About this text: Nature and Principle in Islam

1. Witnesses

2. Base and why

C, because it is a page image of a Qing print in traditional characters, public domain and linkable page by page. W supplies modern punctuation and served as a reading aid; its simplified characters and OCR errors are not followed.

3. Collation

Root Classic

• Method: AI (Claude) read each page of C and wrote the traditional text; W was read alongside, line by line.
• Where W and C differ, C is followed. Differences found (W → C): 唯 → 惟 throughout; 爱分 / 爱生 → 爰分 / 爰生; 万汇 → 萬彙; 是为本性，吸化资之 → 是為木性 (vegetal nature, which the sequence 金性 · 木性 · 生性 requires); 是为党性 → 是為覺性; 卷其过相 → 卷其跡相; 周遍 → 周徧; 凝贰 → 疑貳; 逆悻 → 逆悖; 功用 lists and citations checked one by one.
• Punctuation is editorial: W's, adjusted to the four-character lines of the Root Classic.
• Collated with M (2009), line by line (scripts/chinese/build_lataif.py lists every difference). M agrees with C in every line and every source tag (61 tagged lines), except:
  • M begins chapter 1 with 清哉本然, which C does not have (C, p. 120, opens 最初無稱). Shown in the Arabic column as M's line, in brackets; it pairs with Ma Lianyuan's first line, سبحان من كان كنزاً مخفياً.
  • ch. 4: C 渾融沕合 (our earlier transcription had 吻合, now corrected from the scan: M agrees with C); C 倀倀無之, M 無知 (C followed).
  • ch. 5: M 故日皆朽 (a misreading in M's text layer; C 曰).
  • orthographic only: M 黄, 脉, 内, 髪, 眾, 疎, 恒 for C 黃, 脈, 內, 髮, 衆, 疏, 恆.

Books 1–5 (卷一–卷五)

• Base C2–C6, page by page. The overviews of 卷二–卷五 are printed at the end of the preceding volume. The overview of 卷一 (槩言, here 概言) is not in C2 (it goes from the 目錄 straight to the first diagram); it is taken from P, p. 68.
• AI (Claude) read every page of C2–C6 and P (pp. 68–231) and read W alongside; 183 collation entries and 193 uncertain readings (almost all variant forms kept as printed, or small labels inside the diagrams) are in scripts/chinese/src/apparatus-tianfang-xingli.json, marked in the reader and listed under each passage.
• P agrees with C2 almost character for character. Real differences (C2 followed): P 此體之義 for 此圖之義 (b1-2); P 知之為主宰 for 知其為主宰 (b1-5); P 一功 (misprint) for 一切; P 羣, 溟渣, 誤 for C2 群, 溟涬, 悞; P 流浸其下 for 其中, 鳥獸不生 for 不育, 是為 for 是謂 (twice) (b1-11 to b1-14).
• W's errors are many (OCR): 元稱 for 無稱, 緒龕 for 翕翕, 大川 for 大用, 做黨 for 微覺, 已離 for 已寓, 火 → 人, 麗 → 雨, 萃 → 革, 覺 → 黨; and two that change the sense: 庫而西包含六天 (C2 and P: 七天), and the end of 性理始分圖說, where W drops the second 不可 (C2 and P: 謂之為有為不可，謂之為無為不可). W also skips 而下氣輕 in Hei's note on the four images.
• 卷二–卷五, P against C: few real differences, mostly P's misprints (去里 for 去地, 靈清 for 靈活, 耳自 for 耳目, 意覺 for 五覺, 世界之界 for 之內, 成有 for 或有); P 心居其內 / C 心居至內, P 其有滋養者 / C 其為, P 越發 / C 發越, P 何為 / C 何謂; C 寒媛 is a cutting error, P's 煖 adopted. W omits the heading and diagram of 四際分空 (卷二) and has many OCR errors (e.g. 以人勝 for 以火勝).
• The notes of 清源黑氏 (Hei of Qingyuan), set in the print after each explanation, are kept on lines of their own and shown apart as a later commentator's, never as Liu Zhi's.

