# `scriptures/` — sources & accuracy notes

This file records what each scripture in the `scriptures/` module was imported
from, how it was transformed, and the **soft spots** — what is owed before a
scripture can move from `draft` to `verified`. Read this before the content ships.

## Posture

The repo's standing rule (PLAN.md Conventions) is that content is hand-verified
against canonical text. The scriptures are an authorised exception, on the same
"researched / imported, shipped provisional at `draft`" footing as the Phase 3
kathas and the Phase 8 puja/navratri modules:

- A scripture is **imported from a public-domain digital edition**, not hand-
  typed. The import is deterministic — a downloaded dataset transformed by a
  recorded, re-runnable script — so the result is a faithful copy, not a
  transcription.
- Each scripture ships at status `draft` in the PLAN.md coverage tracker. It
  awaits a human verification pass against a printed canonical edition (Gita
  Press or equivalent) before `verified`.
- ~~Translations are **not** imported — published translations are frequently
  copyright-encumbered.~~ **Revised 2026-05-29:** a translation **is** included
  where a genuinely **public-domain** one exists. The Geeta now carries Shri
  Purohit Swami's 1935 English (life+70 public domain since 2012) on every
  verse — see below. Modern copyrighted translations are still excluded. IAST
  is **generated** from the Devanagari (a deterministic script conversion), not
  imported; where it cannot be generated soundly as strict IAST (the Awadhi
  Sundarkand and Ramcharitmanas), a **best-effort romanisation** is generated
  instead and the `iast` field documents that it is a pronunciation aid, not a
  Sanskrit IAST claim. See each scripture below.

---

## `bhagavad-geeta` — Bhagavad Geeta

**Source.** The `gita/gita` dataset on GitHub (`github.com/gita/gita`,
`data/verse.json` + `data/chapters.json`), retrieved from the `main` branch.
The repository is released under the **Unlicense** — an unconditional public-
domain dedication — so the text may be redistributed without restriction.

**What was imported.** Per verse, only the Devanagari verse text (`verse.json`
`text`). Per chapter, the Sanskrit chapter name in Devanagari (`chapters.json`
`name`) and a short English chapter title (`name_meaning`). The dataset's
`transliteration`, `word_meanings`, `translation`, and `commentary` were **not**
imported — the `transliteration` field is an ad-hoc scheme (`śh`, `ṣh`, `ch`),
not the strict IAST the repo requires, and the translations/commentary are
attributed to named authors. The IAST shipped on each verse is generated
separately (see "The IAST" below), not taken from the dataset.

**Structure verified.** All 18 chapters were checked: the per-chapter verse
count matches the dataset's own metadata **and** the canonical verse counts
(47, 72, 43, 42, 29, 47, 30, 28, 34, 42, 55, 20, 35, 27, 20, 24, 28, 78), and
every chapter's verse numbers are gap-free and sequential. Total **701 verses**.

**Recension.** The text is the **701-verse recension** — chapter 13 carries 35
verses (the verse `prakṛtiṃ puruṣaṃ caiva…` is counted). Printed editions that
quote a round "700" omit that verse from the count; no verse content differs.

**The Devanagari — repairs applied, all verified.** The source `text` needed
two kinds of correction during the original import, plus a further six classes
of repair from the 2026-05-20 cross-source audit. All transforms are applied
by re-runnable scripts (the original generator + `scripts/audit-fix.cjs`).

Original import (2026-05-19):

1. *Line breaks.* 479 verses use `\n\n` between pādas; 222 carry the whole
   verse on one line, the speaker tag and pādas run together
   (`…भगवानुवाचऊर्ध्वमूलम…`, `…व्रज।अहं…`). The script inserts newlines at two
   boundaries only — after a verse-initial speaker tag (one of five exact
   tags), and after a lone daṇḍa (`।`) glued to the next character.
2. *A transposed i-matra.* 25 verses mis-encoded the निश्चित / त्रिविध family —
   a virama and an i-matra placed adjacent (`्ि` or `ि्`), which is malformed
   Devanagari (a vowel sign cannot sit beside a virama). The repair swaps the
   matra onto the consonant the virama joins. This **only reorders** the three
   characters involved.

Audit-driven repairs (2026-05-20; full record in `AUDIT.md`):

3. *Missing instrumental-plural suffix.* 17.25 ended `…मोक्षकाङ्क्षि।।17.25।।`
   in the source; canonical (IIT Kanpur, SD, Mukundananda) is
   `…मोक्षकाङ्क्षिभिः।।17.25।।`. The script appends `भिः` before the trailing
   marker. One verse changed.
4. *Phantom `र` in शृ-imperative.* 14 verses encoded the imperative root
   `śṛ-` ("listen!") as `श्रृ` (श + virama + र + ृ), inserting an extra `र`
   consonant. Canonical is `शृ`. The script removes the spurious `्र` pair.
   Affected: 2.29, 2.39, 5.8, 10.1, 10.18, 13.4, 16.6, 17.2, 17.7, 18.4,
   18.29, 18.36, 18.64, 18.71.
5. *Letter substitution at 1.34.* Source had `मातुलाः श्चशुराः` (`श्च-`, not a
   Sanskrit word); canonical is `श्वशुराः` (`śvaśura` = father-in-law).
   One verse changed.
6. *Verse-initial vowel at 16.20.* Source started `असुरीं` (short `अ`); canonical
   is `आसुरीं` (long `आ`, the adjectival "demoniacal" form). One verse changed.
7. *Decomposed long-vocalic-R.* Three verses (1.26, 9.25, 10.29) encoded
   `पितॄन` as `पितृ़न` (short `ृ` + combining nukta), a non-canonical Unicode
   form. The script replaces `ृ + U+093C` with `ॄ` (U+0944). Four occurrences
   across three verses. **Side benefit**: the IAST output for these verses was
   producing literal Unicode nuktas (`pitṛ़n` instead of `pitṝn`); the repaired
   Devanagari transliterates cleanly.
8. *Stray nukta at 4.31.* Source had `कुतो़ऽन्यः` with a U+093C nukta between
   the vowel sign and the avagraha (no consonant to modify — pure encoding
   noise). The script removes it. One verse changed.

Total: 21 verses changed by the audit fix.

All changes are **asserted programmatically** by the per-fix expected-count
invariants in `scripts/audit-fix.cjs` (e.g. exactly 14 verses match the
phantom-र pattern; the multiset diff per affected verse matches the documented
character add/remove). Post-fix, `scripts/audit-verify.cjs` re-asserts every
invariant against the compiled `dist/`, including the IAST round-trip for all
701 verses (whitespace + daṇḍa-form + Devanagari-digit normalisation, plus
the documented `oṃ` ↔ `ॐ` toleration). The build aborts on any failure. The
verse content is a provably faithful copy of the canonical text-as-attested,
modulo the documented edition-variant choices.

