| Work / author / dates | Kitāb al-ṭibb al-rūḥānī, a manual of ethics in a medical frame, companion of the Manṣūrī (the preface says so) · Abū Bakr Muḥammad b. Zakariyyā al-Rāzī (Rhazes), c. 250–313 AH / c. 865–925 CE. The preface does not name the amir or give a date; neither is asserted. English title in the reader: «Spiritual Medicine». |
| Part added | Preface, the plan of the book and chapters 1–2, Kraus 1939 pp. 15–32: 18 passages, 2,959 words (2,979 tokens with dashes). Pages: preface p. 15; plan pp. 16–17 (line 13); chapter 1 pp. 17–19; chapter 2 pp. 20–32 (ends at p. 32 line 9; chapter 3 begins on the same page). The scoping note's "chapter 2 = pp. 19–31" is corrected: the page tags of OpenITI come at the end of a page, so its figures were one page early. |
| Reader route | /read/razi-tibb · apparatus /read/razi-tibb/about · published 13 Oct 2026 (FLAGS M). |
| Status | working edition, published 13 Oct 2026 · Arabic: Kraus 1939 read from page scans by AI (two sonnet subagents + the editor-AI for p. 32), compared with OpenITI, reviewed against the page by AI (opus) review, 13 Oct 2026, not by a person · fully vowelled by AI (Kraus's own printed marks kept, shown apart) · analysis CAMeL + an AI sweep of the flagged words only · English (readable only; no literal line) and word glosses by AI · reviewed by: AI (opus) review, 13 Oct 2026 (not a person; razi-tibb-review.md, FLAGS L) |
0. What the text says
The preface: the amir asked, after hearing of a treatise on reforming character written for friends in Baghdad, for a short book called Spiritual Medicine, the counterpart of the bodily-medicine Manṣūrī (¶1); the book has twenty chapters, listed (¶2). Chapter 1: reason is God's greatest gift and the standard; passion must not rule it but be schooled by it (¶3–4). Chapter 2: restraining passion is the first principle (¶5–6); passion and nature seek the present pleasure; weigh, and restrain even without a foreseen harm (¶7); addicts lose the pleasure and cannot stop (¶8); those who deny the soul's independence and those who affirm it (¶9); the latter argue from man's build and the beast's contentment (¶10–13: the objections, the thousand dinars, the king who found Paradise a burden); the ascetic company, and why the soul after death lies outside the book (¶14); Plato's three souls and their organs (¶15), the defect and excess of each soul and the medicine of body and soul (¶16), the span of life and the fate of the soul that leaves the body, closing with "a summary of the view of Plato and, before him, Socrates the one who withdrew" (¶17); and the close: every view restrains passion, the lowest rung is to rein it in so far as no present harm follows (¶18). Passages and Kraus pages are in the table in the reader's section list; the Plato summary is ¶15–17.
1. Witnesses
| Siglum | Witness | Licence | Location |
|---|---|---|---|
| K | Paul Kraus (ed.), Rasāʾil falsafiyya Part 1, Cairo 1939, pp. 15–32 | likely public domain, not legally cleared (as the Sīra's §11) | read from the NYU Arabic Collections Online scan (auc_aco000121, https://sites.dlib.nyu.edu/viewer/node/24761, hdl 2333.1/8gthtdkh; high-resolution PDF mc.dlib.nyu.edu/files/books/auc_aco000121/auc_aco000121_hi.pdf, 336 canvases, canvas = printed page + 12, verified here on pp. 15, 17, 18 and 32) |
| ف ق ك ل | the manuscripts of Kraus's apparatus: ل BM Arabic 1530 (Baghdad 759 AH, his base), ف Vatican Arabic 182, ق Cairo DKM taṣawwuf 2241 (732 AH), ك al-Kirmānī's excerpts (probable); from his introduction pp. 5–7, read by the AI review | public domain as works | known through his footnotes |
| O | OpenITI 0313IbnZakariyyaRazi.TibbRuhani.Shamela0012834BK1-ara1 (Beirut 1982 reprint of Kraus, typed) | CC BY-NC-SA 4.0 | committed whole at data/sources/openiti/ (SHA-256 8ea0c71d…); witness only |
| — | Arberry, The Spiritual Physick of Rhazes (1950); Adamson, Al-Rāzī (2021) | in copyright | cited only; not opened, not consulted |
Page images (13 Oct 2026): each of the 18 passages is marked, line range by line range, on Kraus's pp. 15–32 (NYU IIIF, canvas = page + 12; nothing rehosted); boxes from Vision line detection aligned to data/sources/razi-tibb/kraus-reading/kraus-lines.json, checked by eye on several pages. One crop: the probable «يُرِى» (p. 17 l. 11, printed «يُزِى»). Same Kraus volume, same OpenITI book (12834: BK1 here, BK2 the Sīra) and same rights reasoning as the Sīra (razi-sira.md §11): see that record; not repeated.
