Who Wrote the Game? Strata of Authorship in the Earth-Trisolaris Organization's Three Body, with a Note on Its Courtesies

1. Introduction

On the ninth day of the third month of Bunker Era, Year 9, a records clerk of the ETO Prosecution Records unit opened the sixth box of seized headsets and began to log the session recordings stored inside it. The seventeenth recording, which the clerk's marginal note calls unusually clean, shows a player standing in a bare hall while King Wen of Zhou lays his forecast of the sun's return before King Zhou of Shang. In the course of the audience King Wen turns to the player, asks for counsel in a voice of careful politeness, and at the end of the interview thanks the player in a courteous phrase that the clerk copied out twice, as though it might be wanted later (ETO Prosecution Records, BkE 9).

Nothing in the box says who wrote that courtesy.

The question has been put to the fairy tales for a decade and almost never to the game, though the game was circulated by the Earth-Trisolaris Organization (ETO) to recruit intellectuals, and Wang Miao, who played it, reconstructed Trisolaran astrophysics from what he found there. A thing that recruits presumably has a hand behind it and a listener in front of it. I write in Bunker Era, Year 11 (BkE 11), when Trisolaris has been destroyed and no Trisolaran original can now be consulted, so every attribution below is an inference from the internal evidence of recordings and from nothing else. My claim is that Three Body is best read as a stratified text, in which a human frame was set over material of Trisolaran origin and then furnished with the courtesies a human recruit would expect. (Stratified? Layered? Palimpsest, heaven help us. Strata, then, since the title has already committed me, and no further nouns.) The simplest picture is a lodging house. The original rooms are still there, but someone has put up partitions, papered them in a pleasant pattern, and hung pictures of ancestors on the paper. A visitor sees a drawing room, and the surveyor sees a wall.

This paper does not decode the game, and it takes from the Yun Tianming tales a method alone, the lesson that a text composed under an audience's eye bears the marks of that audience — marks that, once noticed, can be counted.

2. Corpus and Approach

The corpus is 34 session recordings from the ETO Prosecution Records, declassified series, recovered from headsets seized from members and recruits of the Organization. Boxes 1 to 6 hold 52 recordings. Fourteen were set aside because the opening segment is damaged or truncated and four because they duplicate another session, leaving 34 of the cleanest sessions, and not all that were played. In these 34 recordings 441 passages were segmented, a passage being a continuous stretch of speech, narration or environmental text which treats one matter. Of these, 380 bear on one of four matters and were coded, at a mean of 11.2 per session; the other 61 concern controls, log-in and similar business and were left uncoded. The 380 were sorted into four strata by content: the recurring civilisations that rise and are destroyed, the historical figures who appear as players or characters, the chaotic and stable eras that govern those risings, and dehydration. The first stratum has 96 passages, the second 118, the third 104 and the fourth 62.

Each passage received one of three labels, of which the first, H, marks human composition, meaning idiom, address or allusion that presupposes a human upbringing. The second, T, marks translated Trisolaran material, meaning content that appears to have been rendered into human speech from a source whose conventions are not human, and which betrays the rendering by a gloss, a flatness of valuation or a plain statement of what a human author would have framed. The third, U, marks passages that could not be assigned. I coded all 380 passages, and a second reader of the Visiting Readers' Panel coded them independently without sight of my labels. Agreement was 317 of 380 passages. The 63 disagreements were settled by discussion, and the consensus labels are those reported below. The practice of coding a made text for the marks of its audience follows the Committee's handling of the collated tales (IDC, BkE 4).

The 380 passages are nested in 34 sessions and are not independent. The intraclass correlation of the H label, estimated from the session counts by one-way analysis of variance, is 0.065, and the design effect (one plus the mean passages per session less one, times that correlation) is 1.66 for all 380 passages and smaller for a subgroup, since a subgroup has fewer passages per session. Every interval below is a Wilson interval at the effective sample size, the count divided by the design effect, and it assumes that subgroups share the overall correlation. Kappa is 0.74 (95% CI 0.67 to 0.82), widened by the square root of the overall design effect.