4. Decisions (substantive)

5. Quotations and citations

Liu Zhi's interlinear source notes are kept as 〔…〕 and shown as green chips. His own list (C, pp. 113–114) pairs each Chinese title with the Persian or Arabic title in Chinese letters: 勒瓦一合 = 昭微經 (Jāmī, Lawāʾiḥ), 額史爾 = 費隱經 (Jāmī, Ashiʿʿat al-lamaʿāt), 默格索德 = 研真經 (Nasafī, Maqṣad-i aqṣā), 密邇索德 = 道行推原經 (Rāzī, Mirṣād al-ʿibād), 古爾阿尼 = 寶命真經 (the Qurʾān), 默瓦吉福 = 格致全經 (reading uncertain). The identification of the four Persian books follows Murata (JMIAS 36, 2004); the others are marked AI.

6. Characters and readings

Readings (pinyin) are modern Mandarin from Unihan, not Liu Zhi's pronunciation. Radicals and strokes: Unihan kRSUnicode, kTotalStrokes. Components: cjk-decomp (Apache-2.0), automatic. Kangxi Dictionary entries: the Wikisource transcription of the 1716 dictionary (531 of 641 characters found; the rest link to their page).

7. Segmentation and IDs

One passage per chapter (#c-1 … #c-5); characters are addressed <chapter>.<n> (?w=5.12), punctuation excluded. Technical terms are matched longest-first from src/content/zh/terms.json.

Payload

The reader page carries the metadata, every passage heading and the whole Root Classic (with the Arabic column); the five books come from public/data/zh/tfxl/b1.json … b5.json (scripts/chinese/split_zh.py, run by build_zh.py), fetched after the first paint, the book a ?w= or #c- link points into first. Page HTML (dev server): 8.79 MB → 1.11 MB (gzip 734 KB → 132 KB).

8. Diagrams

The ten diagrams of the Root Classic (C, pp. 132–141) are shown as page images (720 px WebP) with their Chinese titles; the English titles are AI. The sixty diagrams of 卷一–卷五 are shown beside their explanations, as page images linked to the scan; the characters written inside each diagram are in the transcription notes.

8a. Arabic column: Ma Lianyuan, al-Laṭāʾif

Nūr al-Ḥaqq (Ma Lianyuan 馬聯元, d. 1903) put the Root Classic into Arabic line by line. It is shown beside the Chinese of chapters 1–5 (toggle "Arabic"), cited as "Nūr al-Ḥaqq (Ma Lianyuan), al-Laṭāʾif, as printed in Murata et al. 2009, p. N". Checked against the page images by AI; not yet proofread by a person.

Alignment method (scripts/chinese/build_lataif.py, output app/src/data/zh/tfxl-lataif.json):