**The IAST.** `ScriptureVerse.iast` is generated from the repaired Devanagari
with `@indic-transliteration/sanscript` — a deterministic, mechanical script
conversion, not an interpretation. It is verified per verse by a round-trip:
the IAST is converted back to Devanagari and compared with the source at the
akshara level (all 701 pass; the check tolerates `sanscript`'s harmless
re-ligature of `oṃ` → `ॐ`). The IAST is still machine-generated and has not had
a human Sanskrit-reviewer pass — see soft spots.

### Soft spots — owed before `verified`

1. **No human verification against a printed edition.** The content is faithful
   to the `gita/gita` dataset *and* to the audit-corrected canonical text
   (cross-checked against IIT Kanpur, Sanskrit Documents, Mukundananda), but
   none of those sources has itself been checked here against a printed Gita
   Press (or comparable) edition. That pass is owed.
2. **Verse-number markers.** Each verse keeps the source's trailing marker
   (`।।1.1।।` etc.). These use ASCII digits and are the dataset's annotation,
   not strict scripture punctuation. Verse **18.78** carries a malformed marker
   in the source — `।।18.78।`, a single trailing daṇḍa — and is reproduced
   verbatim. A reviewer may normalise the markers (or strip them) in the
   verification pass.
3. **The IAST is machine-generated.** It is a faithful, round-trip-verified
   transliteration of the (audit-corrected) Devanagari, but has not had a human
   reviewer pass; it ships at `draft` like the rest.

**English translation (added 2026-05-29).** `ScriptureVerse.english` now
carries **Shri Purohit Swami's** translation (*The Geeta: The Gospel of the
Lord Shri Krishna*, 1935). Purohit Swami died in 1941, so the work is public
domain under life+70 (since 2012), independent of the `gita/gita` dataset's
own Unlicense dedication — it is the only one of the dataset's five English
translators who is unambiguously public domain (the others — Sivananda d.1963,
Adidevananda d.1983, Gambirananda d.1988, Sankaranarayan, modern — are not and
were deliberately not used). Imported per-verse by
`scripts/add-geeta-translation.cjs`, joined on the dataset's `verse_id`
(1–701) and **alignment-asserted**: the script parses each verse's `।।C.V।।`
marker and checks it against the expected global index for the 701-recension,
aborting on any mismatch (all 701 pass). The translation is itself `draft`: it
is a faithful copy of Purohit Swami's text, not reviewed against a printed
edition, and his 1935 idiom is occasionally free/Victorian rather than literal.

**Hindi bhāvārtha (added 2026-05-29; pilot Chapter 1, then extended to all 18
chapters / 701 verses).** `ScriptureVerse.hindi` carries a Hindi **meaning
gloss (bhāvārtha)**, not a copied translation. This is a deliberate, authorised
exception: **no public-domain, verse-aligned Hindi translation exists** to
import the way Purohit's English was — the standard Hindi sources are all
copyright-encumbered (Swami Ramsukhdas / Gita Press *Sādhaka Sañjīvanī*, d. 2005
→ PD 2076; Swami Tejomayananda, living; the Goyandka Gita Press Hindi, d. 1965 →
PD ~2036). So the Hindi here is **researched and authored** on the same footing
as every `meaning.hi` field elsewhere in the package (mantras, prayers,
ashtottara, the kathas) and the Sundarkand English: it conveys the *sense* of
each verse, cross-referenced against standard Hindi commentaries for accuracy of
meaning but **not transcribed** from any one of them (copying would infringe).
The Sanskrit Devanagari — the devotional item — is untouched and remains the
sourced canonical text; the bhāvārtha is an explanatory aid for readers who
cannot read Sanskrit, alongside Purohit's English.

The gloss is stored inline per-verse in `bhagavad-geeta.ts` (the `hindi` field),
which is the single source of truth; it was authored one chapter at a time and
injected by a one-off generator (since removed). **Status: complete — all 18
chapters, 701 verses**, shipped `draft`.
Chapter 1 was authored first as a pilot (2026-05-29); chapters 2–18 were then
authored on the same footing. It remains provisional and **owes a human reviewer
pass against a printed Gita Press edition** — `audit-verify.cjs` asserts only
that the gloss is *present* on every verse (and that no chapter has gaps), which
is **not** a check of its correctness (a machine cannot verify an authored
meaning; that is exactly the human-review debt). The Sundarkand Hindi bhāvārtha
was authored on the same footing (see the `sundarkand` section below).
4. **Chapter English titles** are the `gita/gita` `name_meaning` strings
   (e.g. "Arjuna's Dilemma"); they are descriptive, not a fixed canonical
   rendering, and a reviewer may prefer the traditional "Yoga of…" forms.
5. **Chapter colophons (added 2026-05-29).** Each chapter now carries its
   traditional closing colophon in `Chapter.colophon` —
   `ॐ तत्सदिति श्रीमद्भगवद्गीतासूपनिषत्सु ब्रह्मविद्यायां योगशास्त्रे
   श्रीकृष्णार्जुनसंवादे <अध्यायनाम>ो नाम <क्रमसूचक>ोऽध्यायः।।`. This is the
   fixed canonical formula, identical across printed editions and Sanskrit
   Documents, varying only by the chapter name (taken from `titleSanskrit`,
   readability hyphen dropped for ch.13) and the ordinal — so it is *sourced
   formulaic text*, not an authored rendering. `colophon.iast` is machine-
   generated and **round-trip-verified** (Dev→IAST→Dev, danda-normalised),
   the same bar as the verse IAST; asserted in `audit-verify.cjs`. No
   translation is shipped (the formula is paratext, not a verse). See
   `AUDIT.md` §A4.
6. **Edition variants left as the source has them.** Three readings where IIT
   Kanpur / `gita/gita` differ from SD / Mukundananda — 2.33 (`चै` vs `चे`),
   2.64 (`वियुक्तैस्तु` vs `विमुक्तैस्तु`, 3-vs-1 sources agree with shipped),
   18.68 (`इमं` vs `इदं`) — kept as IIT/source has them. A reviewer may switch
   to the alternate reading. See `AUDIT.md` §A3.

---

## `sundarkand` — Sundarkand

**Edition.** Tulsidas's **Sundar Kand** — the fifth kand of the Ramcharitmanas
(Awadhi), the version recited in homes as "Sundarkand path". This was chosen
over the Valmiki Ramayana's Sanskrit Sundara Kanda; PLAN.md Phase 4 has been
corrected to match (it previously specified the Valmiki text, while the repo's
Hanuman puja already described it as the Ramcharitmanas canto).

**Source.** The `WirelessAlien/Ramcharitmanas` dataset on GitHub
(`main/data/verses/sundarkanda.json`, `main` branch) — released under the
**Unlicense**, an unconditional public-domain dedication. The dataset is
Devanagari-only and carries no translation or commentary.

**What was imported.** The 60 doha-numbered segments of the kand, each holding
its chaupai couplets, doha/soratha, and (where present) chhand, with the
structural labels (चौपाई, दोहा/सोरठा, छंद) and the doha numbers as the source
prints them. Plus, restored in the 2026-05-20 audit, the three opening Sanskrit
mangalacharan shlokas (`शान्तं शाश्वतम्…`, `नान्या स्पृहा…`, `अतुलितबलधाम्…`).
Mapped as a single `Chapter` with 63 `ScriptureVerse` entries — verses 1–3 are
the mangalacharan, verses 4–63 are doha-segments 1–60.

