层全量作三轮定向补筛：

轮	筛法	命中	真漏网
一	“以后”单用（一阶只收“后之”“日后”等）	6	3：20.36、22.78、28.106
二	缺字批（含“□”“缺”“阙”者，关键词可能已佚）	17	1：13.3
三	榜册与败亡语（幻榜、芳讳、副册、又副、正册、抄没、败落、树倒、食尽、鸟投林、了局、收缘）	8	4：5.125、6.48、18.17、22.97

三轮真漏网合计 8 条；抽样所得的两条，补筛中又各自筛出，计在这八条之内。八条逐条归位后，七条落入已裁定的组；18.17（又副册三断词作晴雯、袭人、香菱）与第五回正文不合，记不采八（第 5 节）。轮三原还计有 13.14，“‘树倒猢狲散’之语，今犹在耳，屈指三十五年矣”，叹的是批者自己三十五年前所闻，不是书中后事；按上面的口径只叹不指，v0.2 送审后核定为非前指，不计漏网（6.3 节）。补筛同时查出 75.3 庚辰“缺中秋诗俟雪芹”，是底本自记的未完标记，报存不裁。

第四轮（丁轮）。写本文时，把二阶三轴从样本扩到丙层全量，又补三种筛法。前三轮所得十条（含报存的 75.3；其中 13.14 后来核定为非前指，仍是登记行）中，有 7 条又中了丁一（5.125、13.3、13.14、20.36、22.78、22.97、28.106），其余三条不中本轮各筛；表中命中数已扣除这 7 条。各筛之间的重叠，在合计时去重。

筛	筛法	命中	真漏网
丁一	二阶三轴，扫丙层全量	316	12
丁二	单字“伏”（排除伏侍、伏待、埋伏、降伏等）	22	6
丁三	兆、后日、下半部、逝后、不谬	25	6
丁四	好了歌注的句下批，按位置查（第一回，1.167 至 1.183）	16	11

四筛合计命中 366 条，真漏网 26 条，另报存 2 条。26 条中，18.114 至 18.116 三条（元春所点四出戏里前三出的批），甲19 原已论及，只是没有分列批号；其余 23 条是真的漏看了。其中有熟知的前指：24.3 靖本“伏芸哥仗义探庵”，24.61 庚辰“此人后来荣府事败，必有一番作为”，6.1 甲戌回前批“伏‘二进’‘三进’及巧姐之归着”；有总篇写作时已先查出、交本文补登的 42.1（总篇 6.3 节）；还有好了歌注每一句下的“某某一干人”一组。v0.1 曾说三轮之后没有新增兑现缺口，这一轮推翻了这个说法：76.3“不是算贾敬，却是算赦死期也”，复原本原先没有写贾赦之死（6.3、6.4 节）。

漏网的原因有三：一阶词表里没有单字“伏”（“伏芸哥仗义探庵”“中伏贾家之败”），也没有“后来”“后日”“兆”；二阶三轴原先只扫了样本；解谶的句下批（“宝钗、湘云一干人”）通常不含任何关键词，只能按位置查。由此补的工序见第 8 节。

4.3 终版表回归对账

关键词表是分批加词的，各批用表不同，批次之间会留下夹缝。例如 1.94 甲戌侧批“又出一警幻，皆大关键处”：第一批词表无“大关键”，后来的回扫表有此词，却只报新收，两头都没有登记。所以审计收尾时，以终版词表在全量上重跑一次，与表内登记的批号逐号对账；登记数与机筛数不符，就逐条追平。本轮对账追出 1.94 与 20.2 两条，已补裁；现在登记与机筛都是 244 条，一一相符。缺字批的关键词可能就在阙文里，机筛无解，须人工过。

5 处理与分流

5.1 处理档

档	意思
已兑现	复原本情节层满足了批语所说的事
改兑现	批语的字面兑现了，所指被改写；或时序、位置有变通
半兑现	批语所说的几层里，兑现了一部分，其余不采
前八十回内已应	批语所说的“后文”在前八十回之内已经出现；复原本即便另有延伸，也不以之为后三十回的约束
非前指	评点、赞叹、考据、笔法、读法指示一类，不指认后事
理解不同	批语对人物、词义或后事意义的理解与本项目不同；照录，不算兑现，也不算不采
不采	本项目不采这条批语所说的事，注明型
保留	存而不判

另有两个标注。所指推定：批语只说“伏线”“伏下后文”“千里伏线”“将散之兆”，没说伏的是什么事，所指是本项目推出来的。这类登记为“已兑现（所指推定）”，写明推定的依据；它兑现的是本项目的推定，不是批语的明言。甲层的甲02、甲04 也属此类，经过复核，也只证明复原本满足了本项目的推定，不证明批语只容许这一解。否定：批语说的是“只此一见”“不再”一类否定命题，复原本到其落定之处仍不写其事，即为满足，登记为“已兑现（否定）”，如 18.128、44.2、25.104。

另有两种去向。移专题五：解谶之批，与所谶之文一并在专题五处理（界线见 5.3 节）。报存：底本或成书状况之记（如“缺中秋诗俟雪芹”），不属本审计口径，存而不裁。

“已兑现”不作证据。已兑现记的是登记之实。复原本多是依批而写的，兑现由构造所致；兑现的多少，不作复原本近于原稿的证据。第九十回回目直接用了第二十一回回前批所举的十四字（甲21），是直采，更不作证据。

5.2 三条规矩

1. 计数不变，登记分层。一处【】仍算一行，统计不变。一条批里有几层话，就分层登记：主处理记其中跨到八十回以后的那一层，其余各层写入“分流与不采”一栏与说明。
2. 否定命题要看落定处。“不再”“只此一见”“其事虽未行”这类批语，要到事情最终落定的地方才算兑现；前八十回里只露出苗头、或者一时受阻的，记为前八十回内之端，不算整条已应。
3. 重出之批同处登记。不同本子里同文的批，登记在同一处，注明“重出”。

5.3 解谶之批的界线

谶文本身（判词、曲文、灯谜、诗谶、酒令、好了歌注），在专题五处理。批语自己的字里说出了一件可以核的事（远适、为尼、不再、败、死、诔之类），就是一条前指，在本文登记，注“抄送专题五”；只点一个“谶”字，只说“伏后文”“将来形景”，或只把谶文对到某人某处（“某某一干人”“潇湘馆、紫芸轩等处”），就移专题五。测法是一句：批语明确指向书中可辨认的人或家，并说出可核的事，才留本文。所指可由批语明引或明确回指的已存正文确定，不限在该批以前：“树倒猢狲散”出第十三回秦氏之口，说的是贾家，所以引这句俗语的 5.125（批在第五回）、22.97 都留本文。“谁”可由所引之语确定，“何事”却须在批语字面：79.23“偏是此等事体等到”，事是代词，所以移。泛称不算“谁”：1.177“言父母死后之日”只解释“日后”的时点，“父母”未指明哪一家，随谶文移专题五。这是本项目在两篇专题之间分工的工作规则，不是关于一切伏笔的普遍定义。专题五照单接过本文留下的解谶批，不重复登记。现在抄送专题五的有十二条：1.168、1.170、5.125、18.114 至 18.117、18.128、22.97、22.103、22.105、79.5。

5.4 分流与不采

编号	名称	原审计表中的名称
分流一	非前指	甲型
分流二	前八十回内已应	乙型
分流三	理解不同	（新立）
不采四	与本项目所取主旨有分歧	丙型
不采五	情节布置不同：本项目已定的情节与批语所说不能并存，属创作取舍，不涉主旨	（新立）
不采六	原稿已失或作者自删，不复原其原形	丁型
不采七	参级本单出，无准级旁证（只关权重，不关真伪）	戊型
不采八	与前八十回正文不合，从正文	己型
不采九	两批互异，取其一	庚型

分流一至三不是拒绝：非前指无可采，前八十回内已应无须采，理解不同是两种读法并存。不采四至九才是拒绝，而没有一型是“批语有误”。不采四是第 1 节立场的主体；不采五容许“作者选了另一种写法”止于创作取舍，不必每次上升成主旨分歧。不采七全表至今没有用上，照录其定义。原审计表记“脂批自误”的 18.17 改入不采八，理由只写与第五回正文不合、从正文。原记“脂批内部冲突”的甲12 一层，先改入不采九；v0.3 送审后改记不采四，因为不采九要两条批语字面相抵，而 67.1 字面说的是湘莲（6.1 节）。不采九现在只用于 18.17 的一层：它把平儿推在又副册，与 6.48“想亦在副册内”相抵，本项目取 6.48（6.3 节，补记条〇〇三）。两批互异，只记取了哪一边，不判另一边错。总篇所记“明确不承账六条”，现在分入不采四、不采六与分流三（附录四）。

