Created: 24-10-2026 · Last updated: 24-10-2026

title: PD — Poona Dictionary / EDSHP (1948–, unfinished)

PD — An Encyclopaedic Dictionary of Sanskrit on Historical Principles (1948–)

The atlas's first unfinished dictionary. Every other chapter measures a finished artefact; this one measures a fragment — 35 fascicle-volumes, 6,056 pages, still inside the first letter — and asks what a dictionary 4–6 % done can nonetheless tell us. Chapter built from the committed data/pd/ family; no source edition is re-parsed here.

Trust Block

• Evidence: committed data/pd/ artifacts — pd_quotations_vs_pwg.tsv (4,093 shared lemmas), pd_volume_homogeneity.tsv (35 ledger rows, 24 measured), pd_dcs_metrics.json, pd_canon_ledger.tsv. The machine text is the Cologne PDScan 2020 edition of fascicles 1976–2009.
• Limitations: letter a- only, machine text to ~apaca- (2009); print runs to 2019. PD is not comparable to a finished dictionary on coverage — only on the span it has covered.
• Validation: npm run build; the numbers below are re-derivable from the committed TSVs, and the canon bases are gated in pd_canon_ledger.tsv (--check green).
• Owner repo: csl-atlas.
• Next use: open PDScan for an entry, then the canon ledger before quoting any coverage percentage.

1. Overview

An Encyclopaedic Dictionary of Sanskrit on Historical Principles — the "Poona Dictionary", PD/EDSHP — is a Deccan College project conceived by S. M. Katre in 1948 on the OED model: senses in chronological development, each carried by dated citations from a paper corpus of quotation slips. It is the most granular Sanskrit dictionary ever attempted, and it is nowhere near finished.

The chapter runs two registers. The comparative one asks what can be said against finished dictionaries on the span PD covers; the fragment one asks what an unfinished dictionary can still measure about itself. Neither substitutes for a chapter a finished dictionary would get.

2. Comparative register — how dense, on the span it covers

PD has traversed only ~4–6 % of the alphabet a complete dictionary covers, but on that span it is far denser than any of them. In each reference's Sanskrit-collated headword list, the position of apaca (PD's frontier) gives the headwords in the same range (HeadwordLists/now-2026):

That is the headline: PD's first letter already holds as many lemmas as the entire Petersburg dictionary (104,959 against 106,082), for about 4 % of the alphabet. The density is not an artefact of counting homographs — it is the dictionary's method: every attested compound and derivative gets its own entry and its own citations.

On the lemmas both dictionaries share, the density shows up in citations too. The 4,093 lemmas present in both PD and PWG, measured with one quotation rule on both sides: PD carries a median 4.0× the attestations (mean 8.97×, p25 2.0×, p75 9.75×; 19.7 % of shared lemmas at 12× or more), and 8.55× in aggregate (157,480 PD attestations against 18,423 PWG). Below parity: 340 of the 4,093 lemmas, where PWG is the denser entry.

The red line is the median. The distribution is long-tailed: most shared lemmas sit between 1× and 5×, but a fifth are at 12× or more, and the tail is where PD's encyclopaedic ambition lives.

3. Fragment register — what an unfinished dictionary can still measure

A fragment cannot be profiled like a completed dictionary, but it can be tested for editorial stability across its own production. Three measures, all from the 24 fascicle-volumes the machine text carries:

1. Entry size is stable. The median entry is 3 printed lines in every one of the 24 measured volumes and in all 8 bound volumes. The mean runs 4.22–7.56 lines by fascicle — variation in tail length, not in the typical entry.
2. Source marking declines, and not in a clean step. The share of entries carrying a bracketed dictionary source falls from 2.06 % in vol. 1 to 0.39 % in vol. 8, but the series is non-monotonic (pre-v7 average 1.94 %, v7 0.88 %, v8 0.39 %; fascicle 7.2 = 1.88 % against 7.1 = 0.53 %). The cheap conversion signatures are absent — general bracket use is stable at 11–17 % of entries, long source brackets number 29 in the whole corpus, OCR near-misses of the seven source sigla number 3 — so the fall reads as editorial/content, not a conversion artefact (medium-high confidence; a print check of one v7 and one v8 page is the falsification test still owed).
3. Citation mass per page is high and steady. 419,890 attestations over 6,056 printed pages is ≈69 attestations per printed page across the 35-volume record.

The flat median in the first figure and the falling bars in the second are the same finding from two sides: PD's entry shape is remarkably constant across four decades of production, while its acknowledged inheritance from earlier dictionaries thins out — a fact about the dictionary's sources, not about its ambition.

4. What this chapter cannot tell you

• It cannot rank PD against a finished dictionary on coverage. Only ~4–6 % of the alphabet exists; §2's ratios are density on the covered span, not completeness.
• Its denominators are not interchangeable. "118 covered works" divides by anything from 1,118 to 2,031 depending on the base; the canon ledger labels every one, and a percentage without its base is a defect.
• It cannot see print pages 2009–2019. The machine text stops at 8-4327a (~apaca-); the printed record goes further into a-.
• It inherits a known trap. MahāBhā. is the Mahābhārata, MahāBh. is the Mahābhāṣya — similarity clustering fuses them; only the locator shape separates them.

5. Read next

• PD × DCS coverage report — what the largest lemmatised Sanskrit corpus does and does not hold of PD's canon.
• pd_canon_ledger.tsv — every canon base and its 118/N.
• PD knowledge page — source-exhaustive facts, do-not-quote list, three verbatim entries.
• PDScan 2020 — the machine text itself.

Sources

• data/pd/pd_quotations_vs_pwg.tsv, pd_quotations_summary.json — the 4,093-lemma comparison, one quotation rule on both sides.
• data/pd/pd_volume_homogeneity.tsv + H6947 report — 35 ledger rows, 24 measured.
• data/pd/pd_dcs_metrics.json, pd_canon_ledger.tsv — coverage and canon bases.
• reports/PD_DCS_CORPUS_COVERAGE_2026.md §8.2/§8.5 — the density yardsticks and the completion horizon.

Chapter added 24-10-2026 — DeepSeek Flash (deepseek-flash) via DSH; figures read the committed TSVs, no number is typed into prose that is not in data/pd/.