**Structure verified.** 60 doha-segments, doha 1 → 60 in order (segments 39,
49, 56 carry क/ख split dohas, which is normal for the text). The original
import is whitespace-only for the doha-segments: the script asserts that the
cleaned text, with all whitespace stripped, is byte-identical to the source
— so the doha content is a faithful copy and only line breaks were normalised.

**Audit-driven repairs.**
- (2026-05-20) One missing ृ matra in segment 44 (`दुष्टहदय` → `दुष्टहृदय`,
  "evil-hearted"), confirmed canonical by both Drik Panchang and Sanskrit
  Documents Tulsidas. See `AUDIT.md` §B.
- (2026-05-29) One stray-nukta encoding defect in doha-segment 25
  (`गर्ज़ा` → `गर्जा`, "[the monkey] roared"). The source carried ज + U+093C
  (ज़, the Persian-z form), which does not occur in Tulsidas's Awadhi —
  `अट्टहास करि गर्जा कपि` ("laughing aloud the monkey roared"). The spurious
  nukta was removed (same defect class as the Geeta's 4.31 stray nukta). The
  romanisation script's residual-nukta assertion is what surfaced it.

**Mangalacharan provenance.** Confirmed identical text across three
independent authoritative sources: Drik Panchang
(`drikpanchang.com/lyrics/ramcharitmanas/sundarkand/sundarkand-lyrics.html`),
Sanskrit Documents Tulsidas
(`sanskritdocuments.org/doc_z_otherlang_hindi/sundarakaaNDa.html`), and
BhaktiBharat. Where the three differed only on minor orthography
(`पुङ्गव`/`पुंगव`, `चूड़ामणिम्`/`चूडामणिम्`), the form matching the rest of the
corpus's homorganic-nasal / no-spurious-nukta style was chosen.

### Soft spots — owed before `verified`

1. **No human verification against a printed edition.** As with the Geeta, the
   content is faithful to the dataset (with the one repair) and to the audit-
   confirmed mangalacharan, but no printed Gita Press Ramcharitmanas has been
   collated against here.
2. **Romanisation (added 2026-05-29) — best-effort, not strict IAST.**
   `ScriptureVerse.iast` now carries a romanisation for all 63 verses,
   generated by `scripts/add-sundarkand-roman.cjs`: `sanscript`
   Devanagari→IAST, then a post-process that converts the two Awadhi retroflex
   flaps to ISO-15919-style forms — **ड़ → ṛ** and **ढ़ → ṛh** (sanscript
   emits `ḍa़` / `ḍha़` with a stray U+093C nukta; the script rewrites these,
   dropping the spurious inherent-a when an explicit vowel follows). The script
   **asserts no residual nukta** survives. The three opening *mangalacharan
   shlokas* are Sanskrit and romanise as clean IAST; verses 4–63 are Awadhi and
   the field is therefore a **pronunciation aid, not a Sanskrit IAST claim** —
   the flap convention (ṛ/ṛh) collides notationally with vocalic-ṛ, and
   candrabindu nasalisation renders as `~`. A human reviewer pass on the Awadhi
   romanisation is owed.
3. **English — researched/authored, shipped `draft`.** No public-domain
   English of Tulsidas's Sundarkand exists pre-aligned to this edition's 60
   doha-segments for a faithful deterministic import (Growse 1877 is public
   domain but continuous prose; Gita Press and modern web translations are
   copyright-encumbered). With Ishank's explicit authorisation (2026-05-29),
   `ScriptureVerse.english` now carries a **researched prose rendering** of
   each segment, authored from the segment's Awadhi content — the same
   "researched, shipped draft" footing as the Phase 3 kathas and the
   puja/navratri modules, injected by `scripts/add-sundarkand-english.cjs`. It
   conveys the **meaning of each doha-segment, not a word-for-word line
   translation**, and is provisional: it awaits a human reviewer pass against a
   printed Gita Press Ramcharitmanas (with attention to the few interpretive
   spots — e.g. the rendering of doha 59's `ढोल गवाँर…` line, and the
   bracketed glosses supplied for elliptical half-lines).
4. **Hindi bhāvārtha — researched/authored, shipped `draft` (added 2026-05-29).**
   `ScriptureVerse.hindi` carries a **modern-standard-Hindi meaning gloss** of
   each of the 63 segments. Tulsidas's original is archaic Awadhi, not fully
   transparent to a present-day reader, so a contemporary-Hindi *bhāvārtha* is a
   genuine reading aid (the long-standing टीका tradition for the Ramcharitmanas).
   Same posture as the Geeta bhāvārtha and the Sundarkand English: no public-
   domain verse-aligned Hindi टीका exists (modern published commentaries are
   copyright-encumbered), so the gloss is **authored from each segment's sense,
   not transcribed** from any one of them. It renders the **whole segment**
   (chaupai couplets + closing doha) into clear Hindi prose, conveys meaning
   rather than a line-by-line translation, and is provisional — it owes the same
   human reviewer pass against a printed Gita Press Ramcharitmanas. Stored inline
   in `sundarkand.ts`; `audit-verify.cjs` asserts only that the gloss is
   *present* on all 63 segments, which is not a correctness check.
5. **Structural labels are inside the verse text.** Each verse's `devanagari`
   includes the printed section labels (चौपाई / दोहा/सोरठा / छंद), kept because
   printed editions carry them. A consumer wanting only the verse lines can
   strip them. The three mangalacharan shlokas have no structural labels.

---

## `ramcharitmanas` — Ramcharitmanas (full seven-kand)

**Edition.** Tulsidas's **Ramcharitmanas** (śrīRāmacaritamānasa) — the complete
Awadhi retelling of the Rama story in seven kāṇḍas (Bāla, Ayodhyā, Araṇya,
Kiṣkindhā, Sundara, Laṅkā, Uttara). Phase 17 of the build plan; the
Valmiki-vs-Tulsidas decision for the `ram` epic was resolved to **Tulsidas**,
consistent with the already-shipped standalone `sundarkand` (which is this
work's fifth kand).

**Structure.** One `Scripture` (`slug: 'ramcharitmanas'`, `deity: 'rama'`) with
**seven chapters, one per kand**, numbered 1–7. Each chapter's verses are the
**doha-numbered segments** as the source prints them — a segment holds its
chaupai couplets, doha/soratha, and (where present) chhand, with the printed
structural labels (चौपाई / दोहा/सोरठा / छंद). Verse counts: Bāla 361, Ayodhyā
326, Araṇya 46, Kiṣkindhā 30, **Sundara 63**, Laṅkā 121, Uttara 130 — **1077
total** (1014 newly imported here + 63 in the reused Sundar Kand).