6 结果

6.1 甲层

甲层 25 项的现登记如下：

审计号	批号	本名	批语要点	处理	复原本回次
甲01	1.185	甲戌眉批（靖眉同）	“走罢”二字真悬崖撒手	非前指	—
甲02	6.10	甲戌侧批	是数十回后之正脉也。真千里伏线	已兑现（所指推定）	100、101
甲03	6.49	无本名（标“脂批”）	观警幻情榜方知余言不谬	已兑现	110
甲04	7.40	甲戌双行夹批、眉批	闲闲一笔，却将后半部线索提动	已兑现（所指推定）	99、104、105
甲05	8.140	甲戌眉批	按警幻情榜，宝玉系“情不情”	改兑现	110
甲06	18.19	庚辰眉批	至回末警幻情榜方知正、副、再副及三四副芳讳	改兑现（全谱一层不采四）	110
甲07	19.36	庚辰双行夹批	下部后数十回“寒冬噎酸虀，雪夜围破毡”	已兑现	102
甲08	19.80	庚辰双行夹批	后观《情榜》评曰“宝玉情不情”，“黛玉情情”	改兑现	110
甲09–11	20.20	庚辰眉批	茜雪至“狱神庙”方呈正文；“花袭人有始有终”；五六稿被借阅者迷失	已兑现（茜雪、袭人两层）；迷失稿一层不采六	92、93、94、106
甲12	21.65	庚辰双行夹批	若他人得宝钗之妻、麝月之婢，岂能弃而为僧哉	改兑现（弃而为僧一层不采四）	97、105
甲13	25.107	甲戌眉批	叹不得见玉兄“悬崖撒手”文字为恨	已兑现	106
甲14	26.24	甲戌眉批	“狱神庙”红玉、茜雪一大回文字惜迷失无稿	已兑现（迷失稿一层不采六）	93、94
甲15	26.120、26.153	庚辰眉批；甲戌	惜“卫若兰射圃”文字无稿	已兑现（迷失稿一层不采六）	87
甲16	27.79、27.81	甲戌侧批；庚辰眉批	“狱神庙”回内方见；此系未见“抄没”“狱神庙”诸事	已兑现	91、92、94
甲17	31.3	庚辰	后数十回若兰在射圃所佩之麒麟正此麒麟也	已兑现	87、106
甲18	42.6	无本名（标“该批”）	狱庙相逢之日始知“遇难成祥，逢凶化吉”实伏线于千里	半兑现（不采五）	94、99、100
甲19	18.117	庚辰双行夹批	所点之戏剧伏四事，乃通部书之大过节、大关键	已兑现	91、95、96
甲20	13.8、13.89	靖；甲戌	魂托凤姐贾家后事二件……其事虽未行；命芹溪删去	已兑现（删稿一层不采六）	92、105
甲21	21.3、21.4	蒙回前批	未见后三十回犹不见此之妙……“薛宝钗借词含讽谏，王熙凤知命强英雄”	已兑现（直采）	90
甲22	20.45	庚辰双行夹批	袭人出嫁之后，宝玉、宝钗身边还有一人	半兑现	97
甲23	28.1、28.110	庚辰；甲戌	后回与袭人供奉玉兄宝卿得同终始	半兑现（不采四）	81、94、97、106
甲24	41.14	靖眉批	他日瓜州渡口，客示劝惩，红颜固能不屈从枯骨（残文，校读）	已兑现	100、105、110
甲25	67.1	靖	末回“撒手”……是何必削发。青埂峰时缘了证情，仍不出士隐梦	改兑现	105、106、110

合计：已兑现 15 项（其中所指推定 2 项：甲02、甲04），改兑现 5 项，半兑现 3 项，非前指 1 项（甲01），不采 1 项（甲11，迷失稿）。

几条要说明：

• 甲02：“略有些瓜葛，是数十回后之正脉也”，批语说刘姥姥家这门瓜葛是数十回后的正脉，没说是什么事；所指（刘姥姥一行救巧姐，第一〇〇、一〇一回）是本项目推出来的，记已兑现（所指推定）。
• 甲03：6.49“观警幻情榜方知余言不谬”，“余言”即 6.48“想亦在副册内者也”（脚注一）。第一一〇回情榜平儿原有册页，但那一页夹在“余下册页”一串里，没有写属于哪一册。甲层核对者当时已记此疑：“册次未标，这一层只能算不冲突，不能算坐实”（附件05），复核维持时没有接。v0.3 送审后审读者再指出，作者补写第一一〇回，平儿一页落在副册（补记条〇〇三）。副册一层记补写后的兑现，不追溯为原稿已经满足。
• 甲04：眉批只明言“后半部线索”，没说是什么事；作者读作惜春与智能儿最终同在庵中，记已兑现（所指推定）。复核证明复原本满足这一推定，不证明批语只容许这一解。智能儿在前八十回本是水月庵的尼姑，复原本不另写她还俗，她就仍是。第九十九回惜春致书智能儿：“当年同姐姐说的那句顽话，我如今是当真了”，回应第七回的戏言；第一〇四、一〇五回二人同在正觉庵。同处夹批“又伏后文”一层，所伏的智能儿，第十五、十六回已有正文，记分流二。
• 甲06：全谱一层记不采四，出于“情榜是活的”这一判准（总篇 6.2 节），不以一榜定人，不是情节布置的缘故。
• 甲08：前段是对宝玉、黛玉的长篇评议，记分流一；批末说“后观《情榜》评曰‘宝玉情不情’，‘黛玉情情’”，与 8.140 同类。第一一〇回榜上有“林黛玉　情情”；宝玉不在榜上，“情不情”落在回目、薄命司的断喝与宝玉自述，同甲05，所以记改兑现。两评的意思，本项目另有读法：批语以“情不情”为褒（8.140“凡世间之无知无识，彼俱有一痴情去体贴”），本项目读作病，把“情情”读作“敬人如敬己”（总篇 6.1、9.2 节），这一层记分流三。
• 甲12：这一条总篇 6.4 节已拆作两个命题。“得而弃之”一层，按复原本的时序应说“定而弃之”：婚约已定，宝玉婚日四更翻墙出走，“妻”的名分是走后才追定的，所以记改兑现。“弃而为僧”一层，原记不采九，以为与 67.1“是何必削发”互异。v0.3 送审后审读者指出：不采九要两条批语字面相抵，而 67.1 字面说的是湘莲（见甲25），与 21.65 相抵的只是本项目的推读。现改记不采四，理由是判准：宝玉悟后回到人间，为居士而不剃度（总篇 6.4、7.2 节）；67.1 的含意只作旁证。
• 甲14、甲15：批语叹狱神庙、射圃的文字迷失无稿。本项目另写了这两处（第九十三、九十四回；第八十七回），记已兑现；“迷失无稿”是成书状况之记，不是复原本的规格，原稿之形不复原，这一层记不采六，同甲11。26.153 两次盲拆对“前回”的时间解释有出入，送审后补核，见 7.5 节。
• 甲18：复原本狱神庙相逢的是宝玉与茜雪、小红，与巧姐的“遇难成祥”无涉；巧姐遭难、得救都在其后（第九十九、一〇〇回）。本项目只兑现了巧姐遇难成祥一层；狱庙相逢与巧姐相连一层，是情节布置不同，记不采五。底本标“该批”，审读者记得通行作靖本眉批；若是，即属参级，待影印本核。
• 甲19：四出戏的批，《牡丹亭》一条（18.117）含“大过节、大关键”，在候选之内；《一捧雪》《长生殿》《邯郸梦》三条（18.114 至 18.116）不含关键词，第四轮按批号补登，归本项（6.3 节）。四事俱落，但四出戏名在复原本里几乎不出现，所伏是事不是戏名。
• 甲20：“其事虽未行”的落定处有两处：第九十二回凤姐病中复述托梦二事，自道“我一亩也不曾置”；第一〇五回宝玉梦中，秦氏说“他没有听，我不怨他”。“遗簪”“更衣”诸文只见于 13.8 靖本（参级），作者自删，记不采六。
• 甲22：分三层。麝月在侧一层兑现（第九十七回）；宝钗一层只落在名分上，婚礼未成，与 20.72 同因，记不采四；“跌扇角口”一层，第三十一回已应，记分流二。
• 甲23：“供奉玉兄”一层兑现；“宝卿得同终始”一层，钗玉不以夫妇终始，出于本项目的主旨取舍，与 20.72、甲22 宝钗一层同因，记不采四（总篇 9.1 节所记不采之由即此）。它与 21.65 是否互异，留作读法：供奉在先、弃在后，也读得通。28.110 另有“宝玉忘情，露于宝钗，是后回累累忘情之引”一层：第三十回宝玉看龄官画蔷“不觉也看痴了”，第三十二回“出了神”拉住袭人诉肺腑，第五十七回“发了一回呆……一时魂魄失守”，记分流二。
• 甲24：残文的句读依总篇 4.6 节所述的校读；依别的读法，复原本不满足此批。这一条的兑现依赖于一种校读，照实登记。
• 甲25：撒手在第一〇六回，不在末回，记改兑现（总篇 9.1 节已登记为软失败）。“仍不出士隐梦”，指第一回士隐梦中僧道携石下世、了结公案这一框架，落在第一一〇回一僧一道携玉归青埂峰。“此虽眷恋，却破迷关，是何必削发”一句有两读：(a)“此”指湘莲，湘莲虽眷恋，却已破迷关，何必削发；(b) 整句说末回撒手者，他已悟，所以不必削发。批在第六十七回回前，紧承第六十六回湘莲挥剑断发，字面说的是湘莲，前两分句记分流二（第六十六回已应）。本项目取其含意：悟不在削发，末回撒手者不必削发，落第一〇五回宝玉为居士、不剃度。这一层是本项目的推读，只作甲12“弃而为僧”一层不采四的旁证。批末“中秋前引即三姐”一句读法未定，保留。

