Created: 23-05-2026 · Last updated: 24-09-2026

title: ARMH — Abhidhānaratnamālā (Halāyudhakośa)

ARMH — Abhidhānaratnamālā (Halāyudhakośa, ~10th c.)

The classical Sanskrit synonymic kosha by Halāyudha (~10th century CE) — a verse-format synonymic dictionary, each verse listing the synonyms of one concept. This page is a data chapter, and it is also the honest mirror case of the kośa model: the CDSL digitization is exploded — one <L> record per synonym, with set boundaries, varga names and gender all not encoded — so this chapter counts exactly what the digitization can show and labels the rest as absent.

Source: csl-orig v02/armh/armh.txt · ARMH GitHub

Trust Block

Evidence: the CDSL v02 armh.txt digitization (revision f4c08c5), measured by the kośa macrostructure builder m10_kosa_macrostructure_model.py (H5328) and reshaped for this page by build-kosa-chapters.mjs (H5329); field inventory from data/parse-rules/armh.json.
Limitations: the exploded digitization encodes neither synonym-set boundaries nor gender, and its <vn> locator varies only at the kāṇḍa level — the chapter states those absences as measured facts, never fills them by inference.
Validation: npm run validate-kosa-chapters (arithmetic + cross-artifact reconciliation, incl. 7,907 records = instance members) and node --test test/kosa-chapters.test.mjs; page renders checked by npm run build.
Owner repo: csl-atlas (handoff H5329, epic E014; model and measures H5328).
Next use: cite as the exploded-digitization row of the kośa macrostructure type; the kāṇḍa profile below is the finest structure the digitization supports.

At a glance


Genre	Classical Sanskrit synonymic kosha (Brahmanical)
Author / date	Halāyudha, ~10th c.
Digitization model	exploded — one `<L>` record per synonym
Records	(= word forms recovered from the verses)
Verse-groups	(`<vn>` locators)
Divisions	kāṇḍas — numeric locators only, no varga tier
Full verses
Set boundaries	not encoded
Gender (liṅga) marking	absent in markup
CDSL role	Lineage source for WIL → MW chain; cited in PWG as `HALĀY.`

The digitization shape, counted

Each record carries the headword (<k1>), a segmented sort key (<k2>), the page-column coordinate (<pc>) — and the verse locator (<vn>), the tag that groups the exploded records back into their verses: 7,907 <vn> tags, one per record. The locator's first field is the kāṇḍa; it never varies below it, so kāṇḍa 1–5 is the finest division the digitization encodes. Kāṇḍa 5 is the homonym book (anekārtha) — identified from its "…api…" formula, not from a heading.

What the digitization does not encode

The kośa model's orderingDevices table measures absence as honestly as presence. For ARMH the absent devices are the point: the verses carry the macrostructure, the digitization does not.

Alphabetical adjacency across the record pairs is — chance level, as expected for a concept-ordered kośa whose records happen to sit in verse order.

Chart Trust Block

Claim: ARMH's CDSL digitization is an exploded kośa — 7,907 one-synonym records regroupable into verse-groups via the <vn> locator, with kāṇḍa as the finest encoded division, and set boundaries, varga names and gender absent from the markup; its macrostructure is countable only down to the kāṇḍa, and the chapter says so rather than inferring the rest.
Evidence label: derived — a deterministic reshape of the committed kosa model artifacts (H5328) and ARMH parse rules; nothing hand-counted.
Source files: src/data/kosa-chapters/kosa_chapters.json with provenance sidecar kosa_chapters.source.json; upstream csl-orig v02/armh at revision f4c08c5.
Generated by: npm run build-kosa-chapters
Validation: npm run validate-kosa-chapters (instance members 7,907 reconciled against parse-rules record_count; explicit checks that ARMH carries no grouped-model measures and that the varga and synonym-set devices stay absent) and node --test test/kosa-chapters.test.mjs (9 tests including four mutations).
Known false positives: none — no detection happens on this page.
Known false negatives: the kāṇḍa-5 homonym identification is inferred (the "…api…" formula), and is labelled so in the devices table; full-verse counts derive from verse-end digits and inherit that method's limits.
Review status: machine-reviewed (deterministic validator); underlying measures reviewed in H5328 (class-verifier PASS).
Owner repo: csl-atlas.
Next action: none open; this chapter and the ABCH counterpart land together (H5329).

ARMH beside AMAR and ABCH

Where AMAR and ABCH digitizations are grouped (one record per verse, <eid> sets, liṅga tags), ARMH's exploded records preserve the words and the verse locator but discard the set structure — the same genre at two digitization depths, which is exactly what the kośa macrostructure schema models with its unsegmented-verse set kind.

Why ARMH is in the atlas

The kosha group as WIL's primary source. Wilson (WIL) and the Fort William College pandits worked from classical kosha sense-divisions to construct English glosses; ARMH is one of the principal sources of those sense-divisions.
Same word, different traditions. aṃśu appears in ARMH's synonym verse for "sun" and for "ray" — the very sense-divisions WIL preserves ("A ray of light, a sun-beam... The sun...") and MW inherits. From DICT_PROFILE.md — The four Cologne koshas: aṃśu in ARMH L369 (names of the sun) and ARMH L411 (words for "ray").
Cited in PWG as HALĀY. Böhtlingk-Roth cite Halāyudha explicitly; Monier-Williams collapses these citations into the anonymous <ls>L.</ls> hedge.

The four CDSL koshas

CDSL repo	Title	Author	Date
ARMH	Abhidhānaratnamālā	Halāyudha	~10th c.
ABCH	Abhidhānacintāmaṇi	Hemacandra	~12th c.
ACPH	Abhidhānacintāmaṇi-pariśiṣṭa	Hemacandra	~12th c.
ACSJ	Abhidhānacintāmaṇi-śiloñcha	Hemacandra (attr.)	~12th c.