**Sundar Kand (chapter 5) is reused, not re-imported.** Chapter 5 is the
already-shipped, fully-layered `sundarkand` scripture's single chapter, spread
in by reference and renumbered to 5 — so it keeps its 63 verses (3 mangalacharan
+ 60 dohas), its best-effort romanisation, its researched English prose, and its
Hindi bhāvārtha. The standalone `sundarkand` scripture stays exported unchanged
for back-compat; the audit asserts the two share the same verse objects.

**Source (the six new kands).** The `WirelessAlien/Ramcharitmanas` dataset on
GitHub (`main/data/verses/{balkanda,ayodhyakanda,aranyakanda,kishkindhakanda,lankakanda,uttarakanda}.json`)
— released under the **Unlicense**, an unconditional public-domain dedication.
Devanagari-only; no translation or commentary. This is the same dataset and the
same `sundarkanda.json` family the shipped Sundarkand was imported from.

**Transform (verified).** Identical to the Sundarkand import: split each
segment's `content` on `\r\n|\r|\n`, trim each line, drop empty lines, join with
`\n`. The transform was confirmed **byte-identical** against the shipped inline
Sundarkand for all 60 of its doha-segments (the only two diffs being the
documented audit repairs at segments 25 and 44), and the generator self-asserts
that whitespace-stripped output equals whitespace-stripped source for every one
of the 1014 new segments — so the import is a **faithful verbatim copy**, line
breaks normalised only. Emitted via `JSON.stringify` (safe double-quoted string
literals), one `src/scriptures/ramcharitmanas/<kand>.ts` per kand, by a one-off
generator (since deleted, per the inline-source-of-truth convention).

### Layering passes (Phase 17 cont., 2026-06-02)

The Devanagari-only first import is being layered up to the Sundarkand's
three-layer surface, kand by kand, on the same **researched, shipped `draft`**
footing the standalone Sundarkand was authored on (Ishank's posture decision,
2026-06-02: match the Sundar Kand's Awadhi model — best-effort romanisation +
researched English prose + Hindi bhāvārtha — *not* the Geeta's strict-IAST +
public-domain-import model).

- **IAST — all six new kands (1014 segments), done 2026-06-02.** A mechanical
  pass with `@indic-transliteration/sanscript` (dev→IAST), the **same romaniser
  logic as the Sundarkand** (`add-sundarkand-roman.cjs`): the two Awadhi retroflex
  flaps are post-processed — `ड़ → ṛ`, `ढ़ → ṛh` (dropping the spurious inherent-a
  before an explicit vowel) — anusvāra = `ṃ`, and the चौपाई/दोहा/सोरठा/छंद labels,
  the `।`/`।।` daṇḍa markers, and the verse numbers all romanise line-parallel to
  the Devanagari. Embedded Sanskrit श्लोक segments come out as near-strict IAST.
  This is a **pronunciation aid, asserted for PRESENCE only** (the flaps have no
  IAST letter, so it is *not* round-trippable). The romaniser was run, verified,
  and **deleted** (inline source of truth). Result: every verse of all seven
  kands now carries an `iast`, zero residual nuktas.
- **Researched English prose + Hindi bhāvārtha — Araṇya (46) + Kiṣkindhā (30),
  done 2026-06-02.** Authored as a **multi-agent workflow**: each ~8-segment
  chunk was rendered by an author agent (given 3 fully-layered Sundar Kand
  segments as the target voice/length, the anchor references below, and the hard
  rule — researched SENSE-rendering of the whole dohā-segment, *never* a copied
  translation, no source named in any user-facing field), then an **adversarial
  reviewer agent** per chunk checked each gloss against the Devanagari for
  faithfulness, dropped content, hallucination, copy-paste tells, and provenance
  leaks, returning a corrected final. The reviewer caught and fixed **14 real
  defects** across the two kands (restored dohās dropped by the author at
  Araṇya 14/15/21/32; corrected `बिनु फर` "blunt/headless arrow" misread as
  "featherless", Araṇya 24; fixed a reversed sense at Araṇya 45 — *apart from* my
  lotus feet, not *forsaking* them; corrected speaker attributions Kiṣkindhā 1/8
  — Rama's words, not Hanuman's/Bali's; removed an invented "frogs" at
  Kiṣkindhā 16 = `मसक दंस` mosquitoes/gnats; fixed the Sampati/Jatayu conflation
  at Kiṣkindhā 26). The verified glosses were collected from the run journal and
  injected inline by a one-off injector (since **deleted**). Both authoring and
  injection match the Sundarkand pattern.
- **References consulted (research anchors — NEVER named in any user-facing
  field):** the Gita Press Mānas ṭīkā sense, `ramcharit.in`, `hindipath.com` —
  the same anchors the Sundarkand English/Hindi and the Hanuman Bāhuk used.

**Three-layer coverage (chapter / kand):** ch3 Araṇya, ch4 Kiṣkindhā, ch5 Sundar
(reused) carry `devanagari + iast + english + hindi`. ch1 Bāla, ch2 Ayodhyā,
ch6 Laṅkā, ch7 Uttara carry `devanagari + iast` (researched English + Hindi
pending, one kand per later session — Bāla 361 / Ayodhyā 326 / Laṅkā 121 /
Uttara 130). The audit's `MEANING_LAYERED = [3, 4, 5]` locks this frontier; bump
it as kands are layered.

### Soft spots — owed before `verified`

1. **Researched English + Hindi remain owed on four kands.** ch1 Bāla, ch2
   Ayodhyā, ch6 Laṅkā, ch7 Uttara are IAST-only for meaning (researched English
   prose + Hindi bhāvārtha pending, one kand per later session). They ship
   `draft`. **Human-pass flags from the Araṇya + Kiṣkindhā authoring** (the
   reviewer's notes, for a printed-edition collation): Araṇya 3 (the Sanskrit
   `शची पतिं प्रियानुजं` śloka — ṭīkā-tradition meaning split, taken here as Indra's
   beloved younger brother / Upendra-Rāma), Araṇya 10 (the dense Sanskrit
   Chandāvalī stuti), Araṇya 19 (the long battle-chand ghoul/yoginī imagery),
   Araṇya 28 (`दंड एक` rendered "for a while" not the literal ~24-minute *daṇḍa*),
   and the 11 reviewer-fixed segments above — all worth a human eye against a
   printed edition.
2. **No opening Sanskrit mangalacharan on the six new kands.** The dataset's
   per-kand files begin at the first chaupai (Bāla opens `बंदउ गुरु पद पदुम परागा`),
   so the famous opening Sanskrit invocations of each kand (e.g. Bāla's
   `वर्णानामर्थसंघानां…` / `भवानीशंकरौ वन्दे…`) are **not in the source and not
   shipped** in this pass — exactly the situation the shipped Sundarkand was in
   before the 2026-05-20 audit restored its three mangalacharan shlokas. A later
   audit pass can restore the per-kand invocations after cross-source
   confirmation.
3. **No human verification against a printed edition; 3 stray-nukta defects
   located.** As with the Geeta and Sundarkand, the content is faithful to the
   public-domain dataset but no printed Gita Press Ramcharitmanas has been
   collated against it. The IAST pass (2026-06-02) surfaced exactly **three stray
   Perso-Arabic nukta defects** in the shipped Devanagari — `ज़ंबु` (Araṇya,
   intended `जंबु`), `ज़ाऊँ` (Uttara, intended `जाऊँ`), and `सुग़ीव` (Laṅkā, intended
   `सुगीव`/`सुग्रीव`) — the same class the Sundarkand audit repaired (`गर्ज़ा → गर्जा`).
   Per the no-silent-edit rule the **Devanagari is left untouched** (defect
   shipped, to be repaired in the later sweep after second-source confirmation);
   the **IAST renders the intended base consonant** (nukta dropped → `jaṃbu` /
   `jāū̃` / `sugīva`). The dropped-`ृ`-matra class and a full printed-edition
   collation still belong to that later audit pass.
4. **Structural labels are inside the verse text** (चौपाई / दोहा/सोरठा / छंद),
   as in the Sundarkand — kept because printed editions carry them; a consumer
   wanting only the verse lines can strip them.

