diff --git a/QUESTIONS.md b/QUESTIONS.md
index b1f01b3..bc07160 100644
--- a/QUESTIONS.md
+++ b/QUESTIONS.md
@@ -44,6 +44,44 @@ an open row.)
 (The page numbers in `qslips.log` can be one out — the mark it reads is the edition page's last one — so this
 table is the authoritative record.)
 
+## Diacritic sweep, 2026-09-15 — 9 corrections, ~54 of 84 flagged occurrences checked
+Prompted by Sammay spotting that `Nastaʿlīq` on orig. p.47 showed a quote, not a ʿayn. Two nets over the
+transcription: a **rare-character inventory** (U+02BF appeared exactly once in 431 pages — that singleton
+was the tell) and a check for **names spelled with more than one accenting** (84 occurrences). Every flag
+was read off the page image at 500–900 dpi; nothing was changed on the strength of the text layer.
+
+**Corrected** (all verified against the image):
+
+| PDF page | was | is |
+|---|---|---|
+| p0049 | `Nastaʿlīq` | `` Nasta`līq `` — the mark is a raised quote, as the typewriter had no ʿayn |
+| p0072 | `Cálidás` | `Cálidās` — second mark is a macron, not an acute |
+| p0256 | `Rajākṛṣṇa` | `Rājākṛṣṇa` — macron over both a's |
+| p0260 ×2 | `Īsvaracandra` | `Īśvaracandra` — acute on the s was missing |
+| p0430 | `Iśvaracandra` | `Īśvaracandra` — macron on the I was missing |
+| p0430 ×2 | `Gītā` | `Gīta` |
+| p0430 | `Munṣī` | `Munśī` — acute above the s, not a dot below |
+| p0431 | `Rtusamhāra` | `Ṛtusaṃhāra` — both dots were missing |
+
+**Two findings that explain most of the rest.** The typescript's **French accents are handwritten** — the
+cedillas under *poinçons*, the acutes on *gravés*, *impériale*, *Napoléon*, the grave on *caractères* are
+pen strokes, added inconsistently (one *Poinçons* on p.226 carries a hand cedilla, three others do not).
+Each instance is reproduced as it stands. And in the **Bibliography the author gives a title in full
+transliteration beside its anglicised form** — `Ānanda Bājāra Patrikā (Ananda Bazar Patrika)` — so most
+Bibliography variants are hers. `Ṛtusaṃhāra` was catchable only because it broke that pattern.
+
+**Still unchecked** (~30): `Giriśa` ×5 (PDF 11, 16, 244, 250, 251), `Bengālī` ×4 (10, 86, 196, 425),
+`Bāṅgalī`/`Bāṅgālī` ×3 (8, 94), `Raphala`/`laphala` (324), `Gouye` (425), `Sarma`/`Poincons` (426),
+`Rama` (428), `Bhagavat`/`Geeta` (423), `Pancanana` (108), `pratyūttara` (117), the two
+Scheme-of-Transliteration rows (15). Each has a sibling occurrence already verified correct, and the last
+five batches (~24 occurrences) found no errors, so the remaining yield looks low — but it is not zero, and
+these are unchecked, not cleared. `Vyākarana` (255) is the authorial inconsistency the colophon already
+cites as deliberately kept; leave it.
+
+**Limitation worth keeping in view.** Both nets depend on *inconsistency*. A name transcribed the same
+wrong way everywhere passes both silently — `Bhăgavăt-Gēētā` was only checked because its rare breves
+tripped the character inventory. So the flagged cases are clean; the diacritics as a whole are not proven.
+
 ## Resolved
 - **p.295 *compliments* and p.348's dropped *to* join the Errata** (Sammay, 2026-09-14). This overturns the
   p.84 precedent for these two: a word-choice slip and a dropped word are corrected and listed like a