Why is a the letter of compounds — and do dictionary entries shrink over time?

Every Sanskrit dictionary is largest at a, and the H1336 study found that 83 % of Monier-Williams's a-entries are dash-marked compounds, not roots. This page asks whether that is unique to a (it is not), why some letters balloon (they head preverb families), and tests a historical claim raised for Śabdakalpadruma and Vācaspatyam: do entries get shorter toward the end of the work as funding fell? Full write-up: LETTER_ANATOMY_AND_ENTRY_SIZE_2026.md.

Q1 — Samāsa share per letter (Monier-Williams)

a is the most compound-dense letter — but u, p, s, v are right behind it, and k (which heads no preverb) falls away. Compound-heaviness is not an a-anomaly; it tracks whether a letter heads a productive preverb family.

Q2 — Upasarga profile: the mechanism

Each large letter is large because it heads a preverb family: v = vi-, u = ud-/upa-, s = sam-/su-, p = pra-/pari-/prati-, a = ā-/abhi-/anu-/apa-/ava-. k heads none.

Every preverb, every dictionary

Each preverb's total sits entirely under its own initial letter, so these are exact counts. vi- leads in every dictionary; the sam/su order flips in Vedic Grassmann (su > sam).

Q4 — Do entries shrink toward the end of the dictionary?

The funding-decay test. Naïvely regressing entry size on alphabetical position confounds the funding signal with the fact that later letters host shorter words (Q3). The verdict below uses an outlier-robust per-letter rank test (each letter analysed on its own, then aggregated) — mandatory because the encyclopaedic dictionaries have single articles over 300,000 characters.

Each tick is one initial letter. PWG / PWK / GRA lean overwhelmingly negative (later entries shorter within every letter); VCP scatters symmetrically around zero — no decay.

Mean entry size by decile of the alphabet

Raw decile means (below) are confounded by letter composition — VCP's fall and SKD's rise both vanish once each letter is controlled. Shown to make the confounder visible, not as the test.

Wave B — PWG entry size vs real publication year

H1416 measured decay against alphabetical position. PWG's <pc> field encodes the volume (1–7), each with a known year, so every entry maps to a real year — PWG entries shrank −14.3 %/decade, and the counter-test shows it is a smooth fade (volumes 2–7 still −15.3 %/decade after dropping the over-detailed vol-1), not a one-time policy break.