---

## `isha-upanishad` — Īśa (Īśāvāsya) Upaniṣad

**Text.** The Īśa Upaniṣad — the closing 40th adhyāya of the Śukla Yajurveda
(Vājasaneyi Saṃhitā), Kāṇva recension, **18 mantras** — framed by the
`पूर्णमदः…` śānti-pāṭha. This is the first of the **Phase 16 principal
Upaniṣads** (Batch A). Unlike the Phase 15 sūktas, the principal Upaniṣads are
presented as **clean (non-accented) Sanskrit**, so the IAST is **strict,
round-trip-verified IAST** (Dev→IAST→Dev, anusvāra = `ṃ`) — the same bar the
Geeta IAST meets, *not* the sūktas' accent-stripped best-effort posture.

**Source & cross-check.** Devanagari imported from a public digital edition and
cross-checked against **two independent Devanagari editions** — a separate
commentary edition of the same text and an independent lyrics edition —
at **95.0%** and **93.7%** token containment (the residual misses are
orthographic: anusvāra-vs-homorganic-nasal and sandhi-boundary variants, not
content differences). The intended sa.wikisource page (`ईशोपनिषद्` /
`ईशावास्योपनिषद्`) is an **empty redlink**, so two other independent editions
were substituted. Digitization sources are recorded internally only and **never
named in any user-facing field**.

**Transforms applied (all deterministic, asserted in `audit-verify.cjs`).**
1. *Vedic nasalisation normalised.* The source prints the Vedic anusvāra/ardha
   forms `ꣳ` (U+A8F3) and `ँ` (candrabindu) where standard non-Vedic editions
   print plain anusvāra `ं` (e.g. `इदꣳ` → `इदं`, `तांस्ते`). Normalised to `ं`
   — the clean-Sanskrit surface used by home/prayer-book editions and the
   confirmed posture for this batch — so the IAST is uniform śāstrīya `ṃ`. Both
   the faithful and normalised forms round-trip; the normalised form was chosen
   for a clean Geeta-bar surface.
2. *Line-wrap hyphens joined.* The poetry-layout hyphens in mantra 8
   (`अव्रण- मस्नाविरं`, `स्वयम्भू- र्याथातथ्यतो`) are typesetting artifacts;
   joined to `अव्रणमस्नाविरं` / `स्वयम्भूर्याथातथ्यतो`.

**Schema.** One `Chapter` (number 1) holding 19 `ScriptureVerse`s — the
śānti-pāṭha as the **leading verse**, then the 18 numbered mantras (the `॥ N॥`
Devanagari-digit markers kept inline, as the sūktas/Geeta keep theirs). The
śānti is recited at both the opening and the close (noted in `about`); the
closing `॥ इति ईशोपनिषत् ॥` is the chapter `colophon` (round-trip-verified
IAST). `deity` is left undefined (a non-deity-exclusive jñāna text).

**English + Hindi.** Researched per-mantra **meaning glosses** (sense-renderings
anchored to **Śaṅkara-bhāṣya**), authored on the same "researched, shipped
`draft`" footing as the Geeta/Sundarkand bhāvārtha and the sūkta glosses — **not
a copied translation**. The audit asserts only that the gloss is *present* on
every verse, which is **not** a correctness check.

### Soft spots — owed before `verified`
1. **No human verification against a printed edition.** Owed, as for every
   scripture in this module.
2. **The IAST is machine-generated** (round-trip-verified but no human
   Sanskritist pass).
3. **Contested mantras flagged for the human pass — mantras 9–14.** The
   `vidyā`/`avidyā` (9–11) and `sambhūti`/`asambhūti`/`vināśa` (12–14) pairs are
   read differently across Vedānta schools; the shipped gloss follows
   **Śaṅkara's** reading (avidyā = ritual action, vidyā = worship of deities;
   sambhūti = the manifest effect / Hiraṇyagarbha, asambhūti = the unmanifest
   cause) and 12–14 are textually intricate. These are the priority items for
   the reviewer.

---

## `kena-upanishad` — Kena (Talavakāra) Upaniṣad

**Text.** The Kena Upaniṣad — from the Talavakāra / Jaiminīya Brāhmaṇa of the
Sāmaveda — in **four khaṇḍas** (khaṇḍa 1–2 in verse: 9 + 5 mantras; khaṇḍa 3–4
in prose: 12 + 9 sections, the latter pair carrying the Yakṣa narrative with
Umā Haimavatī). Framed by the `आप्यायन्तु…` śānti-pāṭha. Same clean-Sanskrit,
strict-round-trip-IAST posture as the Īśa.

**Source & cross-check.** Devanagari imported from a public digital edition and
cross-checked against an independent Devanagari edition (sa.wikisource
`केनोपनिषद्`, which **does** exist) at **98.2%** token containment. Digitization
sources are internal only, never named in user-facing fields.

**Transforms applied (deterministic, asserted).**
1. *Vedic nasalisation normalised* (`ꣳ`/`ँ` → `ं`), as for the Īśa.
2. *Stray ZWJ removed.* Khaṇḍa 1.8 printed the imperative `शृणोति` with a
   zero-width joiner (`श‍ृ`, U+200D) — the same ZWJ-`शृ` rendering noted in the
   Geeta audit. Stripped to canonical `शृ`.
3. *Editorial apparatus dropped.* Khaṇḍa 2.1 carried the source's inline variant
   apparatus `दहरमेवापि var दभ्रमेवापि`; the primary reading `दहरमेवापि` is kept
   (the cross-check edition agrees) and the `var <alt>` dropped. The
   cross-check edition's `मीमाँस्येमेव` at 2.1 is a typo; the primary's
   `मीमांस्यमेव` is kept. A leaked English translator's footnote in khaṇḍa 4.4
   was stripped.
4. *Pluta kept.* Khaṇḍa 4.4 carries the pluti (vowel-protraction) marker `दा३`
   (`…व्यद्युतदा३ … न्यमीमिषदा३…`) — a genuine phonemic feature of the
   adhidaivata lightning-flash illustration. **Kept faithfully**; it renders in
   IAST as `ā3` and round-trips. This is the one non-standard IAST mark in the
   batch.