6.2 乙层

乙层共 216 条批（244 条候选减去甲层的 28 条），初分与现判如下：

初分	条数
乙-1：所指在前八十回之内	112
非前指	45
乙-2：所指跨八十回以后	59

现判	条数
前八十回内已应	117
非前指	45
已兑现	9
已兑现（否定）	2
已兑现（所指推定）	31
不采	1
移专题五	10
保留	1

乙-2 的 59 条，按批语本身说了什么，再分三类（第 7 节）：

类	条数	现判
一类：批语说了具体后事	13	已兑现 8，已兑现（否定）2（18.128、44.2），前八十回内已应 2（44.12、45.15），不采 1（20.72）
二类：只言伏线、后文，所指由本项目推定	31	已兑现（所指推定），未作兑现复核
三类：只点谶、或只把谶文对到人	10	移专题五
其余	5	1.77 改记非前指；3.79、19.109 改记前八十回内已应；79.5 由三类改回本文，已兑现；52.9 保留

几条要说明：

• 前八十回内已应的 117 条，有乙-1 的 112 条，其中 4 条是兑现复核时从原登记的“已兑现”改过来的：1.73“后之甄宝玉”，第二回雨村已讲起甄府那个学生，第五十六回甄家人说“老太太便叫作宝玉”，贾母接“偏也叫作个宝玉”；3.171 以鹦哥伏鸳鸯、以珍珠伏琥珀，第二十回起二人已同在贾母处；7.41“后文补抱琴”，抱琴第十八回出场；52.4“红玉既有归结，坠儿岂可不表”，两事都在前八十回内。另有 38.1 蒙府回前批，是 37.1 的重出，“起社出自探春意，作者已伏下兴利除弊之文也”，第五十六回，原记非前指，照规矩三改记。再有乙-2 的 44.12、45.15、3.79、19.109（见下）。
• 18.128（元春临别说“倘明岁天恩仍许归省”，庚辰夹批“便是不再之谶”）：“明岁不再”一层，前八十回内已应（第五十三回无归省，第五十五回“不独不能省亲”，是太妃欠安时的一时受阻，至第八十回未再归省）；“终身不再”一层，要到第九十一回元妃薨才落定。照规矩二，主处理记已兑现（否定）。这条批语自己说出了谶底，按 5.3 节留在本文，抄送专题五。
• 45.15（“又伏下后文，且又衬出后文之冷落”）：两层都在前八十回内。夜局一层，第七十三回查赌已应；冷落一层，批语说园中夜里人迹不绝、“不但不见寥落”，第七十八回蘅芜苑“房内搬的空空落落的”，第七十九回紫菱洲只剩“几个该班上夜的老妪”，宝玉“领略得如此寥落凄惨之景”，正是对照之应。冷落不是否定命题，5.2 节规矩二用不上；同 44.12 之例，记前八十回内已应，复原本第八十一、九十九、一〇一、一〇四回为延伸。
• 44.2（“闲闲一语伏下后文，令人可伤，所谓‘盛筵难再’”）：第七十五、七十六回中秋已“少了四个人，便觉冷清了好些”，众人“不禁有凄凉寂寞之意”；但“难再”是否定命题，照规矩二要看落定处，这两回只是前八十回内之端。主处理记已兑现（否定），落定于第九十一回查抄；复原本第八十三回贾母之死、第八十七回凤姐的困顿，是其前之衰。v0.3 送审稿仍只记已兑现，与 5.1 节不合，审读者指出后改标。
• 22.103（“此探春远适之谶也。使此人不远去，将来事败，诸子孙不致流散也”）：远适一层已兑现（第八十五回），抄送专题五；“使此人不远去”一层是批者的反事实与寄望，它不断言复原本该写什么，而是对家败之因的一种理解，本项目的叙述不作因果、报应的解释，记分流三。
• 3.79（第三回王夫人问“月钱放过了不曾”，甲戌侧批“不见后文，不见此笔之妙”）：第三十九回平儿已说破“这个月的月钱，我们奶奶早已支了，放给人使呢”，改记前八十回内已应；复原本第八十六回邢夫人查账翻出放贷，是本项目的延伸。
• 1.77（“八字正是写日后之香菱，见其根源不凡”）：八字写的是封氏“情性贤淑，深明礼义”，指日后香菱的品格，不是后事，改记非前指。
• 19.109（“补出袭人幼时艰辛苦状……自是又副十二钗中之冠”）：说的是袭人在又副册的位置，不是谶语。袭人入又副册见第五回，改记前八十回内已应，归情榜组；“之冠”是批者的评断，不作册次规格解。
• 4.63（“带叙出英莲一向之行踪，并以后之归结”）：“以后之归结”可以读作第四回本回写的为薛蟠所买，也可以读作全书后来的结局，两读并登。
• 35.26 蒙府回末总批“此回是以情说法，警醒世人”，因“千里”一词进了候选。它不指认后事，记非前指；它对全书的读法，是以情为必溺之河、以死为终局的警世之教，与本项目“叙述者不作报应的解释”不同，这一层记分流三。总篇所记“不承账”第五条即此。
• 三类中 37.54“又不脱自己将来形景”，批的是湘云的白海棠诗，原登记记作探春，是记错了人，移专题五时改正。

6.3 丙层

丙层 3,855 条，不逐条裁定。四轮补筛所得各条，登记如下。

前三轮八条。七条归入已裁定的组，并照所归之组补记其档：13.3 归甲20，已兑现；20.36“所以后来代任”归甲22，已兑现（第九十七回）；22.78“始终跌不出警幻幻榜”归甲05，改兑现；6.48“想亦在副册内者也”（平儿），第一一〇回平儿原有册页而未标册次，副册一层只算不冲突，作者补写后平儿一页落在副册（补记条〇〇三），记补写后的兑现；28.106 归 45.14（后部宝钗之文），已兑现（所指推定）；5.125、22.97“树倒猢狲散”两条归甲19：这句俗语出第十三回秦氏之口，说的是贾家，谁与何事俱在，按 5.3 节的测法两条都记已兑现（第九十一回），抄送专题五；5.125 批在第五回《飞鸟各投林》曲下。13.3 另有隐去天香楼一层，同甲20 记不采六。18.17 记不采八；其中“余未多及，想为……平儿等人无疑矣”一句承又副册而言，把平儿推在又副册，与 6.48 相抵，本项目取 6.48，这一层记不采九。

同组原还有 13.14“‘树倒猢狲散’之语，今犹在耳，屈指三十五年矣。哀哉伤哉，宁不痛杀！”它叹的是批者自己三十五年前所闻，不是书中后事，只叹不指，记非前指，不计漏网；这一行仍留在登记里，注“补筛所得，核后非前指”。

第四轮二十六条（未经兑现复核）：

• 归入已有之项。6.1 归甲02：“二进”在第三十九至四十二回，记分流二；“三进”是复原本第一〇一回回目；“巧姐之归着”，第一〇一回她自择随刘姥姥去，第一〇三回嫁板儿，已兑现。18.114 至 18.116 归甲19：《一捧雪》伏贾家之败，落第九十一回查抄；《长生殿》伏元妃之死，落第九十一回；《邯郸梦》伏甄宝玉送玉，落第九十六回，都已兑现，抄送专题五。
• 新立一组：贾芸。24.3 靖本“‘醉金刚’一回文字，伏芸哥仗义探庵”，24.61 庚辰“此人后来荣府事败，必有一番作为”。复原本第九十二回，贾芸借着给狱神庙神前上供的名目探狱；第九十四回，他领了差，揣着狱神庙的门路草图去寻醉金刚倪二，同蒋玉菡等计赚狱神庙。24.61 记已兑现。24.3 记半兑现：仗义一层兑现；“探庵”一层复原本没有写，把庵说成狱神庙，牵强，不取，记不采五。
• 42.1（第四十二回庚辰回前批“钗玉名虽两个，人却一身……请看黛玉逝后宝钗之文字便知余言不谬矣”）：钗黛一身是批者的读法；本项目取黛玉为核心、二人各自具足（总篇 6.3 节），记理解不同。第一〇八回黛玉入宝钗梦，可作回应的候选，不作兑现（第 9 节）。
• 25.104（“通灵玉除邪，全部百回只此一见，何得再言”）：否定命题。复原本没有第二次通灵玉除邪，记已兑现（否定）。这一条同时是回数材料（总篇 3.5 节）。
• 所指推定六条。25.63 甲戌“二玉事在贾府上下诸人即看书人批书人皆信定一段好夫妻……岂具不然，叹叹”，45.12 庚辰黛玉自说夫妻而云“画的”“扮的”，“暗隐不吉之兆”：两条推为二玉终未成夫妇，落第九十五回黛玉泪尽、第九十六回“私下结发那一桩，贾府里认不得这门亲”。4.35 甲戌“早为下半部伏根”，批在门子讲护官符四家“一损皆损，一荣皆荣，扶持遮饰，俱有照应”一句下，推为四家连败，落第九十一回查抄、第一〇三回薛蟠锁系。46.8 庚辰“后日更有各自之处也”，75.15 庚辰“总是将散之兆”，推为家败人散。72.6 庚辰“江淹才尽之兆”，批在凤姐梦中被夺锦下，推为凤姐之衰，落第九十回被休、第九十四回末殁。
• 76.3（贾母说“可怜你公公已是二年多了”，庚辰双行夹批“不是算贾敬，却是算赦死期也”）：复原本第九十一回贾赦锁拿收监，第九十二回定谳，革去世职，发往宁古塔效力赎罪，贾琏、邢夫人随行；其后贾赦没有再出场，第一〇一回只由薛蝌一句“大老爷同琏二爷远发了关外”带过，死期未落。作者裁定补写，出补记条〇〇二，现已写入第一〇八回，记已兑现（6.4 节）。
• 好了歌注句下批十一条。1.168“宁、荣既败之后”与 1.170“雨村等一干新荣暴发之家”，批语自己说出了谶底，按 5.3 节留在本文：1.168 已兑现（第九十一回查抄），1.170 已兑现（第一〇七回雨村骤贵），都抄送专题五。其余九条，1.169“潇湘馆、紫芸轩等处”，1.171“宝钗、湘云一干人”，1.172“黛玉、晴雯一干人”，1.174“熙凤一干人”，1.175“甄玉、贾玉一干人”，1.178“柳湘莲一干人”，1.180“贾赦、雨村一干人”，1.181“贾兰、贾菌一干人”，只把谶文对到人或处，移专题五；1.183“太虚幻境青埂峰一并结住”另属全书框架收束之指认，只有处、没有谁，随好了歌注的结构问题交专题五，不作纯地点对应；原记二类的 1.177“言父母死后之日”同属此组，一并移去。专题五的收录应补入好了歌注。