1. M prints the Chinese and the Arabic in the same line order (Chinese on the even pages, Arabic on the odd). The Chinese lines (with Liu Zhi's tags in parentheses) and the Arabic lines are read in page order from the book's text layer and cut into chapters at 第X章 and الفصل. Line counts agree in every chapter (62, 91, 82, 91, and 83 Chinese against 74 Arabic in ch. 5, where the Chinese adds the ten diagram titles and the Arabic the closing ḥamdala); so line i answers line i.
2. The Arabic text layer is repaired by rule: presentation forms to plain letters (NFKC); the font's kāf, which the layer encodes as آ (U+0622), back to ك (a real madda is encoded differently); ا + ﷲ → الله, and الحمد الله → الحمد لله; a final tanwīn that the layer moves to the start of the line is put back. Proofreading (2026-09-28): every Arabic page, pp. 103–153 (26 pages, 395 lines and the five headings), was rendered with pdftoppm and read by AI (Claude, scripts/chinese/proof_lataif.py) in two independent passes: one compared our lines with the image, one transcribed the image blind. A change was made only where both passes agree, and a sample of each kind was checked by eye at 800 dpi. The text layer turns out to drop every kasra (U+0650) and to set some final fatḥatān before the alif rather than on it. 13 corrections (scripts/chinese/lataif_corrections.json, also stored in the data as proof): 3 kasras restored (فبالامرِ 1.10, خِواصّ 2.85, عينِ 5.1), 6 tanwīns set on the alif (كنزاً, ناراً, بحاراً, كبيراً, جسداً, وجوداً), 4 dotted yāʾs as printed (الكرسي 2.43; هي 3.28, 3.39, 3.43: the print has both هى and هي). Four single-pass proposals were rejected after a look at 800 dpi (5.6, 5.8, 5.10, 1.20). Kept as printed: ملا بسة (1.2), درادى (1.15), الآثارالبشرية (4.40). No line was added, lost or re-split, so the line-by-line English is unchanged. Status: checked against the page images by AI; not yet proofread by a person.
3. Our Chinese lines (split at punctuation, Liu Zhi's citations set aside) are aligned to M's with difflib; one line, M's 清哉本然, has no counterpart in C and is shown as M's.
4. Each Arabic word is looked up in the library's word index (app/public/data/usage/*.json) by its bare form, then without the proclitics و ف ب ل ك and the article, with a stoplist of particles and pronouns and a check that the root's strong letters occur in the word in order. 743 of 1,661 words get a root; the guess is shown as automatic analysis (grey) and opens a root card that links to the root's page. It is not a parse.
5. Liu Zhi's source tags are kept per line (from C, and compared with M's: identical). The ten lines tagged 昭微經 (his Chinese Lawāʾiḥ), plus 一而三義 (tagged 同上), are linked to Whinfield's flashes (ZHAOWEI_LINKS in app/src/lib/zh.ts), each with its grounding:
   • Concordance: Chittick's Gleam N (in Murata, Chinese Gleams, 2000, beside Liu Zhi's own Chinese version) is Whinfield's Flash N for Gleams 1–26 (checked passage by passage: e.g. Gleam 18 = the abstraction of genera, Flash XVIII; Gleam 25 = the Reality of realities, Flash XXV); Whinfield's long Flash XXVI holds Gleams 26–30, and Flashes XXVII–XXX the rest. Liu Zhi's chapter numbers run one ahead of the Gleams from Gleam 21, which he splits.
   • Named by the scholars (M, notes to the Head Volume, pp. 154 n. 2 and 157 nn. 28–32; cited, not reproduced): 作為始出 → Gleam 13; 數者其用 → 25; 一而三義 → 17; 故曰皆朽 → 18; 見物皆幻 → 2; 事何非實 → 25. Three earlier AI proposals were wrong and are withdrawn (ch. 1 Flash XXIV → XIII; 故曰皆朽 XXVI → XVIII; 事何非實 XXIX → XXV).
   • AI proposals (M gives no source): 一貫而已 → XXV (medium); 孰云偏駁 → XXVII (medium); 種果全焉 → XXX (the two self-disclosures, in knowledge and then outwardly; medium); 真一還真 → XXV, its closing quatrain (low).
   • No confident match: 即非本然 (ch. 5). No flash says that what departs from the Essence's own requirement is not the Essence; its first tag is the Yanzhen jing (Nasafī), which may be the real source. Method: AI (Claude) read Whinfield, Chittick's Gleams and Murata's English of Liu's Chinese chapters side by side (the private copy of Chinese Gleams used only as an alignment aid), then each result was checked against the passages.

9. Translation

English by AI (Claude), marked. Murata's translation (2009) is in copyright and is not shown.

10. Term mappings

src/content/zh/terms.json (built from scripts/chinese/src/terms.json): 110 terms across the two Chinese texts. Each Arabic or Persian counterpart states its grounding: the text itself, a published scholar (Murata 2004; Murata 2000; Whinfield 1906; Murata, Chittick and Tu 2009, by page; Ma Lianyuan's *al-Laṭāʾif*, the Arabic line that renders the Root Classic line containing the term, by page of M), or AI with a confidence. Arabic roots link to /roots/<root> where the root is in the library.

