about this edition

What Pivot is

Pivot is a machine edition of a body of classical Chinese prose: the source texts with their Korean translation, and the record of every interpretive step taken between a passage and a conclusion about an author.

Historians move between scales of observation, and the analytical steps connecting them usually remain implicit in the finished argument. This edition keeps them. A section record, a claim under a dimension, an essay-level account and a provisional profile are all durable objects here, each with a row, a run and a place on a page — not hidden intermediate data on the way to a result.

Knowledge distillation, in this project, is not the compression of one model into another. It is the machine-assisted construction of progressively more synthetic interpretations, each of which can be traced back to the passages, section records and essay syntheses beneath it, so that a reader may revise the segmentation, accept or reject a classification, modify the taxonomy, or reconsider the larger profile. A more synthetic interpretation is not a more authoritative one.

Nothing on these pages is written by hand. Every page is generated from one SQLite database whose schema is declared in tools/schema.sql; a correction is a change to the database and a rebuild. The same database is served, unchanged, by the query interface behind Search.

The edition presently holds 733 works by six authors, of which the pages of Su Shi 蘇軾 are built in full: 19,857 source lines, 3,103 section records, 8,531 claims, 733 essay-level syntheses, 6 scholar-level profiles, 2 cross-author comparisons, and 40,367 resolved references linking synthetic prose back to the text it is about.

§1

The eight epistemic states

A reader must be able to tell what kind of statement is in front of them without relying on colour. Position, typography, border grammar, reading mark and label each carry the distinction, so the states remain legible in monochrome, in print, and to a screen reader.

Source statement · passageThe classical Chinese line as the anthology has it, with the Korean translation beside it. Nothing is inferred; the only editorial act is the alignment.

Structured observation · sectionA segmentation of one essay into sections, with the sentences each section rests on, its dominant mode and its structural pivots. A boundary is an interpretation and can be disputed.

Model-assisted inference · claimOne claim about one section, filed under one of the ten dimensions, and returnable to the lines it quotes. Set beside the text, never above it.

Human revisionA correction entered by a reader. The superseded row is kept and gains a status; a rejected claim is not deleted.

Essay-level synthesis · essayThe account of a whole work: core thesis, macro-structural sequence, dimension syntheses, dialectical pairs, key concepts. More synthetic than a claim, not more authoritative.

Provisional profile · profileThe scholar-level account of recurrent tendencies in a selected corpus. Conditioned on the selection, and recomputed when it changes.

Historiographical positionA reading proposed by secondary scholarship, which may contest the taxonomy itself. It is a derivation too, from another reader.

Counterfactual extrapolation · counterfactualWhat the text implies would have followed otherwise, and what an agent built from these writings might argue outside its historical experience. Never set in the quotation face, never presented as documented speech.

§2

Provenance

Every distilled row points at a run: the level it was produced at, the model as recorded in the output file, the time, and the lineage the file declares. Where a file recorded no model, the pages say so rather than supplying one. 1,474 runs and 2,199 improvement-log entries are held.

Level	Model as recorded	Runs	Range
comparison	Gemini 3.6 Flash (High Thinking Mode)	1	2026-07-29 – 2026-07-29
compass	not recorded in the file	1	2026-08-14 – 2026-08-14
essay	not recorded in the file	15	2026-07-27 – 2026-07-28
essay	gemini-3.6-flash	718	2026-07-27 – 2026-07-28
scholar	not recorded in the file	6	2026-07-28 – 2026-07-28
section	not recorded in the file	733	2026-07-27 – 2026-07-28

The Korean translation and the classical Chinese text come from the source anthology and are not model output. The two historiographical positions and the human-revision state described above are part of the design of the edition and are labelled wherever they appear.

§3

Citing a record

Every row addressable on these pages has a stable identifier. A citation names the record class, the identifier and the date of the run that produced it. The identifiers do not change when the pages are rebuilt; they change only if the database row changes, and a changed row is a new run.

A work is cited by its slug, which is derived from the anthology order and the title and is stable: su-shi/001-…. A source line adds its sequence number, a section its section number, a claim its record id.

record: PH.CITE
work: PH.WRK.su-shi/003-guan-zhong-lun-1
source line: PH.SRC.su-shi/003-guan-zhong-lun-1#l12
section: PH.SEC.000217 · s1
claim: PH.CLM.001884
essay synthesis: PH.ESS.000141
profile: PH.PRF.000001 · run 2026-07-28
run: PH.RUN.000412

In prose: Playing Heaven, Pivot: the machine edition, record PH.CLM.001884 (essay-level run of 2026-07-28), pivot.playingheaven.ai/su-shi/003-guan-zhong-lun-1/#c1884.