报存三条：75.3“缺中秋诗俟雪芹”，22.104“此后破失，系再补”，22.106“此回未成而芹逝矣”。

其余丙层批语，不逐条裁定。

6.4 补记条

审计查出、作者认可、但不即时动稿的条目，登记为补记条，只记落点与约束，不拟正文。补写之后，登记照实记为补写后的兑现：审计时未兑现 → 作者裁定 → 补写 → 改登记，不追溯为原稿已经满足。

补记条〇〇一：甲07 的“寒冬噎酸虀”，落第一〇二回柳湘莲与宝玉土地祠分饼一场，与“雪夜围破毡”同场。复原本第一〇二回已写入，已结。

补记条〇〇二：76.3“算赦死期”，补写贾赦死于发往宁古塔的途中。落第一〇八回，宝钗北行途中风雪封路的那一处荒驿，由驿卒随口一句带出。当初立的约束：

• 不报丧。贾家所知，止于第一〇一回薛蝌那句“大老爷同琏二爷远发了关外”，此句不改；第一〇九、一一〇回也不补。贾赦之死只落在宝钗这一条路上，同回宝钗死于雪中，这件事也就没有人再带回去。
• 只演不论。驿卒不点名、不评人，不写死状；宝钗听见，只作一个动作，不说评语；叙述者不作报应之语（总篇 4.5 节）。雨村同走这一段路，不牵连到他。
• 时序与路线。宝钗一路“越往北”，原文没有点目的地；这一句等于定了她也走出关的路，驿站须在男女两队分路的岔路口之前。
• 贾琏、邢夫人照第九十二回随行，驿卒至多说有随行的儿子与妇人，不写他们其后下落。
• 只加一两句，不动路引、最后一丸冷香丸、《辞尘网》《钗头凤》、黛玉入梦与埋簪诸段。

写入的只有驿卒一句。补写是一处小范围替换（r53，2026 年 9 月 27 日补）：原稿“记人数，宝钗向他借了两张。”改作“记人数，看了关文，随口道：「前年也押过一个京里姓贾的老犯官，死在这一带。」宝钗听了，向他借了两张。”其后“驿卒说纸须还，宝钗将头上一根细线拆下，替他缠好将断的笔管，驿卒便摆摆手去了”诸句，都是原稿。驿卒没有点名，也没有评语；宝钗听了，只有借纸一个动作，缠笔是应驿卒之求；借去的两张旧纸，就是随后写《辞尘网》的那两张。“前年”与复原本的年序相合：第九十二回“早又开春”，贾赦定谳起解；第九十七回婚期在四月，第九十八回贾政之死在其后的暑天，与起解同在一年；第一〇八回同回称贾政之死也作“前年”。以上约束都守住了，补记条〇〇二已结。

补记条〇〇三：6.48“想亦在副册内者也”（平儿）与甲03（6.49“观警幻情榜方知余言不谬”）的副册一层。本项目的名册里，平儿在副册；但第一一〇回情榜，平儿那一页夹在“余下册页极多”一段里，与鸳鸯、紫鹃诸页一串带过，没有写属于哪一册。复原本不列全部情榜，读者看不出平儿在副册。甲层核对者当时已记此疑（附件05），v0.3 送审后审读者再指出。作者裁定补一笔，不改记不采（附件11）。

补写是第一一〇回的一处小范围改动（r53，2026 年 9 月 27 日补）：

• “余下册页极多”一段，删去“平儿页上有杏花·摇篮·半本账。”一句，平儿不再夹在一串册页里一闪而过；
• 其后“宝玉没有抬头，只将副册又揭开一页。”改作“宝玉没有抬头，只将副册又揭开一页，却是平儿，画着杏花·摇篮·半本账。”

下一段“副册页动，空中便另起一支”是原稿，这一支曲随之成了平儿的。册页上的杏花、摇篮、半本账，原稿本有，只挪了位置。长榜名单保留混排，没有重排。补写之后，6.48 与甲03 的副册一层记补写后的兑现，不追溯为原稿已经满足；18.17 把平儿推在又副册的一层，随之记不采九（6.3 节）。补记条〇〇三已结。

6.5 合计

层	批数	现判
甲	28（25 项）	已兑现 15 项（其中所指推定 2 项），改兑现 5 项，半兑现 3 项，非前指 1 项，不采 1 项
乙	216	前八十回内已应 117，非前指 45，已兑现 9，已兑现（否定）2，已兑现（所指推定）31，不采 1，移专题五 10，保留 1
丙	3,855	补筛漏网 34 条已登记：已兑现 13，已兑现（所指推定）7，已兑现（否定）1，改兑现 1，半兑现 1，理解不同 1，不采 1，移专题五 9；补筛所得、核后非前指 1（13.14）；报存 3；整理者校注 2；其余 3,815 条不逐条裁定。补登各条都未经盲拆复核（7.6 节）
合计	4,099

照总篇的约定，这些数字随审计续做而更新，以随文数据为准。

7 兑现复核

7.1 查的是什么

兑现复核查的是本项目自己的登记：本项目说某条批语“已兑现”，复原本的情节层是不是真的满足了那条批语所说的事。它不碰批语的真伪。可能的结果只有两种：维持，或本项目的登记要改。

总篇称这一步为“反向审计”，说它“把批语拆成可证伪的命题，再用复原本文本去证伪”。那句话容易读成要证伪批语，不合第 1 节的立场。本文起改称“兑现复核”，总篇下一版随改。

7.2 范围

登记为兑现的，并不都能复核。复核时，乙-2 登记为“已兑现”的有 62 条，真正说了具体后事、能拆出命题的只有 13 条；只言伏线、所指由本项目推定的有 34 条，这一类查不了批语，只能改登记为“所指推定”；谶语类 11 条，移专题五；另 4 条原登记有误，实为前八十回内已应。（此后审读与第四轮补筛又改了几条：二类的 1.77、3.79 改记（6.2 节），1.177 随好了歌注一组移专题五，二类现为 31 条；三类的 19.109 改记前八十回内已应，79.5 改回本文，加上 1.177，三类现为 10 条。）

所以本轮的范围是：甲层登记为兑现的 23 项，加乙-2 一类 13 条，共三十六个核查单元（甲层以审计项计，一项可含两条同文的批）。二类不作复核；第四轮补登的各条是复核之后才查出的，也未复核。

在此之前，已做过六组定向测试（总篇 9.3 节），并在写总篇时对几条作过复核。

7.3 做法

盲拆。先请一位分析者，只看批语与所批正文（批语前约 260 字、后约 120 字），不看复原本，也不看审计表，把每条批语对后事说了什么拆成命题，写成“谁｜何事或何种处境｜何时｜是否含否定或未然”；并用前八十回全文查，这条批语所说的后文是否在前八十回之内已经出现。这样做，是为了不把本项目的答案读进题目里。

复核。再分两组核对：甲层 23 项一位核对者，乙-2 一类 13 条一位核对者，每条只经一位核对者。核对者拿盲拆出的命题逐条对复原本，逐字摘出引文，按以下准则判定：

1. 身份与处境是否兑现；批语的修辞不作逐项规格核（如“缁衣乞食”，见下）；
2. 同名异指，按语义复核；
3. 时序：批语说的先后，复原本里是否成立；
4. 施受：谁对谁做；
5. 否定与未然（如“其事虽未行”）也要核；
6. 整段读完再判；
7. 盲拆指出前八十回内可能已应的，判断是否属实。

核对者的建议，由作者逐条裁定。

执行者。盲拆者与两位核对者都是新开的 AI 会话，模型为 Claude Opus 5.5，与撰写登记的归并方是同一个模型，日期 2026 年 9 月 27 日（UTC）。盲拆者的工具调用记录显示，它只读了盲拆材料与前八十回底本（另列过一次所在目录，见附件12）。“没有给它复原本”有记录可查；“它此前从未接触本项目的答案”只能说到新会话里没有本项目的上下文为止，模型相同，独立性有限。提示词与运行记录见附件12。

