Chapter authored per Decision 29 Tier C. Position 8 in the atlas ordering — the first Sanskrit-Sanskrit lexicon, where the 18-block framework stops applying.
Śabdakalpadrumaḥ ("the wishing-tree of words"), an encyclopedic Sanskrit-Sanskrit dictionary in seven volumes compiled by Rājā Rādhākānta Deva and his circle, published Calcutta 1822–1858. SKD is the first indigenous-Indian large-scale lexicon in the modern Sanskrit-philological tradition: encyclopedic in scope, monolingual in language (Sanskrit headword + Sanskrit definitions), and rich with citations from literary works and prior kośa sources via inline iti <source> prose. Within CDSL, SKD marks the genre boundary of the present framework: where MW, PWG, PWK, AP, BEN, CAE, WIL are structured bilingual dictionaries amenable to the 18-block detector, SKD and VCP are encyclopedic Sanskrit-Sanskrit works whose source-discipline lives in prose, not in tags. The block apparatus developed for MW does not apply.
The 18-block apparatus developed for MW in PAPER.md §3 is genre-bound to structured bilingual dictionaries. SKD violates the framework's three core preconditions:
No <lex> grammatical category tag. SKD marks gender, voice, declension class, etc. inline in Sanskrit prose ("X iti puṁsi" — "X is masculine", literally "X-thus-in-masculine"). The MW detector that searches for <lex>m.</lex> finds zero matches.
No <ls> source citation tag. SKD cites sources via the formula "… iti " inline in the gloss prose. Inline iti appears 72,176 times across SKD's 42,531 records — 1.70 per record (vs MW ≈ 0). This is the operative citation apparatus; the digitisation does not extract it as <ls> because the prose context does not survive tagging.
Long encyclopedic prose entries. Mean entry length is 532 characters in SKD (vs ~80 in MW). The "kernel + enrichment" block-economy that defines MW does not have a counterpart — SKD entries are paragraphs of prose with quotations woven through, not slot-and-block compositions.
The block apparatus is therefore inapplicable to SKD. Block-presence by name shows mostly empty:
(The bars largely flatline at zero — this is the visual signature of the genre boundary.)
SKD's source-discipline is inline prose rather than tagged. The pattern is the formula "… iti " — literally "thus says " — which marks the preceding quotation or paraphrase as drawn from the named work. From analysis/CROSS_DICT_PROFILES.md Part B:
Metric
SKD
(vs MW for contrast)
Records
42,531
286,561
<ls> tags
0
311,932
Inline iti
72,176
≈ 0
iti per record
1.70
≈ 0
Mean entry length (chars)
532
~80
A typical SKD entry quotes Manu, Amarakośa, Bhagavad-Gītā, Yajurveda, Līlāvatī, or a Pāṇinian sūtra by name in the prose. The reader who knows the iti-formula reads the source-attribution off the page; the digital detector that searches for <ls> does not.
The European tradition (PWG, MW, etc.) treats every source citation as a separable structural slot: a <ls> tag occupies a fixed position in the entry, machine-distinguishable from the gloss it accompanies. The indigenous kośa tradition (SKD, VCP, and the prior Amarakośa / Medinīkośa / Hemacandra works SKD synthesises) treats every source citation as a prose feature of the gloss itself — the source is part of what the gloss says, not a tagged annotation about the gloss. This is not a technical limitation of the medium (SKD was printed in 7 large volumes with full typographic apparatus); it is a design choice rooted in the encyclopedic-kosha tradition's preference for woven, paragraph-length, citation-rich definitions over the European tradition's slot-and-block compactness.
The implication for the present framework: the 18-block apparatus is not a universal microanalysis tool. It is a tool calibrated to structured bilingual dictionaries — the 7 CDSL works that MW, PWG, PWK, AP, BEN, CAE, WIL belong to. SKD and VCP require a different microanalysis tool, one that takes inline-iti as its citation unit and paragraph-prose-flow as its block unit. This work is not done in the present atlas; it would be a separate paper / future Phase-5 project.
SKD occupies the encyclopedic-synthesis position in the indigenous Sanskrit kośa tradition. It draws on the prior kośa literature (Amarakośa 6th c., Medinīkośa 14th c., Hemacandra Anekārtha-saṃgraha 12th c., Trikāṇḍaśeṣa, Halāyudha, Śabdaratnāvalī, Vaijayantī) and on the Sanskrit literary canon (Vedic, epic, Purāṇic, classical kāvya, dharma-śāstra). It is the first encyclopedic work of the modern Indian indological tradition — Rādhākānta Deva's circle was working under British-colonial patronage but the editorial choices are recognisably kośa-traditional. The four indigenous-kosha repos that resolve MW's <ls>L.</ls> hedge (ARMH, ABCH, ACPH, ACSJ) are the prior kośa works SKD itself synthesises; SKD therefore plays a dual role — encyclopedic kośa in its own right, and the target into which MW's hedges resolve.
PWG cites SKD as <ls>ŚKDR.</ls> 20,109 times — SKD's biggest impact on the European tradition was as PWG's primary indigenous-lexicon source.
Both 19th-century Calcutta compilations rooted in indigenous-Indian scholarship
WIL is bilingual (Sanskrit → English); SKD is monolingual Sanskrit-Sanskrit; WIL is structured-bilingual genre, SKD is genre-bound encyclopedic; the genre boundary in the atlas falls here
Both encyclopedic Sanskrit-Sanskrit; both no <lex>/<ls> tags; both prose-paragraph entries
SKD inline iti density = 1.70/record (the densest); VCP = 0.26/record (much sparser). SKD is the prototype of the genre, VCP is a variant — different editor (Tārānātha), different methodology, but same genre
Tier C template per Decision 29 §29.2. §§ 2–4 replaced with structural-features, prose-pattern, and "why framework does not transfer" sections.
The genre boundary is the chapter's central finding — explicit linkage to PAPER.md §8 (framework is genre-bound to structured bilingual dicts).
Dual role of SKD (indigenous kośa + PWG citation source <ls>ŚKDR.</ls> 20,109×) surfaced in §5.
Effect-size threshold (D19): numerical claims (1.70 iti/record, 0/0/0 tags, 532-char mean length) exceed the threshold by orders of magnitude — these are not borderline findings.