**Schema.** Four `Chapter`s (khaṇḍa 1–4), each with its traditional
`॥ इति केनोपनिषदि <ordinal> खण्डः ॥` colophon (round-trip-verified IAST); the
overall `॥ इति केनोपनिषत् ॥` closes the fourth khaṇḍa's colophon. The śānti-pāṭha
is the **leading verse of khaṇḍa 1** (so khaṇḍa 1 holds 10 verses: śānti + 9);
recited again at the close (noted in `about`). `deity` undefined.

**English + Hindi.** Researched per-mantra meaning glosses anchored to
Śaṅkara-bhāṣya, `draft`, presence-only audit.

### Soft spots — owed before `verified`
1. **No human verification against a printed edition.**
2. **Machine-generated IAST** (round-trip-verified, no human pass).
3. **Terse mantras flagged for the human pass — 2.1, 4.4, 4.5.** The
   teacher–pupil exchange of 2.1 (`यदि मन्यसे सुवेदेति…`), the elliptical
   adhidaivata illustration of 4.4 (the pluta), and the highly compressed
   adhyātma statement of 4.5 (`अथाध्यात्मं यद्देतद्गच्छतीव च मनः…`) are rendered
   to Śaṅkara's sense but are the densest spots; priority for the reviewer.

---

## `mandukya-upanishad` — Māṇḍūkya Upaniṣad

**Text.** The Māṇḍūkya Upaniṣad of the Atharvaveda — the shortest of the
principal Upaniṣads, **12 prose mantras** on Om (AUM) and the four states /
quarters of the Self (Vaiśvānara, Taijasa, Prājña, Turīya). Framed by the
`भद्रं कर्णेभिः…` śānti-pāṭha (the Atharvaveda śānti). Phase 16 **Batch B**.
Same clean-Sanskrit / strict-round-trip-IAST posture as Īśa/Kena.

**Source & cross-check.** Devanagari imported from a public digital edition and
cross-checked against an independent edition (sa.wikisource `माण्डुक्योपनिषद्`)
at **94.1%** token containment (misses are orthographic + the dropped maṅgala,
below). Digitization sources internal only; never named in user-facing fields.

**Transforms applied (deterministic, asserted).**
1. *Vedic nasalisation normalised* (`ꣳ`/`ँ` → `ं`), as for Īśa/Kena.
2. *ZWJ removed* (the `श‍ृ`/`शृ` rendering in the śānti).
3. *Maṅgala preamble dropped.* The source prints a scribal `हरिः ॐ ।` between
   the śānti and mantra 1; dropped (a recitational maṅgala, not part of the 12
   mantras or the śānti proper) and documented here.

**Schema.** One `Chapter` (number 1) of 13 `ScriptureVerse`s — the śānti-pāṭha
as the **leading verse**, then the 12 numbered mantras; closing
`॥ इति माण्डूक्योपनिषत् समाप्ता ॥` as the chapter `colophon` (round-trip-verified
IAST). `deity` undefined.

**English + Hindi.** Researched per-mantra meaning glosses anchored to
**Śaṅkara / Gauḍapāda**, `draft`, presence-only audit.

### Soft spots — owed before `verified`
1. **No human verification against a printed edition.**
2. **Machine-generated IAST** (round-trip-verified, no human pass).
3. **All 12 mantras are dense and technical** (the sapta-aṅga / ekonaviṃśati-
   mukha enumeration, the mātrā correlations). The whole text is a priority for
   the reviewer; the Gauḍapāda Kārikā is intentionally **not** shipped (only the
   12 Upaniṣad mantras).

---

## `katha-upanishad` — Kaṭha (Kāṭhaka) Upaniṣad

**Text.** The Kaṭha Upaniṣad of the Kṛṣṇa Yajurveda (Kaṭha school) — the
Naciketā–Yama dialogue — in **two adhyāyas of three vallīs each** (6 vallīs).
Framed by the `सह नाववतु…` śānti-pāṭha (the Kṛṣṇa-Yajurveda śānti). Phase 16
**Batch B**; the largest of the short Upaniṣads. Same clean-Sanskrit /
strict-round-trip-IAST posture.

**Source & cross-check.** Devanagari imported from a public digital edition and
cross-checked against an independent edition — sa.wikisource, whose Kaṭha is
fragmented across 6 per-vallī subpages (under the parent `कठोपनिषत्`), all six
fetched and concatenated — at **99.7%** token containment. Digitization sources
internal only.

**Verse count is edition-dependent.** This source's division gives **120
mantras** (vallī counts 29 / 25 / 17 / 15 / 15 / 19), gap-free and **identical
to the independent sa.wikisource division**. Some printed editions cite **119**
(with vallī 2.3 holding 18), counting the closing śānti differently or merging a
boundary verse. The text is the same; only the verse boundaries differ — the
same situation as the Geeta's 701-vs-700 recension note. Shipped faithfully as
the cross-checked sources divide it.

**Transforms applied (deterministic, asserted).**
1. *Vedic nasalisation normalised* (`ꣳ`/`ँ` → `ं`).
2. *Line-wrap hyphens joined* (poetry-layout artifacts, as in Īśa 8).
3. *Colophon-split fix (parser).* The 6 vallī colophons were used to segment the
   text; care was taken not to consume each vallī's last verse-marker `॥ N॥`.

**`द्ह` ↔ `ध` round-trip toleration.** In Kaṭha **2.2.2** (the famous *haṃsaḥ
śuciṣat…* verse) the word **`सद्होता`** (*sad* + *hotā*) carries the `द्+ह`
juncture. `sanscript` renders `द्ह` as `dh`, which collides with `ध`, so the
back-transliteration yields `सधोता` and the strict round-trip would fail. Both
sources (SD + sa.wikisource) attest `सद्होता`, so the Devanagari is kept faithful
and the audit folds `द्ह` ↔ `ध` symmetrically — a documented `sanscript`
artifact toleration, exactly like the Geeta's `oṃ` ↔ `ॐ` toleration. (The
`व्योमसद् अब्जा` epithet-string in the same verse is kept spaced as the sources
print it.)

**Schema.** Six `Chapter`s (one per vallī, numbered 1–6), each with its
traditional `॥ इति काठकोपनिषदि <adhyāya>ाध्याये <vallī> वल्ली ॥` colophon
(round-trip-verified IAST). The opening śānti-pāṭha is the **leading verse of
vallī 1** (so ch.1 holds 30 verses: śānti + 29); the **closing śānti is verse
2.3.19**, numbered as the source has it (which is why 2.3 = 19), and is glossed
as the closing śānti. `deity` undefined.

**English + Hindi.** Researched per-mantra meaning glosses anchored to
**Śaṅkara-bhāṣya** — including the famous chariot analogy (1.3.3–4),
`उत्तिष्ठत जाग्रत` (1.3.14), `न जायते म्रियते` (1.2.18), and
`न तत्र सूर्यो भाति` (2.2.15) — `draft`, presence-only audit.