2. Base and why
K, read from its page images (editor's instruction: Kraus is the base; OpenITI a witness). Two sonnet subagents read pp. 15–23 and 24–31 from 170 dpi renderings (their notes record the places read without zoom); the editor-AI read p. 32. scripts/razi_tibb/clean_reading.py cleans the readings; build_base.py builds the base with the corrections in scripts/razi_tibb/fixes.json. Kraus's printed vowel marks are kept in the base; his chapter headings are the section labels (Arabic in headings).
3. Collation with OpenITI (collate.py → data/sources/razi-tibb-collation.json)
- Base 2,959 letter-words; OpenITI 3,645 on the same range (lines 40–361). 49 differing spans at build, 45 after the AI review's corrections (now 26 words differ); at build: 30 words differ, 13 words Kraus has that OpenITI lacks, 4 OpenITI has that Kraus lacks, and 2 duplicated blocks.
- OpenITI gaps: no page is missing on pp. 15–32 (the Sīra's failure mode, a dropped page, does not occur here). Instead OpenITI repeats text: a block of 422 words (OpenITI lines 212–245: from «أتنغص هذا النعيم …» back through the king's speech to «… فإني أتنغص», at Kraus p. 26) and a block of 265 words at the end of chapter 2 (lines 341–361: the text from p. 30 «يوية والنطق …» to the end of the chapter, printed once by Kraus). The scoping word count (3,152 words for ch. 2) was inflated by these 687 words; chapter 2 has 2,452 words. The page-tag locations are untrustworthy for these blocks.
- Small differences are mostly OpenITI's slips or normalised spellings against Kraus's print («سألنيها», «إخوانى», «الرديّة», «أفلاطون»/«فلاطن», «الحيوة»). The places where Kraus's text differs from what sense expects are queued (review A).
- Kraus's supplements (angle brackets): none were found on pp. 15–32; none is marked editorial. (The convention added for the Sīra,
editorialon the words, is ready inscripts/razi_tibb/mark_words.py.)
4. Decisions (substantive; scripts/razi_tibb/fixes.json)
| Where | Reading adopted | Rejected | Reason |
|---|---|---|---|
| p. 17 l. 11 (plan, ch. 18) | «يُرِى» ×2 (emendation) | «يُزِى» (as printed, certain at native resolution) | the clause is in ms. ل only; OpenITI يرى; form IV 'shows' is the only construable reading (review A1; shown as probable) |
| p. 15 l. 4 | «سألنيها» (emendation) | «سألنها» (as printed, = OpenITI) | the printed form is ungrammatical; «سألنيها» is ms. ف's reading per Kraus's note 4 (review A8) |
| pp. 21, 23, 28, 30 | «يتهيأ», «المخدوعة», «كجوهر», «الجسم» | «تهيأ», «المخدوع», «لجوهر», «الجسد» (first reading) | misreadings corrected by the AI review on the page (= the print and OpenITI) |
| p. 26 l. 13 | «أَتَنَغَّصُ» | «أنتغّص» | the print reads so at native resolution (review A2; certain) |
| four places | word division restored: «وتكون» (p. 26), «هلاكها» (p. 23), «إذا كان» (p. 32), «بما وهب» (p. 19) | the print's gap or run-together | mechanical |
| p. 18 l. 1 | «وأجداها» kept | OpenITI «وأجدها» | clearly printed on a 300 dpi crop (queue A3) |
5. Quotations
No āya is quoted on these pages. One unattributed verse in ¶13 («وهل ينعمن إلا سعيد مخلد … ما يبيت بأوجال»); identified by the AI review as Imruʾ al-Qays, lāmiyya «ألا عِمْ صباحًا», ṭawīl, from memory: UNVERIFIED against a dīwān (which reads «يَعِمَنْ»). The text is not registered in scripts/quran_refs.py.