A caution about order governs the inquiry from end to end. A recording preserves the order in which a player met the material and not the order in which the material was composed, so the argument from the order of telling that I have made for another corpus (see Carrow-Mabey on the painter's tale) carries over here as a method of asking and not as a method of answering. And since the Organization's own channel to Trisolaris ran, after their arrival, through sophons, which attend every human record of the period, it cannot be shown that any recording was untouched. The recordings are taken as the Prosecution unit received them. The astrophysical content has an independent check in the working catalogue of Wang Miao's reconstruction (Moravcik, BkE 5). It is used only to confirm that the game was probably accurate where it was technical, not to assign anything to a hand.

3. The Historical Figures and the Courtesy of the Invited

The distribution of labels across the strata is what the data decide, because the strata were sorted by content and the labels by form, so that nothing in the coding rules obliges a stratum to take a particular label (Table 1). Of the 118 passages on historical figures, 97 were assigned H, 4 were assigned T and 17 were left undecided. The H share is 82.2% (95% CI 73.6 to 88.4), against 32.3% in the civilisations, 36.5% in the eras and 8.1% in dehydration, so that the figures are the one stratum that is predominantly human in composition. The figures who appear, among them King Wen of Zhou, Copernicus, Newton and von Neumann, seem to speak in the manner of an educated man's anecdote about them. I could trace their remarks to no source beyond that manner, and perhaps none exists.

What marks these passages as human is a habit of address, and the habit comes with a warning. Address to the player is one of the marks that assigned H, so the 71 of the 97 H passages (73.2%, 95% CI 63.0 to 81.4) that turn to the player, ask for an opinion or thank the player for one largely restate the coding rule and are not a second line of evidence. What the rule could not force is that the human label should gather in this stratum and not another. A text with no audience has no reason to be polite. Nobody else had reason to be. A Trisolaran source, whose speakers display what they think, would have little use for courtesy as a means of persuasion. The politeness is therefore probably a human addition, and of a particular sort — the politeness of a recruiter who wishes the recruit to feel invited.

Courtesy is not universal among the 97. The 26 H passages without address seem to include figures who command or condemn in place of consulting, among them the emperor Qin Shi Huang, and a recruiter need not be gentle everywhere to be gentle where it counts.

4. Dehydration and the Plain Voice

The fourth stratum is the opposite case, and the distribution shows it. Of its 62 passages, 41 were assigned T, 5 were assigned H and 16 were left undecided, so that the T share is 66.1% (95% CI 53.4 to 76.9), against 22.9% in the civilisations, 27.9% in the eras and 3.4% in the historical figures. Dehydration supplies 41 of the 96 passages assigned T anywhere in the corpus (42.7%, 95% CI 32.8 to 53.3). The register probably decides the matter. The passages state that a body gives up its water, that it is stored, and that it is restored, and they do so without a trace of the relief, dread or wonder which a human author would almost certainly have added to such a matter. Of the 41 T passages, 33 (80.5%, 95% CI 65.9 to 89.8) carry no valuation of any kind. Flatness of valuation is one of the marks that assigned T, so that share is partly a property of the coding and not a separate finding; what the coding could not force is where the T passages fell, and they fell in dehydration.

I take the flatness to be, probably, a translator's fidelity and not a human author's choice. Trisolaran speech is displayed thought, and a rendering of it into human words has nothing to render but content (Eskelinen, BkE 6). The human reader supplies the feeling, and the game does not. The few places where a feeling does appear are the 5 passages assigned H, and they may be the hand of the frame reaching in to soften a plain fact for the player.

Work on dehydration in the period was done almost wholly from terrestrial comparators (Oyelaran-Voss, BkE 6), and a plain statement of the process in a source that predates them is a datum worth keeping.

5. The Eras and Civilisations: A Contested Middle

These two strata do not divide so cleanly, and the failure to divide is the finding. For the civilisations the consensus labels are 31 H, 22 T and 43 U in 96 passages, and for the eras 38 H, 29 T and 37 U in 104. Taken together, the 200 passages split into 69 H, 51 T and 80 U (34.5%, 25.5% and 40.0%), and the undecided are the largest class. Here the translator and the framer seem to have worked on the same sentence, perhaps at the same desk.