Added 2026-09-28 for 卷二–卷五 (34 entries; 32 text groundings, 31 page citations of M, 23 of the Laṭāʾif, 12 AI proposals): the natures of Book 3, 繼性 (the Nature of Continuity; Nasafī's Ascribed Spirit, al-rūḥ al-iḍāfī, M p. 512), 本性, 氣性, 活性, 發育, 堅定, 靈活, 知覺 (Ma: al-rūḥ al-insāniyya, al-nafsāniyya, al-ḥayawāniyya, al-maʿdiniyya, al-malakūtiyya, al-idrāk wa'l-iḥsās); the cosmos of Book 2, 七洲 (the seven climes, al-aqālīm al-sabʿa: Ma, p. 119; M p. 322), 四際 (the four boundaries of the air; Ma gives no technical term; ṭabaqāt al-hawāʾ is AI), 七政 (the seven planets, AI high), 四時, 形器, 性品; 知能 (al-ʿilm wa'l-qudra, Ma), 功用 (khāṣṣa, Ma; M p. 285 n. 6), 全體, 大用, 自然 (iqtiḍāʾ, Ma p. 147), 終古, 一息, 本量 (qadar, M pp. 570–71 n. 3), 障礙, 順逆, 疑信, 歸真, 名相 (asmāʾ wa-ṣuwar, Ma p. 149; M p. 538), 三一, 渾化, 代理 (khalīfa, M p. 512), 體竅, 胚胎. Method: AI (Claude) drafted each entry from our text, the Laṭāʾif lines and M's pages (read privately, paraphrased); every Laṭāʾif citation was then checked by script against the Arabic line and its page, every M page against the page text, and four entries were corrected by hand (繼性, 代理, 本量, 胚胎).

Names: 爾里父 (Book 4 · 12, Diagram 4.9) is perhaps ʿārif, "the gnostic", an unnamed Sufi authority (M p. 474 n. 1, tentative); 偶栽爾 (Book 5 · 10, Diagram 5.8) transliterates ʿUzayr (Ezra), but Liu Zhi probably meant ʿAzīz al-Dīn Nasafī, whose ʿAzīz differs by one dot; M traces the saying to Nasafī's Maqṣad-i aqṣā on God's storehouses and relates it to the Hidden Treasure (M p. 548 nn. 2–3). Both are in the term list with that grounding.

11. Rights

Scan: public domain. Wikisource: CC BY-SA (used as a reading aid; its punctuation followed). Nothing in copyright is shown.

12. Open questions

• The Arabic column has been checked page by page by AI (see §8a); a person's proofreading is still wanted, especially of vowel signs.
• 即非本然 (ch. 5, tagged 研真經又昭微經) has no confident match in the Lawāʾiḥ; look in Nasafī's Maqṣad-i aqṣā.
• Identify 清源黑氏 securely: the postface in vol. 6 is signed 朝陽黑鳴鳳 (Hei Mingfeng of Chaoyang), which supports identifying the annotator as Hei Mingfeng; 清源 and 朝陽 are both place names to be explained.
• 爾里父 = ʿārif is Murata's tentative guess (2009, p. 474 n. 1); the saying itself is still to be found in a Persian or Arabic source.
• The diagrams' small ring labels (卷二 pp. 8, 18, 30, 35; 卷三–四 several) are only partly legible in C; read with P where clearer; recorded in the transcription notes, not in the reader.
• Term mappings for 卷二–卷五: added (§10); 四際 and 七政 have no scholarly Arabic yet (AI only).
• 青尼大道 in Hei's note on 真理流行圖 (b1-6): Laozi, or an error for 仲尼?
• The flash-by-flash links to the Lawāʾiḥ (six named by Murata et al., four AI) are to be checked against the Persian.

13. Change log

• 2026-09-28: the Arabic column checked page by page (13 corrections); the Zhaowei jing lines linked with grounding (6 from Murata et al. 2009, 4 AI, 1 open); 34 term mappings for 卷二–卷五 and the two names; the reader loads Books 1–5 on demand.
• 2026-09-28: Root Classic 1–5, diagrams, terms, AI English.
• 2026-09-28: 卷二–卷五 (62 passages, 30,283 characters, 48 diagrams), same pipeline as 卷一.
• 2026-09-28: 卷一 (overview and the twelve diagrams explained, with Hei's notes), collated with P and W; the Root Classic collated with M (one correction, 沕合); Ma Lianyuan's Arabic column; 11 new term mappings.