6. Vowelling (the difference from the Sīra: FULL)
- Every word is vowelled: short vowels, shadda, tanwīn, sukūn, case endings (by sonnet 5.5, in the editor session). The AI marks are labelled AI exactly as the site labels AI vowels: per-character
vowelMask,'t'= a mark the print carries (Kraus's, as read from the page images),'a'= added by AI. The reader's rule for a text vowelled mostly by AI (mostlyAi) marks with the dashed amber underline the doubtful words (those where CAMeL's best analysis disagrees: 115) and the one reading shown probable («يُرِى» ×2, an emendation; «أتنغّص» was confirmed by the AI review); the About page and the text's banner say the whole vowelling is AI; the word panel's "text before AI" shows the print's marks only. - Stats: 549 words carry marks the print itself prints (642 marks); AI added 10,183 marks, so 2,959 of 2,959 words are marked; Kraus's marks and the AI's can be told apart by character. Every vowelled passage was checked by
data/sources/razi-tibb/vowelling/check.py: same letters word by word (hamza seats and Kraus's spelling kept), no printed mark lost; none was rejected. - Doubtful vocalisations queued: 20 words (
amb.json; shown in the word panel) plus the 22 translation doubts:razi-tibb-review.md. - Kraus's tanwīn placement (on the alif) is kept where printed. Printed marks that look like misprints are kept as printed and queued («الْإِنْسَانِ», «مُغْتَبِطَةَ», «قَبَلِهِ», «مُرَوِيِّةً»).
- No authoritative fully vowelled edition exists to compare with.
7. Segmentation and word IDs
18 passages: the preface, the plan, chapter 1 in two, chapter 2 in fourteen (our own paragraphing, by the author's turns «فنقول / فأقول / قالوا / ويرى» and the short last lines of paragraphs on the printed page). IDs rt:<passage>:<word> (punctuation rt:<passage>:p<n>); each passage records its Kraus pages and the word at which each page begins, and its OpenITI lines with a GitHub link (for ¶13 the range includes the repeated block). The manuscript's folio breaks (Kraus's «|» and margin numerals ٤٤ظ … ٤٩ظ) are recorded in folios.
8. Analysis layer
CAMeL Tools (calima-msa-r13 + MLE) via scripts/fusus/analyze_ch.py razi-tibb.json razi-tibb-analysis.json rt, constrained by the marks: 2,959 words, all analysed, 1,815 with a root, 476 distinct roots, 115 vowel conflicts. The AI sweep is narrow, not the Sīra's whole-text sweep: a script flagged the 74 words CAMeL could not analyse or tagged as names (prop/"NO ANALYSIS"); each was read in its sentence and 67 were corrected (60 roots set; rootFix/lemmaFix with the reason; scripts/razi_tibb/make_sweep.py, apply_fixes.py). The other ~2,880 words were not read one by one. The AI (opus) review of 13 Oct 2026 checked 40 more at random (5.3 % wrong) and set the roots CAMeL left unresolved or wrong for 99 words by lemma (data/sources/razi-tibb/review/1.json). A random sample of 60 of them was read in context: 1 wrong (وعظم tagged a verb: fixed), so the error rate on unflagged words is perhaps a few per cent (small sample). The contextual glosses for the word panel come from the translation pass; function words use one uniform gloss (FUNC in assemble_ai.py) and the many-context ones (من, أن) are therefore only approximate.
9. Commentaries aligned
None here. Al-Kirmānī's refutation (al-Aqwāl al-dhahabiyya) is quoted in Kraus's footnotes (the ✲ notes on pp. 15, 17, 19, 22 …); those notes are recorded in the apparatus (kirmani) and not shown.
10. Translation
Readable AI English for each passage, own wording, from the vowelled Arabic with the glossary in data/sources/razi-tibb-meta.json (ṭibb rūḥānī "Spiritual Medicine", hawā "passion", ʿaql "reason", nafs "soul", the three souls, zamm "reining in", shared with the Sīra where the words overlap: ṭibb rūḥānī, hawā, ʿaql, nafs). Arberry 1950, Adamson 2021 and all other translations: cited only, not opened, not consulted for wording. Per the editor's rule of 13 Oct 2026 no literal line is generated or shown; the per-word contextual glosses stay for the word panel. 22 doubts and 15 term notes are recorded, all ✦ AI. The AI (opus) review of 13 Oct 2026 settled the 22 doubts, read the whole English against the Arabic and made 16 revisions; no person has reviewed it. Triangulation (§17) was not run.