### Soft spots — owed before `verified`
1. **No human verification against a printed edition** (incl. the edition-variant
   verse division noted above; a reviewer may prefer the 119-count split).
2. **Machine-generated IAST** (round-trip-verified, no human pass; the one
   `द्ह`↔`ध` toleration is documented above).
3. **Dense / figurative mantras flagged.** The adhyātma-yoga and avyakta/puruṣa
   hierarchy (1.3.10–13, 2.3.7–8), the *aṅguṣṭha-mātra puruṣa* verses (2.1.12–13,
   2.3.17), and the Vedic *haṃsaḥ śuciṣat* epithet-string (2.2.2) are rendered to
   Śaṅkara's reading and are priorities for the reviewer.

---

## `mundaka-upanishad` — Muṇḍaka Upaniṣad

**Text.** The Muṇḍaka Upaniṣad of the Atharvaveda — three muṇḍakas of two
khaṇḍas each (**6 sections, 64 mantras**), in verse. Framed by the
`भद्रं कर्णेभिः…` śānti-pāṭha (the Atharvaveda śānti, shared with Māṇḍūkya/Praśna).
Phase 16 **Batch C**. Same clean-Sanskrit / strict-round-trip-IAST posture.

**Source & cross-check.** Devanagari imported from a public digital edition and
cross-checked against an independent edition (sa.wikisource `मुण्डकोपनिषद्`) at
**98.5%** token containment. Digitization sources internal only; never named in
user-facing fields.

**Transforms applied (deterministic, asserted).**
1. *Vedic nasalisation normalised* (`ꣳ`/`ँ` → `ं`).
2. *ZWJ removed* (in the śānti's `शृ`).
3. *Scribal maṅgala dropped.* The source prints `ॐ ब्रह्मणे नमः ॥ १॥` between the
   śānti and muṇḍaka 1.1 (and would otherwise duplicate verse "1"); dropped (a
   recitational maṅgala, like Māṇḍūkya's `हरिः ॐ`) and documented.
4. *SD parenthetical variant-readings dropped.* Several mantras carried the
   source's inline alternates in parentheses — e.g. `(संदधीत)` (2.2.3),
   `(जयति)` (3.1.6 — kept primary `जयते`, India's motto reading), `(पराय)` /
   `(संप्रतिष्ठितः)` (2.2.6–7), `(आत्मा वृणुते)` (3.2.4), and a 1.2.2/1.2.3
   boundary alternate; the primary (main-text) reading is kept, the parenthetical
   alternates dropped (the Kena-`var`-apparatus posture).

**Schema.** Six `Chapter`s (muṇḍaka.khaṇḍa, numbered 1–6), each with its
traditional `॥ इति मुण्डकोपनिषदि …मुण्डके … खण्डः ॥` colophon (round-trip-verified
IAST). The opening śānti-pāṭha is the **leading verse of section 1** (so ch.1
holds 10 verses: śānti + 9). `deity` undefined.

**English + Hindi.** Researched per-mantra meaning glosses anchored to
**Śaṅkara-bhāṣya** — including the parā/aparā distinction (1.1.4–5), the spider
image (1.1.7), the two birds (3.1.1–2), and `सत्यमेव जयते` (3.1.6) — `draft`,
presence-only audit.

### Soft spots — owed before `verified`
1. **No human verification against a printed edition.**
2. **Machine-generated IAST** (round-trip-verified, no human pass).
3. **The karma-kāṇḍa critique (1.2) and the cosmogony (2.1) are dense**; rendered
   to Śaṅkara's reading and priorities for the reviewer.

---

## `prashna-upanishad` — Praśna Upaniṣad

**Text.** The Praśna Upaniṣad of the Atharvaveda — **six questions (praśnas)**
put by six seekers to the sage Pippalāda, in prose (**67 sections**). Framed by
the `भद्रं कर्णेभिः…` śānti. Phase 16 **Batch C**. Same posture.

**Source & cross-check.** Devanagari imported from a public digital edition.
sa.wikisource `प्रश्नोपनिषद्` / `प्रश्नोपनिषत्` are **empty redlinks**, so — as
with Īśa — an independent edition was substituted (shlokam.org all-verses page),
at **100% token containment**. Digitization sources internal only.

**Transforms applied (deterministic, asserted).**
1. *Vedic nasalisation normalised* (`ꣳ`/`ँ` → `ं`).
2. *ZWJ removed* (in the śānti's `शृ`).

**Section markers.** Praśna numbers its sections `॥P.N॥` (praśna.section — e.g.
`॥ १.१६॥`), not the plain `॥N॥` of the other Upaniṣads; kept inline as the source
has them. They round-trip (the danda-and-dot fold in the round-trip norm).

**Schema.** Six `Chapter`s (one per praśna, numbered 1–6), each with its
`॥ इति प्रश्नोपनिषदि …प्रश्नः ॥` colophon (round-trip-verified IAST). The opening
śānti-pāṭha is the **leading verse of praśna 1** (so ch.1 holds 17 verses: śānti
+ 16). `deity` undefined.

**English + Hindi.** Researched per-section meaning glosses anchored to
**Śaṅkara-bhāṣya** — the Rayi/Prāṇa cosmogony (P1), the supremacy-of-Prāṇa
hymn (P2), the five prāṇas and nāḍīs (P3), the three states + deep sleep (P4),
Om-meditation (P5), and the ṣoḍaśa-kala Puruṣa (P6) — `draft`, presence-only audit.

### Soft spots — owed before `verified`
1. **No human verification against a printed edition.**
2. **Machine-generated IAST** (round-trip-verified, no human pass).
3. **The Prāṇa-stuti (P2.5–13) and the prāṇa-physiology (P3, P4) are dense and
   figurative**; rendered to Śaṅkara's reading and priorities for the reviewer.

---

## `aitareya-upanishad` — Aitareya Upaniṣad

**Text.** The Aitareya Upaniṣad of the Ṛgveda — from the Aitareya Āraṇyaka
(2.4–2.6), in **three adhyāyas** (the first of three khaṇḍas; **33 mantras**):
creation from the Self alone, the Self's entry into the body and self-recognition
as Brahman (Idandra → Indra), the three births, and the mahāvākya
`प्रज्ञानं ब्रह्म` (3.3). Framed by the `वाङ् मे मनसि…` śānti. Phase 16 **Batch D**.
Traditionally chanted *sasvara*, but shipped from a clean **de-accented** edition
under the **same strict-IAST posture** as the six short Upaniṣads — Option A,
settled with Ishank (a clean cross-checkable edition exists, so no best-effort
fallback was needed).

**Source & cross-check.** Devanagari imported from a public digital edition and
cross-checked against sa.wikisource `ऐतरेयोपनिषद्` (a real, populated page) at
**~96% token containment**; the residual misses are sandhi/word-split variants
plus one wikisource typo (`किञ्चचन` for the primary's correct `किञ्चन`, 1.1.1).
Digitization sources internal only.