一处须照实交代的。盲拆提示词在“修辞与规格”一栏举的例子，是“缁衣乞食只是出家人的代称”，这正是作者对 22.105 的裁定，等于把答案带进了题目。重拆时已改（7.5 节）。两位核对者的提示词，在复核准则第一条里举了同一个例子；核对者看得到现登记与作者既有的裁定，并被要求照裁定执行、有疑另写（设计如此：复核查的是登记是否属实，不是替作者重裁），所以在 22.105 这一点上，核对并不独立。22.105 的这条修辞裁定，照登为作者的读法：汉传佛教自隋唐农禅并重（百丈怀海“一日不作，一日不食”之说，见《五灯会元》卷三百丈怀海章），托钵乞食不是庵中常制，复原本写惜春在正觉庵清苦度日，只核身份与处境。这是一般的制度常态，只说明作者为何不取“乞食”的字面，不据以断定批语字面必非实指；它也不作通用的豁免。

所据。复原本第八十一至一一〇回四个合集，各记校验值（附录三）。

7.4 结果

三十六个单元中，维持二十个，改登记十六个：

批号	原登记	复核后	理由
1.185（甲01）	已兑现	非前指	批的是第一回甄士隐“走罢”随道人去，是本回的事，不是前指；原先因“悬崖撒手”一语挂到了第一〇六回
8.140（甲05）	已兑现	改兑现	复原本的情榜只列女子；“情不情”落在第一一〇回回目与薄命司那一声断喝，不在榜上
19.36（甲07）	半兑现	已兑现	补记条〇〇一已写入第一〇二回，原登记过时
26.120、26.153（甲15）	改兑现	已兑现	射圃从原拟的第九十四回挪到第八十七回，只是本项目内部挪回次，内容全写了
42.6（甲18）	已兑现	半兑现	见 6.1 节
67.1（甲25）	已兑现	改兑现	撒手不在末回；“士隐梦”的兑现位由第一〇八回改为第一一〇回，第一〇八回士隐所说是雨村的“权势梦”，同名异指
21.65（甲12）	改兑现	改兑现（说明改）	“得而弃之”改说“定而弃之”；不采一层注型
28.1（甲23）	半兑现	半兑现（补）	补兑现位：茜香罗所伏的袭人嫁琪官在第八十一回，供奉补第九十四回；不采一层注型
7.40（甲04）	已兑现	已兑现（补）	夹批一层改记前八十回内已应
20.45（甲22）	半兑现	半兑现（补）	“跌扇角口”一层改记前八十回内已应（第三十一回）
18.128	已兑现	前八十回内已应	第五十五回已有“不独不能省亲”；复原本第九十一回元妃薨是延伸
20.72	已兑现	不采	批说“后文成其夫妇时”，还有“谈旧之情”；复原本婚礼未成，二人未以夫妇相处
26.81	已兑现（第九十六回）	已兑现（第一〇四回）	第九十六回的潇湘馆还是三月青竹；冬日查园，竹梢黄了，帘子撤了，在第一〇四回
43.20	已兑现（第九十、九十二回）	已兑现（第九十四、九十六回）	“短命”要落在凤姐之死；她死在第九十四回末那一夜，第九十六回平儿方知
44.12	已兑现	前八十回内已应	“黄黄脸儿”所伏病根，第五十五回“面目黄瘦”、第七十二回已应；复原本延伸至死
45.15	已兑现	前八十回内已应	所伏的园中夜局，第七十三回已查出；园中冷落改记第八十一、九十九、一〇一、一〇四回

另有几条维持而补注：甲06 不采一层注型；甲09、甲16 兑现位补第九十二回；甲19 另三出戏的批分列批号 18.114 至 18.116；甲20 兑现位补第一〇五回；22.105 的说明由“自耕自食”改作“山庵清苦”；16.14、44.2 补兑现位。

改动升降两向都有：降档五条（1.185、8.140、42.6、67.1、20.72），升档两条（19.36、26.120），改记前八十回内已应三条（18.128、44.12、45.15），改兑现位两条（26.81、43.20），其余四条（21.65、28.1、7.40、20.45）改说明、补兑现位，或把其中一层改记前八十回内已应。正向登记的毛病，主要是偏宽：只要人在、事在，就记兑现；没有先查批语说的“后文”是否在前八十回内已经应了；也有因关键词而挂错位的。

此后送审，审读者又指出几条一处【】里有几层话、却只给了整条一个判断，于是立了 5.2 节的规矩，又改了四条：19.80 由非前指改为改兑现（情榜两评）；18.128 由前八十回内已应改为已兑现（否定），落定于第九十一回；45.15 分层，冷落一层已兑现；38.1 照 37.1 改记。v0.2 送审后，又按审读意见改了：45.15 冷落一层改记前八十回内已应（第七十八、七十九回），主处理的名目与第一次复核后的登记相同，理由却不同：第一次复核判冷落一层在第七十九回“只见端倪”，主张落在复原本，本版依审读意见把第七十八、七十九回读作应；甲02、甲04 加“所指推定”标注；甲23“宝卿得同终始”一层改注不采四；22.103“使此人不远去”一层改记分流三。v0.3 送审后，又改了三处：甲12“弃而为僧”一层由不采九改记不采四；44.2 改标已兑现（否定），落定于第九十一回；甲03 与 6.48 的副册一层，甲03 复核维持之后，审读者指出正文未写明，作者补写第一一〇回（补记条〇〇三），记补写后的兑现，18.17 的平儿一层随之记不采九。现登记以第 6 节为准。

加上此前查出的：甲20 曾拿林家祖茔的词频当贾家祖茔的证据；3.58 兑现位误记；22.105 曾以“乞食”字面作规格而判半兑现；甲24 曾拿残文硬断而判不承；甲12、甲22、甲23 写总篇时改档。靠复核改动过的登记，已不是少数。这正是复核值得做的理由。

7.5 重拆

v0.1 送审时，审读者发现数据里“所批正文”两栏有十三行提取有误：十二行混进了相邻批语的残段，一行（16.74）前文截短。病因是先按字数开窗、再去批语，窗口边上切断的半条批语去不掉。盲拆材料的上下文用的是同一种办法，只是窗口较长：40 条里有 15 条（16 处），窗口边缘混进了相邻批语的残段。紧贴批语的正文多数是对的；只有 21.4、13.89、26.153 三条回前、回末批，前后本来没有正文，窗口里取到的全是别的批语。

于是作者裁定，对这 15 条重做：

1. 修正提取。先把整回的批语剥净，再截取；截到回首标“回首”，含回目，截到回末标“回末”。全表“所批正文”两栏也照此重提，十三行与审读者的复算逐条相符。
2. 另开盲拆者。换用另一个模型 Claude Sonnet 5，新会话；工具调用记录显示只读了重拆材料与前八十回底本。提示词里带了作者裁定的那个例子，改为中性说法。
3. 比对。新旧两份命题逐条比对，事先只立三种出入：新拆出旧版没有的后事命题；同一命题的人、事、时间、否定有别；对“前八十回内是否已应”的判断不同。
4. 核对。有出入的，另开一个 Sonnet 5 会话核对复原本，再由作者裁定。

当时比对，以 7.40、13.89、22.105、28.110 四条送核。v0.2 送审后，审读者指出 26.153 的时间解释也有出入：旧拆说“前回”不能限在第二十六回以前，重拆读作第二十四至二十六回一带，并说射圃“本应与之相近”。归并方原把它当作小差别，没有送核，这不合事先所立的三种出入，于是另列补核。合计无出入 10 条，有出入 5 条。

核对者建议 7.40、22.105 维持，作者照准；核对者看得到现登记里作者对 22.105 的裁定，这一条上仍不算独立。26.153 的补核者（另一个 Sonnet 5 新会话）查得：四样侠文在第二十四、二十六、二十八、四十七回，“前回”不能限在第二十四至二十六回；批语只叹射圃文字迷失，没有规定射圃在第几回；第三十一回批明说射圃在“后数十回”。建议维持，作者照准；重拆所加的邻近回次一层，不采作复原约束。补核输出见附件14。另两条，核对者的建议没有采纳：

• 13.89，核对者建议删去第一〇五回兑现位，理由是第一〇五回只有贾母灵柩归金陵祖茔一事。但第一〇五回宝玉梦中，秦氏说“他没有听，我不怨他”，正是“其事虽未行”的旁证。核对者以“祖茔”“家塾”“供给”“置田”“则例”诸词检索第一〇五回，没有读到秦氏那一段；而交核对的登记只写了“兑现位另记第105回”，没有写出依据，依据是 v0.2 才补进说明的，这一半是登记之失。第一〇五回兑现位保留。
• 28.110，核对者以复原本全文没有“忘情”二字，建议把“后回累累忘情之引”一层记为未兑现。但“后回”不必在八十回以后，前八十回里已经累累可见（6.1 节甲23）。这一层补登为前八十回内已应。

两处都出现以关键词命中代替语义判断的问题；28.110 还误把“后回”限为八十回以后。换了模型，同一个病根照样出现。核对输出里另有几句时序与计数的错话，原样留存，勘误列在附件12 第六节。

