diff --git a/docs/arz/oracles.md b/docs/arz/oracles.md
new file mode 100644
index 00000000..ec9ddc8c
--- /dev/null
+++ b/docs/arz/oracles.md
@@ -0,0 +1,10 @@
+# Egyptian Arabic (arz): parked — thin, inconsistently voweled
+
+UniMorph `arz` has ~1,320 verb lemmas, but only **636** carry both a
+3sg-masc perfective and imperfective (the principal parts the templatic
+engine seeds from) — 48%, far below the 99.5% lemma-coverage bar. The
+vocalisation is also inconsistent (كان unvoweled next to سِمِع partly
+pointed), so a rule engine can't reliably match the target orthography.
+kaikki.org has **no** Egyptian Arabic verb dump, so there is no second
+oracle and no way to cross-check. Revisit if a fuller, consistently
+voweled dialect lexicon (e.g. CALIMA-EGY output) becomes downloadable.
diff --git a/docs/syc/oracles.md b/docs/syc/oracles.md
new file mode 100644
index 00000000..2997ba57
--- /dev/null
+++ b/docs/syc/oracles.md
@@ -0,0 +1,10 @@
+# Classical Syriac (syc): parked — sparse seeds, participle-heavy
+
+UniMorph `syc` has ~755 verb lemmas / 17k rows, but after stripping the
+Syriac vowel points only **269** lemmas (36%) carry both a 3sg perfect and
+imperfect, so principal-part seeding covers nowhere near the 99.5% lemma
+bar. The paradigm is unusually participle-heavy (the active/passive
+participles inflect for person and dominate the rows) and the roots are
+written defectively (2–3 consonants), which the templatic engine can't
+disambiguate from the citation alone. kaikki has no usable Syriac verb
+extraction. A dedicated Aramaic FST oracle would be needed; parked until then.