**Transforms applied (deterministic, asserted).**
1. *Vedic nasalisation normalised* (`ꣳ`/`ँ` → `ं`).
2. *ZWJ removed* (in `श‍ृणोति`, 3.1).
3. *A stray source line-slash fixed* (`ग्रहीतु/न्` → `ग्रहीतुं`, 1.3.5) and a
   *double-avagraha artifact* (`एवऽऽत्मानं` → `एवात्मानं`, 2.1).
4. *Adhyāya 2/3 colophon spelling corrected* — the source wrote `ऐतर` for
   `ऐतरेय`; the canonical `इत्यैतरेयोपनिषदि …ऽध्यायः` form is used (round-trip-verified).

**Schema.** Five `Chapter`s (Adhyāya 1 · Khaṇḍa 1/2/3, Adhyāya 2, Adhyāya 3),
each with its own `॥ इति …खण्डः/ऽध्यायः ॥` colophon (round-trip-verified IAST),
mirroring Muṇḍaka's khaṇḍa-grain (the chapter-grain was confirmed with Ishank).
The `वाङ् मे मनसि` śānti is the **leading verse of ch.1** (so ch.1 holds 5 verses:
śānti + 4); the source repeats it at the close (noted in `about`). `deity`
undefined.

**English + Hindi.** Researched per-mantra meaning glosses anchored to
**Śaṅkara-bhāṣya** — `draft`, strict round-trip audit (Option A).

### Soft spots — owed before `verified`
1. **No human verification against a printed edition.**
2. **Machine-generated IAST** (round-trip-verified, no human pass).
3. **The creation cosmology (1.1–1.4 — the lokas ambhas/marīci/mara, the
   organ→deity emanations) and the idam-adarśa → Idandra wordplay (1.3.13–14) are
   terse and figurative**; rendered to Śaṅkara's sense and flagged for the reviewer.
4. **Vāmadeva's ṛṣi-verse (2.5 = Ṛgveda 4.27.1)** is cryptic; rendered to Śaṅkara's reading.

---

## `taittiriya-upanishad` — Taittirīya Upaniṣad

**Text.** The Taittirīya Upaniṣad of the Kṛṣṇa Yajurveda — three vallīs:
**Śīkṣāvallī** (12 anuvākas), **Brahmānanda/Ānandavallī** (9), **Bhṛguvallī**
(10) — **31 anuvākas**, mixed prose + verse. Carries the convocation
`सत्यं वद धर्मं चर` (Śīkṣā 1.11), the pañcakośa analysis (annamaya…ānandamaya),
the ānanda-mīmāṃsā, and Bhṛgu's quest. Phase 16 **Batch D**. Traditionally chanted
*sasvara*, shipped from a clean **de-accented** edition under the **same
strict-IAST posture** (Option A).

**Source & cross-check.** Devanagari imported from a public digital edition and
cross-checked against sa.wikisource at **~95% token containment**. The
sa.wikisource parent `तैत्तिरीयोपनिषद्` is a **TOC stub**; the text lives on three
per-vallī subpages under a **misspelled parent** `तैत्तिरीयोपनिषदत्`
(`…/शिक्षावल्ली`, `…/ब्रह्मानन्दवल्ली`, `…/भृगुवल्ली`), found by grepping the stub's
hrefs (cf. the Kaṭha fragmentation). That second edition is itself a
Śaṅkara-bhāṣya text (commentary interleaved), so it served for cross-check only,
not as primary. Residual misses are sandhi/word-split variants. Digitization
sources internal only.

**Transforms applied (deterministic, asserted).**
1. *Scribal maṅgala dropped* — `ॐ श्री गुरुभ्यो नमः । हरिः ॐ ।` before the text.
2. *Vedic nasalisation normalised* (`ꣳ` → `ं`); *ZWJ removed*; *a stray backtick
   removed* (`वे` + `` ` `` + `द`, 1.3.4).
3. *Editorial anuvāka topic-headings stripped* — the source prefixes anuvākas with
   descriptive labels (`संहितोपासनम्`, `व्याहृत्युपासनम्`, `ब्रह्मानन्दमीमांसा`, …, 19
   in all, incl. the glued pair `सदाचारप्रदर्शनम् । ब्रह्मानन्दानुभवः`); not mūla,
   removed (a no-leak scan asserts none remain in any verse).
4. *A leaked Latin translator-note removed* — `3 for prolonging the vowel in the
   form । अऽऽ ।` embedded at Ānanda 2.6.1 (an editorial gloss on the pluta
   `गच्छती३`); the mūla pluta is kept.
5. *One typo corrected* — `भुतानि` → `भूतानि` (Bhṛgu 3.2.1; the 2nd source and the
   text's own parallel verses read `भूतानि`).

**Markers & pluta.** Verse = the `॥N॥` unit within an anuvāka (numbering resets per
anuvāka). The Bhṛguvallī sāman (3.10.5–6) carries chant-elongations written with
the pluta digit `३` (`हा ३ वु`, `ऋता३स्य`, `अहमन्नादोऽ३हमन्नादः`); these round-trip
as-is (sanscript maps `३`↔`3`) — no special fold needed.

**Schema.** 31 `Chapter`s, one per anuvāka (Śīkṣā 1–12 / Ānanda 1–9 / Bhṛgu
1–10) — the per-anuvāka grain confirmed with Ishank, mirroring Muṇḍaka/Kaṭha so
each natural division keeps its own colophon. Each chapter carries a full
`॥ इति <vallī-locative> …ऽनुवाकः ॥` colophon (round-trip-verified IAST; the
source's bare `इति …ऽनुवाकः` is given the standard vallī-locative prefix to
distinguish the three vallīs). Śīkṣā anuvāka 1 **is** the `शं नो मित्रः` śānti and
anuvāka 12 the closing repeat; the `सह नाववतु` śānti is the **leading verse** of
the Ānanda & Bhṛgu anuvāka-1 chapters (each also repeated at its vallī's close —
noted in `about`). `deity` undefined.

**English + Hindi.** Researched per-section meaning glosses anchored to
**Śaṅkara-bhāṣya** — `draft`, strict round-trip audit (Option A).

### Soft spots — owed before `verified`
1. **No human verification against a printed edition.**
2. **Machine-generated IAST** (round-trip-verified, no human pass).
3. **Many passages are technical or subtle**: the saṃhitā/medhā/vyāhṛti/pāṅkta
   meditations (Śīkṣā 1.3–1.9, ritual); the sad/tyad & satya/anṛta cosmology and
   the contested `असन्नेव स भवति` (Ānanda 2.6); the `रसो वै सः` and
   fear-from-difference (2.7); the ānanda-mīmāṃsā calculus (2.8); the triumphal
   sāman (Bhṛgu 3.10.6). All rendered to Śaṅkara's sense and flagged for the reviewer.
4. **SD-edition spellings kept faithfully** where both sources agree or where
   correction would be conjectural — e.g. `एदद्वा` (3.10.1, both editions),
   `अन्तहृदय` (std. `अन्तर्हृदय`, Śīkṣā 1.6.1), the joined `सर्वेलोक`/`सर्वेवेदा`
   sandhi (1.5.2–3). Flagged for the human Sanskritist pass.