7.6 局限

• 二类没有复核。乙-2 中 31 条只言伏线、所指由本项目推定的，只改了登记，没有复核；它们兑现的是本项目的推定。
• 甲08 没有复核。19.80 在兑现复核时登记为非前指，不在范围内；v0.1 送审后按裁一改记改兑现（情榜两评），没有经过盲拆复核。
• 补登的各条没有复核。第四轮补登的 26 条、前三轮补登的 8 条照组补档，以及按 5.3 节改回本文的 79.5，都是复核之后才定的，未经盲拆。其中登记为兑现、又能拆出具体命题的，与一类同型，下一版先核，共十五条：13.3、20.36、22.78、5.125、22.97、6.1、18.114、18.115、18.116、24.3、24.61、25.104、1.168、1.170、79.5。最先核的是 79.5（改回本文，所凭是一个“诔”字）、25.104（否定命题，须全扫第八十一至一一〇回）、24.3 与 24.61（新立之组）、22.78（改兑现）。所指推定各条同二类；移专题五各条不在本文；18.17 已在总篇 9.3 节测过；42.1、13.14 无兑现可核；76.3、6.48 是补写后的兑现，只核约束与补写之句（6.48 原在此列，补写后移出）。
• 盲拆与复核都由 AI 完成。原盲拆者、原核对者与归并方同为一个模型，重拆者与重核者换了模型，独立性都有限。所有改登记，都经作者逐条裁定。
• 修辞裁定是作者的读法。22.105 的“缁衣乞食”作出家人的代称，是作者的读法，照登，不作通用豁免；所举佛教制度的常态，只说明为何不取字面。
• 复核所据是复原本的一版。复核用的是附录三所列四份合集。其时第九十六至一〇〇回已各有更晚的封稿（r54），字句有出入；审读者抽核登记所引各句，在 r54 本都在。复原本还在修订；只改字句的，登记不变；改了情节层取舍的，登记随改；登记所引各句，下一版逐句对最新封稿。

8 误判的病根与工序

已查出的误判，归起来有几种病根：

1. 拿残文当定本（甲24 初判）；
2. 拿词频当语义（甲20 的祖茔；重拆核对时的 13.89、28.110，其中 13.89 登记当时没写依据，也有份）；
3. 拿批语的修辞当实务规格（22.105 的“缁衣乞食”初判）；
4. 引曲文只读半支（总篇 9.1 节所记）；
5. 凭关键词定层，不先回看所批正文（1.185 挂错位）；
6. 定为跨八十回前指之前，没有先查前八十回（1.73、3.171、7.41、52.4、44.12、3.79、45.15 冷落一层等）；
7. 记错人（37.54）；
8. 补记条落实以后，没有回改登记（19.36）；
9. 一阶词表缺单字“伏”，缺“后来”“后日”“兆”；二阶筛只扫了样本（24.3、24.61、6.1、45.12 等）；
10. 解谶的句下批不含关键词，按词筛不出（好了歌注一组）；
11. 一处【】里有几层话，只给整条一个判断（19.80、18.128、45.15、38.1）；
12. 提取上下文时先开窗后去批，窗口边上混进别的批语；
13. 盲拆提示词里带了本项目的裁定；
14. 登记只写兑现位，不写依据（13.89）；
15. 比对新旧命题时，把时间解释的变化当成小差别，没有送核（26.153）；
16. 只叹不指的批语，因关键词进了漏网（13.14）；
17. 定型时把本项目对批语的推读当作批语字面（甲12 的不采九）；
18. 核对者记下的疑点，裁定时没有逐条回应（甲03 的册次）；
19. 立了规矩，没有回扫已登记的各条（44.2 说明里已认作否定命题，主处理却未标）。

由此立的工序：

1. 残文先校后裁；
2. 同名异指以语义复核（祖茔、庵、麒麟、册、梦）；
3. 修辞不作规格解，只核身份与处境，但须说出为何是修辞；
4. 整支、整段读完再引；
5. 定层之前先回看所批正文；
6. 定为跨八十回前指之前，先在前八十回里查一遍；
7. 补记条落实，回改登记；
8. 终版表回归对账；
9. 缺字批人工过；
10. 兑现复核先盲拆命题，再对复原本；
11. 二阶筛扫全量，不只扫样本；
12. 单字前指词（伏、兆）与“后来”“后日”另筛；
13. 解谶的句下批按位置过；
14. 一条批里有几层话，分层登记（5.2 节）；
15. 上下文先整回去批，再截取，并断言不含【】；
16. 盲拆提示词里的例子，不得取自本项目的裁定；
17. 登记写兑现位，同时写依据；
18. 比对新旧命题，照事先所立的三种出入逐栏过，不自行判为小差别；
19. 定为漏网之前，先按口径分出只叹不指者；
20. 记两批互异之前，先确认两条批语字面相抵；
21. 核对输出里的疑点，裁定时逐条回应，采与不采都写明；
22. 新立一条规矩，回扫全部登记。

9 批语里的另几种下半部

把不采的批语、半兑现里没采的那一半、两批互异时舍去的那一边、改兑现时变通掉的那一处，以及理解不同的几条合起来看，按这些批语读，可以得到另几种下半部的轮廓。这一节把它们逐一写明，说明本项目为什么不采。本节不重建这些下半部，也不判它们对错；每一种都分开写：批语明说了什么，按这一路可以读出什么，本项目怎样处理。

一、宝玉娶宝钗，成其夫妇，麝月在侧；后来宝玉“弃而为僧”；琪官与袭人供奉“玉兄宝卿”，得同终始。

• 批语明说：20.45“袭人出嫁之后，宝玉、宝钗身边还有一人”；20.72“后文成其夫妇时”，还有“谈旧之情”；21.65“若他人得宝钗之妻、麝月之婢，岂能弃而为僧哉”；28.1“后回与袭人供奉玉兄宝卿得同终始”。
• 按这一路读：宝玉与宝钗真成了夫妇，婚后同处，麝月在侧，由琪官、袭人供奉；宝玉后来弃而为僧。21.65 说弃，28.1 说同终始，两批可读作先后相承（供奉在先，弃在后），也可读作出于两稿。
• 分歧：本项目所取的，是宝玉自择出走，不是得而后弃；宝钗最大的难题是名分，不是被供奉的生活；宝玉悟后回到人间，为居士而不剃度（总篇 6.4、7.2 节）。记不采四；67.1“是何必削发”字面说的是湘莲，其含意只作旁证（6.1 节甲12、甲25）。
• 本项目：宝玉在婚约已定、婚礼临头时走开，宝钗的名分事后追定；麝月在侧、琪官袭人供奉宝玉，各兑现一半。

二、钗黛一身。

• 批语明说：42.1“钗玉名虽两个，人却一身，此幻笔也……请看黛玉逝后宝钗之文字便知余言不谬矣”。
• 按这一路读：这一条首先是批者对第四十二回的读法，钗黛“合而为一”，可以是人物结构与文体上的解释，不一定是一件事。它所期待的，是黛玉逝后宝钗的文字能印证这种读法。
• 本项目：读黛玉为核心，二人各自具足（总篇 6.3 节）；《情情赋》说：“尔自为尔，非为吾之宗极；吾亦自为吾，不待尔而后明。”记理解不同。黛玉逝后，宝钗确有文字，第一〇八回黛玉入宝钗梦，说的是“那是你自己写的。这一笔，没有人替你落”。这一场可作回应的候选，但它写的是二人各自，本文不拿它来兑现此批。

三、宝玉以“情不情”入情榜；情榜列出全谱，定出正、副、再副及三四副。

• 批语明说：8.140“按警幻情榜，宝玉系‘情不情’”；19.80“后观《情榜》评曰‘宝玉情不情’，‘黛玉情情’”；18.19“至回末警幻情榜方知正、副、再副及三四副芳讳”。
• 按这一路读：情榜男女同列，册次定下。两评在批语里是褒语。
• 分歧：情榜是活的，不定死（总篇 6.2 节），记不采四；“情不情”“情情”之义，本项目另有读法，记理解不同。
• 本项目：第一一〇回写情榜，只列女子，不列全谱；“情不情”落在回目与薄命司的断喝里。

四、狱庙相逢之日，才知巧姐“遇难成祥，逢凶化吉”的伏线。

• 批语明说：42.6“狱庙相逢之日始知‘遇难成祥，逢凶化吉’实伏线于千里，哀哉伤哉！此后文字不忍卒读。”
• 按这一路读：狱神庙的相逢与巧姐的得救连在一起。
• 本项目：狱神庙相逢的是宝玉与茜雪、小红；巧姐之救在其后，由刘姥姥一行完成，写作偶遇，不作阴功之报（总篇附录五《留余庆》一行）。这一处是情节布置不同，不涉主旨，记不采五，此批只兑现一半。

五、芸哥仗义探庵。

• 批语明说：24.3 靖本“‘醉金刚’一回文字，伏芸哥仗义探庵”。
• 按这一路读：贾芸仗义，与醉金刚倪二一回相伏，去探的是一座庵；探的是谁，批未明言。
• 本项目：贾芸的仗义写在狱神庙，第九十二回探狱，第九十四回同倪二等计赚狱神庙；“探庵”一层没有写，把庵读作狱神庙，牵强，不取。这也是情节布置不同，记不采五，半兑现。

六、撒手在末回。

• 批语明说：67.1“末回‘撒手’，乃是已悟”。
• 按这一路读：宝玉悟而撒手，被安排在全书末回；批语未限定它是否为末回最后一事。
• 本项目：撒手在第一〇六回，末回写情榜与玉返青埂峰，记改兑现。宝玉撒手之后还有几回人间的日子，是本项目所写悟后的第三层（总篇 6.4 节）要的。