A seam can be seen in the form of the passages, perhaps more easily than in their content. The cycle itself — in which a civilisation grows in a stable era and is destroyed in a chaotic one — reads to me as Trisolaran content, since it concerns a fact of Trisolaran life. The device by which a destroyed civilisation returns and the player begins again perhaps reads as a human addition. A human player will not go on without it. The names given to the eras, with their suggestion of weather and temper, may carry the cadence of a human gloss over a plain classification (Lindqvist-Amari, BkE 8). I cannot say with any confidence where the gloss ends. (Cadence! You are a philologist, not a conductor. Say instead that the sentences seem to get longer.) I have not counted them, and the third inquiry in the Conclusion is the count; for now the impression is that the passages lengthen and acquire subordinate clauses at the points where I suspect a human hand.

Any design question has an answer even where the attribution has none. A game made to recruit must be playable, and a playable game cannot be a bare transcription of an alien convention. Whoever built the frame had to decide how much of the original a human mind would bear, and the decision is presumably visible in how often a civilisation is allowed to try again, though the restarts have not been counted across the 34 sessions and so the frame cannot yet be called generous or grudging. That choice was probably human, whatever the facts it protected. Lectio difficilior, then, or its philologist's cousin: the harder attribution is the safer one to hold open.

6. Counter-readings

The Luo-Ji reading deserves a full nod, and it has one. If survival is the primary need of a civilisation, a cycle of growth and destruction is only to be expected in any account of civilisations, and a game that shows such a cycle owes its authors, human or otherwise, nothing further (Tarkhanov, BkE 10). The premise stands. The conclusion does not, because a model has an author, and the author's habits decide which cycles are shown and which are left out. (Full? He has praised the premise twice in three sentences; ask what the praise is buying.) The recording of King Wen is a model as well, and it was written by someone who thought a recruit would enjoy being asked.

Partisans of the Lurker (also 'Spirit', 'Spirit Mother' in some translations) have a more interesting objection, and I should like to state it fairly. They hold that periodic collapse is a trace of guidance and not only of competition, and that the repeated destruction in the game might therefore record an order older than either the Organization or Trisolaris (Verhoeven-Salt, BkE 9). That would make my T class too small, since it would credit a third author. I cannot exclude it, and the 80 undecided passages in the civilisations and the eras are exactly where such an author would leave a mark, if there were one.

Neither objection requires the other to be false. This paper argues for compatibility between the Luo-Ji framework and the Lurker hypothesis, in the narrow sense that a text built in strata can hold a dark-forest model in one stratum and a larger history in another without contradiction, as the ecumenical position has always said of the universe at large. A third objection comes from the archive itself: the recordings are of play, and play may differ from the build. That is true, and it is the reason I report the labels as readings of what the Prosecution unit holds, and nothing grander. A recording of a recruit's hour with the game is a sample of the game, and no more.

7. Conclusion

The clerk of the sixth box copied out a courtesy twice. On the evidence of 380 coded passages the courtesy was probably written by a person, to be heard by a person, and the forecast of the sun that King Wen laid before the court of Shang may have been something quite different. That is a smaller conclusion than the one the paper began with, and I think it a safer one. It does not say who the person was. It may never be able to.

Four further inquiries might narrow the question, though none is certain to. The first is the intake register of the Prosecution unit, which would show whether the 34 recordings resemble the seized recordings that were not coded. The second is any surviving build notes, held among the exhibits of the prosecutions, that distinguish the frame from the base. The third is a test for the 80 undecided passages, run by a reader who knows nothing of the strata, to see whether sentence length alone separates the two hands. The fourth is for the physicists, who are better placed than I am to say whether the accuracy that Wang Miao found could survive a human retelling.

If the clerk's courtesy is read again with these in hand, it may turn out to belong to the frame, or to the base, or to neither. I should be glad to be told which, and I would not be astonished to be told that the question had been put the wrong way round.