11. Rights and private layers
As the Sīra's §11 (Kraus "likely public domain, not legally cleared"; OpenITI CC BY-NC-SA 4.0 as a witness only; page images not rehosted; the AI reading of K is our working transcription; translations cited only). The NYU scan was downloaded to the session scratchpad only (169.8 MB) and not committed. Hidden: id in HIDDEN_TEXTS; catalogue/library entries in HIDDEN_CATALOGUE/HIDDEN_LIBRARY; word detail in app/src/data/hidden/words/razi-tibb/ (never under public/); HIDDEN in scripts/build_client_data.py; not in scripts/build_data.py EXTRA_TEXTS, so no public root page is touched. To publish: follow FLAGS K.
12. Open questions / to review
razi-tibb-review.md: A readings (settled by the AI review on the page images; A1 an emendation), B OpenITI's repeats, C apparatus and sigla, D vowelling doubts, E translation doubts, F analysis, G claims. The review is settled except C2–C3 (apparatus, al-Kirmānī) by the AI (opus) review of 13 Oct 2026; Kraus's introduction was read for the sigla only.
13. Connections (editor's proposals; none is a verified lineage)
razi-sira(same author, same Kraus volume): the Sīra (¶3–4, ¶9) gives the argument and says its detail is in the Ṭibb; the Ṭibb ¶2's chapter list, ch. 2's «reason against passion» and the practice of reining in passion are the manual. Scholars read the Ṭibb as more ascetic than the Sīra (scoping note citing Adamson, UNVERIFIED here): shown, not smoothed. The Sīra's «likeness to God» formula is not in these chapters.- Plato on the site (
alcibiades-1-mirror,republic-sun-line-cave,phaedrus-wings): ¶15–17 report «Falāṭun» and the three souls; the tripartite soul is the Republic IV/Phaedrus background as a parallel (reception), not a lineage; the passage is attributed «to Plato in al-Rāzī's report», not a translation of any dialogue. No dialogue is on the site for Socrates the withdrawn one (¶17) beyond the Sīra's. ikhwan,hayy-tufayl,gnomologia-mubashshir-siwan: as in the Sīra record §13 (unexamined here).- The reverse link (Sīra → Ṭibb) is added at publication (a public text must not name a hidden id).
14. Companions (app/src/content/companions/razi-tibb.json)
Five link-out items carried over from the Sīra's checked list (SEP entry, NYU volume, Wellcome records, Hackett): not re-fetched. No recording or reading of the text found.
15. Checks (13 Oct 2026)
ESLint on the changed files: clean. node scripts/design_audit.mjs --lint: 0 off-system values. Public build (no PRIVATE_LAYERS) next build --webpack: passes; /read/razi-tibb is not generated; razi-tibb is absent from app/public/ and from .next/static (client chunks); it appears only in server-side trace lists (*.nft.json), by design. Private dev server (tibb-private, port 3140): /read/razi-tibb, /about, /library, /read/razi-sira and /data/hidden/words/razi-tibb/p3.json answer 200; no literal English line on the page. Screenshots (scratchpad tibb/tibb-1280.jpg, tibb-375.jpg, not committed).
16. Change log
- 2026-10-13: record created; page-image reading, base, collation, full AI vowelling, CAMeL analysis with the flagged-word sweep, readable AI English, versions and companions; wired hidden; launch config
tibb-private(port 3140). Draft PR; not merged. - 2026-10-13: AI (opus) review, 13 Oct 2026 (not a person). Page images at native resolution: 47 spans and the doubtful readings settled; 4 misreadings and 4 printed marks corrected (
fixes.json); 3 full pages re-read, 0.15 % letter errors; sigla identified. 20 vowel doubts settled; the 2-paragraph check found 1.0 % errors, and 3 moods and the corrected words were fixed. 40-word analysis check, 5.3 %: roots set by lemma for 99 words (data/sources/razi-tibb/review/1.json). 22 English doubts settled and 16 revisions; glosses rekeyed. «أتنغّص» is no longer probable. Rebuilt: build_base → collate → assemble → analyze_ch (CAMeL) → apply_fixes (sweep, review) → mark_words → assemble_ai → reader_data. - 2026-10-13: published (FLAGS M): unhidden, library and catalogue next to razi-sira, word files public, registered in the data build, connections and search rebuilt, reciprocal link from razi-sira, page notes updated; ROADMAP item for chapters 3–20.
- 2026-10-13: page images added ("show the evidence"): 18 passages on pp. 15–32, 1 crop. See
docs/design/PAGE-IMAGES.md.