七、探春若不远去，家败后子孙可不流散。

• 批语明说：22.103“此探春远适之谶也。使此人不远去，将来事败，诸子孙不致流散也，悲哉伤哉！”
• 按这一路读：探春之去，是家败时子孙流散的一个缘故；她若在，家事或可收拾。这是批者的一个反事实，也是一种寄望。
• 本项目：远适一层兑现。实际远嫁并没有实现这一反事实的前件，所以复原本写探春远嫁，不构成对它的反驳，也不能证明它为真。“使此人不远去”一层，不断言复原本该写什么，而是对家败之因的一种理解；本项目的叙述不作因果、报应的解释，照写散而不灭，记理解不同。

八、以情说法、警醒世人。

• 批语明说：35.26“此回是以情说法，警醒世人……爱河之深无底，何可泛滥？一溺其中，非死不止”。
• 按这一路读：这一条不是前指，但它读全书为警世之教，情是必溺之河，一溺其中，非死不止。
• 本项目：叙述者不作报应的解释（总篇 4.5 节），记理解不同。

九、又副册三断词作晴雯、袭人、香菱；平儿在又副册。

• 批语明说：18.17“又有又副册三断词乃晴雯、袭人、香菱三人，余未多及，想为金钏、玉钏、鸳鸯、苗云（下有按语一句，略）、平儿等人无疑矣”；同回畸笏眉批（18.19，即甲06）自承“是处引十二钗总未的确，皆系漫拟也”；6.48 说平儿“想亦在副册内者也”。
• 按这一路读：又副册三断词乃晴雯、袭人、香菱，香菱不在副册。这与第五回正文香菱在副册相抵，或许是第五回的另一稿，或许是批者的漫拟。又副册其余诸人，批者想为金钏、玉钏、鸳鸯、苗云、平儿等，平儿在又副册；6.48 却说平儿“想亦在副册内”，两批所说的册次不同。
• 本项目：三断词从正文，记不采八；平儿取 6.48，第一一〇回补写平儿一页在副册（补记条〇〇三），18.17 这一层记不采九。

十、原稿已失或作者自删之文。

• 批语明说：狱神庙、射圃诸回的原稿迷失（20.20、26.24、26.120、26.153）；“遗簪”“更衣”诸文，据 13.8 批者自述，曾命芹溪删去（只见于靖本）。
• 按这一路读：下半部原有狱神庙一大回、卫若兰射圃之文，前文原有秦可卿“遗簪”“更衣”诸事。这些不是另一种设计，是失去或删去的文字。
• 本项目：或另立（狱神庙、射圃），或不写（遗簪、更衣），都不复原其原形，记不采六。

这几种下半部，本项目都不采，或只采一部分。它们照录在这里，说明两件事。一是按批语本身，曹雪芹的下半部很可能不止一种设计，批者所见所想各有不同；这是一种可能性，本文不据此断言历史上确有这样几部完整的稿子。二是本项目所写，是依“人是目的”这一判准，在证据仍然允许的读法之间择出来的一种；换一个判准，这里的许多“不采”就可以改成“采”。批语在这里不是被驳倒的对手，而是另几种可能的证人。总篇说“情榜是活的”，对复原本自己也是如此。

10 结语

审计的结论可以说得很短：本项目已识别的、前八十回里指向后事的批语，逐条留了痕；采了哪些，怎样采，不采哪些，为什么不采，读者都可以查。兑现复核又把本项目自己的登记查了一遍，改了不少；补筛又查出一批漏看的，其中有熟知的几条；送审之后，审读者与重拆又逼出了几处分层与几处病根。没有一条批语在这里被判为错；被检验的，只是本项目。

附录一　数据说明

随文发布 专题一_脂批前指审计数据_v0.4.csv，UTF-8（带 BOM）编码，一处【】一行，共 4,099 行，22 栏。有登记的行共 284 行（甲、乙两层 244 行；丙层补筛登记 35 行，即漏网 34 行与核后非前指 1 行；报存 3 行；整理者校注 2 行），其余 3,815 行不逐条裁定。字段：

字段	内容
批号	回.序，须作文本读（6.10 与 6.1 是两条）
回、序	回次；该回之内第几处【】（整数）
标签	底本所标的原样标签，如“庚辰双行夹批”“甲夹批”“该批”
本名	归并后的本名
批型	侧批、眉批、夹批、双行夹批、回前批、回末批、总批等
批语	全文
所批正文前、所批正文后	批语前二十字、后十二字；先把整回的批语剥净再截取，截到回首标“回首”（含回目），截到回末标“回末”
初分层	甲、乙-1、乙（非前指）、乙-2、丙；记初分，现判看“处理”
审计号	甲层各项的号
处理	5.1 节各档与去向（主处理）
分流与不采	其余各层的编号与说明（5.4 节）
兑现位	复原本回次，阿拉伯数字，几回以顿号分隔，起止写作“至”（如“81至110”）；无兑现位的记“—”
说明	理由与备注
复核	判法：已实际参加兑现复核者（三十六个核查单元，以及 7.2 节定范围时查出改档的 1.73、3.171、7.41、52.4 四条），按当次核查记“复核维持”或“复核改登记”，其后作者裁定的变动记入“复核记录”；未参加者，才按当前处理区分“未复核”与“不复核”。五值之一：复核维持（在兑现复核范围内，主处理与兑现位未因复核改换；只改说明、补兑现位、注型、改记一层的，也在此类）；复核改登记（主处理改档，或兑现位改换）；未复核（登记为兑现而未经盲拆复核）；不复核（非兑现登记：前八十回内已应、非前指、理解不同、不采、保留、移专题五、报存、排除）；补写后兑现（76.3、6.48，未参加兑现复核，兑现出于作者裁定后的补写；甲03 参加过复核，按判法仍记复核维持，补写记入复核记录）。无登记的行空
复核记录	复核栏的细节：改档来由、补兑现位、重拆有无出入、补筛所得、裁定号等。7.4 节所说“改登记十六个单元”，口径比复核栏宽，含改说明、补兑现位与改记一层
抄送专题五	解谶之批留本文登记而抄送专题五者，记“是”
一阶命中词	终版 45 词中命中之词
补筛轮次	轮一至轮三、丁一至丁四；只标补筛所得的登记各行，丁一在全量中的 316 条命中不逐行标，可用附件13 重算
抽样、二阶命中轴	382 条样本记“是”；二阶命中轴只标样本内二阶筛命中的 27 行，记“命”“时”“叹”

本名归并规则。

• 标签以“甲戌”或“甲夹”“甲眉”“甲侧”“甲嗯”起首者，归甲戌；“甲嗯”只 1.158 一例，标签作“甲嗯和了人风。蒙侧批”，疑有讹；
• 以“庚辰”“庚讲”起首者，归庚辰；
• 以“己卯”“己双”起首者，归己卯；
• 以“蒙”起首者，归蒙府；底本误作“梦”者，一并归蒙府；
• 以“靖”起首者，归靖藏；
• 第一回卷首戚蓼生序文，归戚序；
• 标“校者注”者，是整理者的校注，单列为整理者校注；
• 其余，如“该批”“脂批”“题曰”或无标签者，归无本名。

一处【】内含几家同处之批的，按起首的标签归。标签讹写、按规则归入无本名的边例：17.125“前庚辰双行夹批”、37.25“辰夹批”（显属庚辰），35.16“郊批”、16.151“列”（未明）。本文照规则计，不另改归。

底本一处异常。第二十四回 24.9 庚辰眉批之后，底本多出“笏叟。】”四字，是前一条批语末三字的重出，全书仅此一处无头之“】”；提取上下文时剔除，批语计数不受影响。

附录二　筛词与抽样

终版一阶词表（45 词）：伏线、伏脉、伏下、又伏、总伏、伏后、伏笔、千里、后文、后回、后数十回、后半部、数十回后、三十回后、后三十回、迷失、无稿、情榜、谶、狱神庙、狱庙、射圃、卫若兰、悬崖撒手、有始有终、寒冬、雪夜、将来、日后、后书、后部、观者记之、末回、后之、遥与、后事、结果、大过节、大关键、大节目、同终始、他日、末世、归结、后半。匹配前先去掉“伏侍”“伏待”。匹配法：在【】内全文（连同本名标签）作子串匹配。

二阶三轴：

• 命运：死、亡、夭、终身、散、败、抄、出家、为僧、削发、嫁、卖、流落、乞、冤、泪尽、了结、收场、下场、应了、验；
• 时序：后来、此后、异日、来日、他年、再看、且看、方知、始知、方见、至此了、以后；
• 叹款：叹叹、哀哉、伤哉、可伤、悲夫、痛哉、不忍、血泪、三十年、丁亥、壬午、辛卯、己卯冬、畸笏、笏叟、脂砚。

抽样。丙层池：依回次、回内序号排列，凡【】内全文（连同本名标签，去“伏侍”“伏待”后）不含终版一阶词者，共 3,855 条。以 Python 的 random.seed(5432)、random.sample(池, 382) 抽取，按回次、序号排序。审计时所存的样本原件 bing_sample.txt（md5 9053f235c613a2313817dba78e5799e1）与本文重抽的结果逐字相同。样本批号、原件与抽样程序随文发布（附件01、附件13）；程序默认先找 附件01_bing_sample.txt，也可在命令里指定原件路径。本附录与第 3、4 节的各项计数，可用附件13 的筛查程序由底本重算。

补筛。轮二缺字批的筛词为“□”“缺”“阙”。第四轮：丁一，二阶三轴扫丙层全量；丁二，单字“伏”，排除伏侍、伏待、伏首、伏低、埋伏、伏罪、伏法、伏惟、降伏、伏在、伏身、伏枕、伏气；丁三，兆、后日、下半部、逝后、不谬；丁四，第一回 1.167 至 1.183 好了歌注句下批。第四轮的命中数，扣除前三轮已登记而又中丁一的 7 条；不扣时丁一命中 323 条，四筛合计 373 条。

附录三　复原本所据的一版

兑现复核所据的复原本合集（这是复核当时的历史快照，校验值只标识当时所用的一版，不随复原本修订更新）：

回次	稿次	md5
81–88	r54 卷封	2c1295a7a536e1b190bece936380fdfe
89–95	r54 封稿	611f11220808e11762346b83beffa4c4
96–103	r53 封稿	481064b2d768b33c6bae291c0e14384c
104–110	r53 封稿	7ef2c176cab5f5c9fbfb982d7c4d55b1

第四轮补登各条、重拆与 26.153 补核的核对，也据这一版查；唯两处据补写之后的稿子，其余与上表一版同：76.3 据第一〇八回补驿卒一句之后的稿子；6.48、甲03 的副册一层，据第一一〇回补平儿一页之后的稿子。两处都是 r53，2026 年 9 月 27 日补，改处与前后文照录于 6.4 节。

复核之时，第九十六至一〇〇回已各有更晚的封稿（r54，各回卷首记 9 月 25、26 日），字句与上表 r53 合集有出入，如第一〇〇回刘姥姥交成窑钟的话（附录四第 8 条）。审读者抽核登记所引各句，在 r54 本都在，登记不受影响；下一版，登记所引各句逐句对最新封稿。复原本以后的修订，只改字句的，本文登记不变；改了情节层取舍的，登记随改，出新版。

附录四　与总篇 v0.8 的出入

以下几处，以本文为准，总篇下一版随改（中英同步）：

1. 名称：“反向审计”改称“兑现复核”，9.3 节标题与术语表同改。9.3 节所说“把批语拆成可证伪的命题，再用复原本文本去证伪”、术语表所说“把批语拆成可证伪的命题，以复原本去证伪”，改作“把批语拆成具体命题，查复原本是否真的满足；所查的是本项目的登记，不是批语的真伪”；“无可证伪命题”改作“所指推定（无具体命题可查）”。
2. 9.3 节表：第五行“脂批自误，从正文”改作“与第五回正文不合，从正文”（本文不采八）；第六行“脂批内部冲突，以参压准”改作“依判准不采，67.1 的含意为旁证”（本文不采四，见第 11 条）。本文的不采九（两批互异，取其一）现在只有 18.17 平儿一层一例，两边都是准级，不是以参压准；总篇如需举不采九之例，用这一条。
3. 8.2 节：第三回问月钱一例，改说批语的后文第三十九回已应，复原本第八十六回是延伸，删去“隔了八十三回”。
4. 表一第十九行：去掉 1.77 这个锚（本文改记非前指）。
5. 表一第三行：锚“第四十二回批‘狱庙相逢之日’”只证狱庙相逢之有，可不改，加注“专题一记半兑现”。
6. 表二：计数照本文 6.5 节更新。丙层一条改作：补筛四轮，登记漏网三十四条（另有一条核后为只叹不指），新增兑现缺口一条（76.3），已由补记条〇〇二补写于第一〇八回；好了歌注句下批大部移专题五。抽样一条，“漏网率”改说“检出比例”。
7. “明确不承账六条”（8.2 节、9.1 节）：改说各条按型登记。原六条分属：甲06 全谱一层，不采四；甲11 迷失稿，不采六；甲20“遗簪”“更衣”，不采六；22.103 反事实与寄望，理解不同；35.26 因果说教，非前指兼理解不同；28.1“宝卿得同终始”，不采四（与总篇 9.1 节所记理由一致）。
8. 表三妙玉行与附录五妙玉行：“刘姥姥托还成窑钟”两处，按第一〇〇回 r54 封稿改为“刘姥姥白给成窑钟”（封稿原文：“这个原是你们府里出来的，这几年我一向带在身上。姑娘认得，就留着罢。”）。本篇复核所据的 r53 合集仍作“替我还了这一注人情罢”，两者属版本变化，不据旧稿反疑 r54 的白给裁定。这一条与专题一的登记无关，是总篇的版本同步。
9. 3.4 节或 4.1 节：补一句立场：批语是需保留来路的证词材料，有对所见稿本的记述，也有批者的推断与评议；本文不采的批语，只说分歧所在，不判其误。
10. 6.4 节：“得而弃之”一语，改说“定而弃之”，或加一句说明。
11. “以参压准”各处（4.1 节判准择读一段、6.4 节“其二，弃而为僧”、表一第十一行、8.2 节甲层一条、9.3 节表第六行、9.4 节“宝玉悟性的第三层”、术语表）：67.1“此虽眷恋，却破迷关，是何必削发”有两读，字面说的是湘莲；本项目取其含意，即末回撒手者不必削发。两批字面不相抵，所以本文把 21.65“弃而为僧”一层记为依判准不采（不采四），67.1 的含意只作旁证。总篇以此为“以参压准”之例的各处，改说依判准不采（判准择读），67.1 的含意为旁证；术语表“以参压准”一条删去，或注“本项目今无实例”。

附录五　随文附件

本文随附三件：正文（PDF）、审计全表（CSV）、附件包 专题一_附件包_v0.4.zip。附件包内是附件01 至 14，文件名如下表；另有 读我.md 一件，说明程序怎样运行。

件	文件	内容
数据	专题一_脂批前指审计数据_v0.4.csv	审计全表，4,099 行，22 栏（附录一）
附件01	附件01_抽样.md、附件01_bing_sample.txt	抽样的做法、样本批号、二阶筛在样本中的 27 条命中；审计时所存的样本原件
附件02	附件02_盲拆输入_40条.md	第一次盲拆的输入；有 7.5 节所说的提取缺陷，照原样留存
附件03	附件03_盲拆命题_40条.md	第一次盲拆的输出
附件04	附件04_复核时登记快照.md	兑现复核时交给核对者的登记（公开版删改一处，见附件12 第四节）
附件05	附件05_复核_甲层.md	甲层 23 项的复核记录（公开版删改两处，见附件12 第四节）
附件06	附件06_复核_乙层一类.md	乙-2 一类 13 条的复核记录
附件07	附件07_重拆输入_15条.md	修正提取后的重拆输入
附件08	附件08_重拆命题_15条.md	重拆的输出
附件09	附件09_重拆比对_4条.md	交重拆核对者的比对材料（四条）
附件10	附件10_重拆核对_4条.md	四条出入的核对
附件11	附件11_裁定记录.md	作者对复核建议与审读意见的裁定，至 v0.4 发布
附件12	附件12_运行说明.md	各次 AI 会话的用途、模型、时间、读过的文件、提示词原文；公开版删改记录；核对输出中不取的陈述
附件13	附件13_程序/	筛查重现.py（第 3、4 节与附录二的各项计数）、抽样重现.py（丙层抽样）、上下文提取.py（全表“所批正文”两栏）
附件14	附件14_补核_26.153.md	v0.2 送审后对 26.153 的补核：比对材料与核对输出（7.5 节）

附录六　致谢与 AI 使用说明

感谢 Zesi Chen、Zhen Zhang 在项目中给予的大量分析与帮助。

本项目使用了多家 AI 辅助，包括 Claude（Anthropic）、ChatGPT（OpenAI）、Gemini（Google）、DeepSeek、GLM（智谱）、Meta AI、Grok（xAI）。本文所据的审计，由 AI 按作者所立的口径执行筛选、登记与复核；兑现复核中的盲拆、核对与重拆，也由 AI 分别承担，所用模型与会话见 7.3、7.5 节与附件12。所有裁定由作者作出，文责由作者自负。

参考文献

• 秦汉（Han Qin）《人是有情的：〈红楼梦〉后三十回诠释性复原的缘起、方法与结论》，预发布版 v0.8，Zenodo，concept DOI: 10.5281/zenodo.22985375。
• 江山阁隐制作《红楼梦脂批本完美版》，网络流传电子本（PDF；据制作者介绍，以庚辰本为底本，吸取甲戌、王府、戚序等本批语）。
• 曹雪芹《脂砚斋重评石头记》诸本：甲戌本、己卯本、庚辰本、蒙古王府本、戚序本；靖藏本批语（原件已佚，据过录）。
• 曹雪芹著，中国艺术研究院红楼梦研究所校注《红楼梦》（北京：人民文学出版社）。

[^1]: 底本的来源与整理，见总篇卷首说明第三条及其脚注。本名标注从底本；与通行校本的出入，影印本逐条复核列为待办，其中审读者已指出的三处列在最前：18.19“回末／末回”；42.6 本名（底本标“该批”，或作靖本眉批）；6.49 本名（底本标“脂批”，紧接 6.48 甲戌夹批，“余言”即指 6.48，通行辑本或作甲戌夹批一条到底；若是，甲03 本名作甲戌，并与 6.48 同处登记）。